This PDF file contains the front matter associated with SPIE Proceedings Volume 12642, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

In the receiving-end power grid, the proportion of external power supplies and large-scale renewable power generations has increased rapidly, which not only increases the unbalanced power impact of the system, but also weakens the frequency support capability of the system. Therefore, frequency security becomes a bottleneck of the receiving-end power grid, and system frequency response modeling becomes a key concern. In this paper, an aggregation modeling method is presented to build the system frequency response model (SFR). Two types of the SFR model structures are proposed. The first one considers the dynamics of the governors and prime movers of both the thermal and hydro units, which is suitable for the simulation and analysis of the system frequency behaviour following small power disturbances. The second one additionally considers the boiler dynamics of the thermal units, which is suitable for the simulation following large power disturbances. The parameter aggregation method for the SFR model are also proposed. The aggregation SFR model is built based on the detailed model of the east China power grid. The results show that the proposed method can achieve satisfactory performance owing to its accuracy and efficiency.

In order to adapt to the characteristics of high mobility and high efficiency firing of aviation artillery shells, meet the requirements of convenient transfer and rapid formation of security capabilities, the design of an aerial gun pressure and discharge machine is optimized by using the bullet chain supply structure, and the two working processes of pressure and discharge are realized through the mode of interchangeable components, and the key motion process is simulated and verified by using the dynamics software ADAMS. The power on its power shaft is about 130N，comply with index design requirements. The motion process is reliable. The structure has the characteristics of small size, light weight, simple operation, fast discharge speed, stable and reliable quality, easy to carry and transfer, etc., which can meet the needs of high-intensity modern warfare.

To verify the capability of the jack-up platform during earthquake, the seismic analysis based on spectrum response method is studied. First SACS analysis model of the platform is established, then the vibration mode shape and seismic response is analyzed, and finally the loads exacted from seismic response are combined with the gravity loads, and member and joint code check is executed. The calculation and analysis results show that the target jack-up platform meets the strength requirements under the design seismic load.

In order to study the destructive ability of the target debris to the personnel inside the tank, the numerical simulation method is used to simulate the penetration process of different masses of rhombus debris to the personnel equivalent target plate, the kinetic energy and velocity change law during the penetration process of the target debris to the target plate, and the limit penetration velocity of the target debris to the personnel equivalent target is obtained, so as to provide a basis for the study of the penetration of the target debris to the personnel inside the tank. The study provides the basis for the destruction of internal personnel.

In this paper, the lightweight design and research are carried out for a certain type of hydraulic bulging axle housing. In this paper, the local wall thickness of the axle housing is used as the optimization variable, and the maximum stress of the axle housing under the shock load condition is the response constraint. The Box-Behnken test design method is used to determine the best test point and conduct the test, and the response surface function of the liquid expansion axle housing is established. Then the genetic algorithm (GA for short) is used to solve the response surface function, and finally the goal of optimizing the axle housing quality is achieved. The optimized axle housing meets the design requirements while reducing the mass by 7.4Kg, and the weight reduction rate reaches 10.11%.

In this paper, the PyroSim software is used to simulate the fire in an elderly apartment and analyze the vertical and horizontal variation characteristics of temperature, visibility, CO volume fraction, O2 concentration and flue gas layer height caused by fire combustion. The results show that the first floor is the fire floor, which is the most seriously affected by the fire. The temperature quickly reaches 70℃ at about 91-280s. To 54~265s, visibility rapidly decreased to less than 10m. The 1st to 6th floor of stairwell No.3 cannot be evacuated because the temperature reaches the value. Finally, the available time for the elderly evacuation is obtained, in order to provide a basis for the fire evacuation design and safety management of the elderly apartment.

Forecasting natural gas is an important link to maintain the balance between supply and demand of the natural gas industry and plan energy strategy. From the three main dimensions of economy, population and development, this paper selects six influencing factors: GDP, per capita GDP, per capita disposable income of urban residents, the proportion of secondary industry added value in GDP, the number of urban permanent residents and urbanization rate to analyze the prospects of natural gas energy. After pre-processing to improve the added value of the data, a grey forecasting model, a polynomial regression model and a partial least square method were used for modelling, and the three models were coupled to construct a comprehensive forecasting model. The model results show that: 1. The combined forecasting model has outstanding advantages. Under the premise of considering multiple influencing factors, the advantages of the three models are complementary, and the fitting error is 2.42%. 2. The forecast results show that natural gas consumption in Jiangsu Province will exceed 50 billion cubic meters in 2028, and reach 58.748 billion cubic meters in 2031, which is 187% of 2021, with an annual growth rate of more than 2 billion cubic meters.

As a widely studied fundamental block in photonic integrated circuits, multimode interferometer (MMI) is excellent in coupling of multiple light sources with equal intensity. However, unacceptable excess loss occurs if phase-matching is not satisfied at any input port. In this paper, we proceed direct binary search (DBS) algorithm to optimize an inverse designed 3 × 1 MMI coupler with nano-pixel structure and realize high-efficiency coupling of equal input (intensity and phase) sources of 1550 nm fundamental TE mode, with a compact footprint of 2.5 × 2.5 μm² and low excess loss of 0.04dB. We also investigated the possibility of inverse design method to handle the coupling of multiple input sources with arbitrary phase difference among different ports.

In this paper, the SiC MOSFETs with different channel crystal orientation are all fabricated on the 6-inch (0001) 4H-SiC wafer. Preliminary measurements demonstrate that the fabricated 4H-SiC trench MOSFET with <11-20< channel crystal orientation has the highest carrier mobility and the strongest current capability of unit gate width. This high carrier mobility is consistent with the lowest interface state density. Finally, an enhanced 4H-SiC trench MOSEFT with breakdown voltage of 900V, specific on-resistance of 0.8 mΩ·cm² and threshold voltage of 4.2V was prepared based on the channel crystal orientation of <11-20<.

In the complex environment of the substation, small objects such as long-distance cigarettes, cigarette boxes, and lighters have few imaging pixels and lack texture information, making it difficult for convolutional neural networks to extract small object features. In the case of multiple targets, the missed detection rate and false detection rate of small targets are high, and the fusion of model features is insufficient, making it difficult to accurately identify and detect. Aiming at the above problems, a multi-scale small target detection algorithm is proposed. In the prediction part of the network, a more effective decoupling head is designed. In addition, shallow features are introduced to improve the feature pyramid, extract small target features, increase the correlation between multiple targets, and prevent the loss of small target feature information. At the same time, a multi-layer attention mechanism is embedded in the backbone network to increase the regional features of invisible small targets and reduce the missed detection rate. In the post-processing stage, the Focal Loss loss function is introduced to increase the model's learning of positive sample targets and further reduce the rate of missed detection and false detection. The experimental results show that the method achieves a 𝑚𝐴𝑃@. 5: .95 of 0.6350 on the homemade smoking dataset, and 𝑚𝐴𝑃@0.5 achieves 0.9569. For the self-made multi-scene and multi-scale smoking data set, this model has advantages in detection accuracy compared with the current excellent target detection models such as YOLOX and YOLOv5. The experimental results show that the model method can realize the identification and detection of small targets such as cigarettes and lighters under multi-scale targets, which has a certain reference value for anti-smoking measures.

Clean energy is the development direction of the future, dividend period of high-speed development in China, the clean production has become a concept of the production of the whole society, and an important component of the electric power as a clean energy gradually by the attention of the society from all walks of our life, Power transmission and transformation works is the carrier of electric power transmission and transformation, the construction of its perfect degree of the serious influence the use of the society, however, Current construction enterprise competition gradually increase, the market is also quietly reform the existing mechanism, power transmission and transformation project at this stage there are a lot of problems. There are many risks in the construction process, and the control of these risks is difficult. The above factors seriously threaten the safety of construction workers and various equipment. Therefore, it is very important to strengthen the management and control of power transmission and transformation and to ensure the safety of construction workers.

As the cornerstone of economic development, power infrastructure business is of great significance for improving the quality of economic development. Compared with the traditional power infrastructure, the new power infrastructure is more technical and professional, and more intelligent. Taking the application of abnormal object detection in edge computing in line infrastructure construction site as an example, this paper expounds the important role of front-end intelligent sensing devices in new infrastructure construction. Compared with the traditional cloud platform processing method, edge computing technology deploys computing power on the physical equipment and data source side of the near power distribution terminal, conducts data analysis, system operation status situational awareness, and makes independent and rapid decisions on the control execution unit side, effectively making up for the shortcomings of cloud computing, responding more timely, and having higher reliability, so it is more suitable for deployment in business scenarios with unstable network environment and high response requirements. In this paper, a new method of edge computing node deployment is proposed to meet the requirements of distributed edge computing in distribution information physics system, considering information stability and power stability The calculation model of computing information stability is constructed by quantifying the real-time degree, accuracy and integrity of information. The power stability scale is established by observing small signal voltage changes. The system information power mixed entropy is calculated based on the mixed entropy theory, and the optimal deployment of network edge computing nodes is solved Finally, based on the IEEE 39 bus power system standard example simulation, the system power stability, information transmission stability and power grid frequency control performance are numerically analyzed. The experimental simulation results verify that the ECN deployment scheme obtained by the proposed method can effectively improve the network response speed, have stronger reactive power support, and achieve rapid suppression of system frequency deviation.

Tunnel lighting safety is an important condition to ensure the safe operation of the tunnel traffic, CAN bus with its simple structure, high stability, strong anti-interference ability, better scalability, as well as low cost, can be widely used in the tunnel lighting bus. The CAN protocol, however, lacks an encryption function, making it possible for hackers to launch replay attacks. In this paper, we have comprehensively analyzed the security problems of the CAN bus. We examine the security flaws present in the CAN bus, design an attack method for the CAN bus, and describe how to reverse decrypt the bus packet message with the intention of controlling the lighting system. Through the use of replay attack and reverse analysis, the lighting system's control instructions from the standard data packet were successfully decoded, it has been demonstrated through simulation experiment that the CAN bus can be quickly reverse the ordinary data packet control instruction of the system at low cost.

In order to solve the problem of time-consuming multi-graph segmentation and fuzzy boundary segmentation, a full-convolution multi-graph segmentation algorithm based on region of interest location and contour perception was proposed. In this model, the ROI positioning module is serial-ized with the fusion network. The ROI positioning module locates the segmented area, reduces the parameters of the subsequent fusion module and the network load, and improves the training speed of the network model. The multi-sensory domain fusion strategy and dense connection are intro-duced to obtain the global and local context information of the image and enhance the detail rich-ness of the image. Then, the DICE loss function based on contour perception was used to constrain the sensitivity of the network to the image segmentation boundary and enhance the segmentation efficiency. Two public brain data sets were used to evaluate the performance of the proposed algorithm by segmenting the hippocampus. The experimental results show that the time cost and seg-mentation accuracy of the proposed algorithm are improved, and the average DSC, HD95 and JSC of the proposed algorithm reach 96.73%, 2.110 and 93.7% respectively.

As intelligent grid has been fully promoted in recent years, electric information acquisition system has become an important factor restricting its development. In addition to the development of user-side service functions, the communication mode of the acquisition system has also become crucial. In view of the current problems of low data transmission efficiency, inadequate data acquisition and low real-time monitoring efficiency of intelligent meters, this paper proposes a method for collecting and returning electric power data based on an operator's fiber optic network. The method adds an embedded fiber network unit module to the intelligent meter, encodes electric power data and other parameters of the intelligent meter according to the DL/T645 and Q/GDW376.1 protocols of the automatic meter reading system, and realizes the meter reading function, high-speed broadband access network function, and broadband Internet access function for users through the TCP/IP protocol of fiber communication technology and the existing power line carrier technology. The experimental results show that the device has good operating status, high data acquisition success rate, and satisfactory speed test results. This method provides a new direction for realizing the informatization and automation of the intelligent grid, and provides a new way for user broadband access through combination with power line carrier.

In the actual collection process of point cloud data, the measured road point cloud data is incomplete due to the influence of various factors such as collection equipment and object occlusion. This paper proposes a LiDAR road point cloud data completion algorithm based on edge information. The algorithm uses the OSM(OpenStreetMap) map information as structural and semantic prior knowledge to locate erroneous areas and extract road edge information. Then, the algorithm uses the road defective area edge information data sets as input to obtain the semantic information of road data. Finally, the algorithm realizes road point cloud map completion based on the nearest neighbor principle. Research in this area does not rely on the parameters of the hardware devices used for data acquisition, but uses the data itself as a driver to complete the data. Compared with traditional algorithms, this algorithm reduces the average nearest neighbor distance (Er) of road missing areas to 0.08m in road point cloud data completion, while this algorithm has better robustness and more accurate results for defective data completion.

With the continuous progress of society and economy, the management of water resources and the prevention and control of sewage are becoming more and more important. Due to the complex structure of the mechanistic prediction model and the huge amount of computation, the universality is weak, and it is not easy to generalize. In order to improve the prediction accuracy of water quality index data, this paper proposes an ARIMA-GRU water quality prediction model based on wavelet decomposition. The model divides the water quality index data into high-frequency components and low-frequency components through wavelet decomposition and coefficient reconstruction, which are respectively input into the differential autoregressive moving average model (ARIMA) and the gated recurrent neural network (GRU), and then each model is used. The predicted values are combined to obtain the final prediction result. Compared with a single prediction model, the water quality prediction model has better accuracy and higher coincidence with the real value, and has practical application value.

Rice is one of the most important grains in the world and its yield increase and quality improvement have always been the focus of research. Low temperature plasma (LTP) technology is a green agricultural technology, which can increase crop yield and improve crop quality. Accurate yield prediction and evaluation can promote the adjustment of agricultural production structure, the integration of agricultural resources and the healthy development of agricultural industry. It can also help to adjust crop management and commercial decisions (for example, to determine sales prices and marketing plans). In this paper, a plasma rice yield prediction model based on Bi-directional Long Short-Term Memory (Bi-LSTM) artificial neural network is constructed, which can accurately predict plasma rice yield. Compared with Multiple Linear Regression (MLR) and Support Vector Machine (SVM) methods, the results showed that the Bi-LSTM prediction model could well predict plasma rice yield, and the average error of predicted yield was 25 kg per mu (1mu = 666.67m²).

Millimeter wave landing radar is one of the important payloads of deep space landing probes, providing the landing probe with its distance and velocity information relative to the landing surface in the landing descent process to ensure landing accuracy and safety, and the landers of Chinese Chang'e Lunar Exploration Project and Mars Exploration Project are all equipped with millimeter wave landing radar. In this paper, a novel microwave landing radar with hybrid pulsedoppler and continuous wave work mode is presented. Analog and digital circuit integration technology with characteristics of light weight and small size is utilized to miniaturize the subsystems design,. This novel design can be applied to the follow-up project applications such as Chang'e Lunar Exploration Projects and manned lunar exploration.

Air quality indexes (AQI) forecasting plays a vital role in quality of life. Most of AQI prediction methods ignore the nonstationarity of time series, which reducing the accuracy of model forecasts. A novel approach which combines ensemble empirical mode decomposition (EEMD) and autoregressive integrated moving average (ARIMA) model is proposed for AQI forecasting in this paper. Firstly, the AQI sequence is decomposed by EEMD and the noise can be removed by the new threshold method. Secondly, ARIMA model is used to predict the obtained multiple stationary subsequences. Finally, all the separate prediction results are summed to obtain the predicted values of AQI. In terms of the forecasting assessment measures, the proposed model is superior to other methods.

With the proposal of China's dual carbon goals, the carbon emission reduction contribution of battery electric vehicles in use phase has gradually attracted the attention of the industry. At present, the method for calculating carbon emission reduction of battery electric vehicle in use phase is ambiguous and subjective. This study proposes an objective baseline determination method based on the curb weight parameters of vehicles, which is simpler and clearer. On the basis of baseline determination, an calculating model for carbon emission reduction of battery electric passenger vehicle in use phase is built, and the carbon emission reduction is calculated and analyzed. The analysis result shows that the carbon emission reduction of battery electric passenger vehicles in use phase in China presents a basic trend of "high in the east and low in the west, high in the south and low in the north". The vehicle types that make greater contribution to carbon emission reduction are Class A, Class A00 and Class B, and the contribution to carbon emission reduction is obviously related to vehicle age. In the future, the continuous increase in the production and sales of battery electric passenger vehicles will further enhance the contribution of battery electric passenger vehicles to carbon emission reduction.

Optical networks-on-chip (ONoCs) using optical media to interconnect cores, have the advantages of high bandwidth, low power consumption and low latency. At present, mesh, torus and fat-tree are the most widely used topologies in ONoCs. Torus structure has the advantages of higher connectivity due to its annular structure. However, the traditional torus structure still has the problem of high end-to-end (ETE) delay. This paper proposes a new double-path ONoCs based on torus topology, named DPtorus. The DPtorus consists of two layers: one layer is a torus network divided into unit clusters, the other layer is an all-pass optical router. The dual-layer network is connected with each other by optical vias, so that there are two communication pathes between the source node and the target node in the network can be selected. The One is inter-cluster communication and the other is non-cluster communication. According to the comparison of the number of hop of the two communication paths, the communication path with fewer hops between the nodes is preferably selected to reduce the communication cost of the data packet. The network simulation results show that the DPtorus reduces latency and improves throughput compared to the torus network, resulting in better performance and enhanced reliability of network communications for the DPtorus.

The hydraulic system is an important part of the aircraft and is critical to flight safety. Therefore, the realization of fault diagnosis of the aircraft hydraulic system is of great significance to improve the safety and reliability of the aircraft. Aiming at the problem of insufficient fault data of the newly developed equipment, a virtual sample is formed through modeling, simulation and fault injection, which is combined with the real sample of the test bench to train the model. Aiming at the characteristics of uncertainty, nonlinearity and time-varying of hydraulic system, a fault diagnosis method of aircraft hydraulic system based on the combination of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) is proposed. The results show that the proposed hybrid algorithm improves the accuracy of fault diagnosis by 5%~10% compared with SVM and single LSTM, which proves the effectiveness of the algorithm.

In order to predict the key health parameters of complex equipment, this paper constructs an ensemble learning model based on LightGBM with a data-driven idea. Taking the hydraulic system as a typical research object, we have realized the prediction of its flow parameters. Under multiple working conditions, the R² reached 0.9988, and the prediction accuracy and time cost were excellent. Finally, using the relevant tool chain, we deploy the prediction model in the embedded environment to verify the effectiveness of the method.

As a large-scale marine fishery facility, marine ranch plays an important role. At the same time, in the current context of energy conservation and environmental protection, solar energy has the characteristics of clean, environmental protection, high efficiency and easy access, so it undertakes the important task of energy consumption control and energy conservation and environmental protection. However, under the influence of human and natural factors, solar panels inevitably have cracks, shielding and other defects. If it cannot be found and maintained in time, it will seriously affect the power supply efficiency and energy consumption control. Therefore, it is extremely critical to accurately detect the defects of the photovoltaic system in the marine pasture. Aiming at the problems of high false detection rate, unbalanced detection speed and accuracy in the original photovoltaic defect detection method, this paper proposes an improved method based on YOLOv5s. This method embeds CA (Coord Attention) attention mechanism in the backbone network, and uses BiFPN (Bidirectional Feature Pyramid Network) to replace the original PANet to improve the feature fusion ability. The final experimental results show that the precision of the proposed method is 1.2% higher than that of the original algorithm, and the amount of parameters and computation are reduced by 25.3% and 16.5% respectively, and the reasoning speed is the same. Therefore, this method achieves the balance between speed and accuracy, and the reduction of model size also provides the possibility for its further deployment.

Inspired by the generative adversarial network EnlightenGAN, we propose a novel low-light image enhancement algorithm based on unsupervised learning global-local feature modeling (GLFE). The algorithm has two stages: generation and discrimination, including global and local feature modeling network, and global and local discriminator. First of all, Swin-Transformer Block is innovatively introduced in the global feature modeling of the generation stage. Its shift window mechanism can conduct long-distance feature dependence modeling of the input image with less memory consumption, and well extract the features of image color, texture and shape, so as to effectively suppress noise and artifacts. Secondly, in the local feature modeling, the U-net branch based on grayscale spatial attention guidance can well capture the detailed information such as image edges and corner points. In the discrimination stage, deep and shallow feature fusion modules are added to enhance the discrimination ability, and the inconsistency is suppressed by learning the spatial filter contradiction information, so that the shallow representation information and deep semantic information guide each other, and the reasoning is almost no overhead, so that the enhanced image has uniform illumination intensity. Thanks to the synergistic effect of the above three innovative aspects, GLFE can achieve greater performance improvement compared with EnlightenGAN. Compared with the existing low-light enhancement algorithms, the algorithm achieves SOTA level performance in several public datasets.

Distortion-free enhancement on images captured under low-light conditions has always been a challenging problem in computer vision. To address these problems, this paper proposes an end-to-end network, which can learn the map way of low-light images to normal-light images from unpaired low-light and normal-light datasets. The network is consisted of dual branches, the upper branch is a refinement branch focusing on noise suppression, and the lower branch is a global reconstruction branch based on light-weight Transformer. The discrimination network adopts the multi-scale discrimination structure of feature pyramid to enhance the global consistency and avoid local overexposure. Qualitative and quantitative experimental results show that the proposed method can effectively suppress the generation of artifacts and noise amplification of enhanced images.

Taking the Sanjiang Plain as an example, this paper selects typical profiles A-B along the groundwater flow direction. Based on the groundwater flow field in 1980 and 2019 dry seasons and the duration monitoring data of typical monitoring wells from 2018 to 2021, this paper applies GIS analysis, statistical analysis and comparative analysis methods to study the characteristics and causes of the difference in the dynamic changes of groundwater in the east and west directions in the Sanjiang Plain. The results show that: compared with 1980, the groundwater level in Jiansanjiang area has dropped 5-14m in total. The decline rate is reduced from 0.59m to 0.29m per year, and the groundwater level in the central and western plain area shows a small decline or upward trend. The groundwater regime in the central and western plain area is sensitive to artificial exploitation and atmospheric precipitation. The fluctuation range of groundwater level in the mainland can reach 9 meters, and that in Jiansanjiang area is mostly less than 1 meter. Influenced by the paleogeographic sedimentary environment, the characteristics of shallow surface "west sand and east clay" and the spatial differences of aquifer water richness are the main reasons for the significant inter annual and intra annual dynamic differences of groundwater in the east and west.

With the development of the times, "blockchain + innovative social governance" has provided infinite impetus for the development of modern society. As an important part of social and public security, the importance of environmental sanitation to explore its digital value is becoming more and more prominent, so making data networking on the cloud has become a key step. There are many technologies for microbial detection, and the use of ATP (adenosine triphosphate) hygiene detection technology has become a relatively mainstream detection method with its technical advantages such as rapidity, efficiency, convenience and reliability. This paper designs an IoT microbial detector using ordinary silicon photodiodes as light detectors, which reduces the size of the detector, realizes the portability of the instrument, and broadens the application scenarios of the ATP detector.

In the context of the problem of excessive MAPE values in the method of filling in missing big data of electricity consumption information, a method of filling in missing big data of electricity consumption information based on variational self-encoder is designed. Optimising the electricity information pre-processing model, store and manage the various types of raw and application data collected in a classified manner, define the objective function as the algebraic sum of the squared measurement errors, construct an electricity big data tensor filling model, treat missing values as variables, and design a missing filling method based on a variational self-encoder. Experimental results: The mean value of MAPE of the big data missing fill method for electricity consumption information in the paper is: 38.514%, indicating that the performance of the designed big data missing fill method for electricity consumption information is better after fusing the variational self-encoder.

An aircraft four-dimensional(4-D) trajectory prediction method is proposed for the pre-tactical stage. The method is based on point mass model and total energy model to establish aircraft dynamic model and wind correction model, and use the wind forecast data to correct the trajectory, and gradually deduce the 4-D trajectory by computer recursive method. Based on the flight plan and the aircraft performance data in BADA3.11, the 4-D trajectory of the flight can be predicted. Finally, the accuracy of the prediction method was verified for flight KLM888 on the VHHH-EHAM international long route, where the difference between the predicted trajectory flight time and the planned flight time was only 0.73%.

Litchi fruit is rich in water and nutritional value. Machine picking can greatly reduce production costs. But if mechanical damage occurs during picking, it will seriously affect subsequent storage and transportation. In order to reduce the mechanical damage caused by excessive clamping force of the end effector on litchi, this study used the finite element model to analyze the mechanical damage caused by different clamping methods on the pericarp, sarcocarp and putamen of litchi under the basic clamping conditions. The results showed that the effect of clamping the fruit below the waist by double finger lifting was excellent. The maximum deformation of litchi was 42.06μm, the average equivalent elastic strain was 3.90 × 10^-4 and the maximum equivalent stress was 62857 Pa. This method effectively reduced the maximum stress and average strain in the process of grasping and reduced the mechanical damage caused by the end effector to the pericarp and sarcocarp. At the same time, it can provide some reference for the similar characteristics of fresh fruit picking, sorting, transportation and packaging.

With the increasing urban congestion, research on the carrying capacity of urban road network to urban vehicle ownership can provide reference for urban road network planning. From the perspective of urban road network, by considering the influence of the service level of public transport system and the proportion of vehicle on the road, this paper adopts the time-space dissipation method to build an easy operating urban vehicle ownership carrying capacity evaluation method, and applies it to the urban road network of Guiyang for analysis. It is found that the model is simple and easy to operate, and the saturation of urban road network can be greatly improved by increasing the number of motor vehicle passengers and the service quality of public transport system. If Guiyang wants to improve the saturation of road network without changing the road network area, that is, without building new roads and transport infrastructure, it should vigorously develop public transport and improve the public transport sharing rate.

The network structure and line operation have become increasingly complex with the continuous construction of the urban rail transit network (URTN). However, the traditional graph theory has challenges in meeting the abstraction and analysis. The complex network theory was used in this research. The topology analysis model, called Space S Model, was built based on the improved Space L Model. The database includes 56 URTNs around the world. The topology properties were collected. Then all URTNs were classified into two forms and three classifications by cluster analysis. The discriminant function based on properties was built and could evaluate the development of any URTN. The feasibility of the classification method was verified according to the analysis of the Nanjing Metro network.

Traffic flow prediction can provide data support for urban traffic planning, vehicle running conditions, and other fields. Based on the urban survey data, a multi-scale traffic flow model is proposed, and the characteristics of the traffic flow in Xi'an are analyzed. The results show that the relative error of the proposed multi-scale traffic flow model is less than 15% when predicting the traffic flow of the whole urban road network.

To improve the prediction accuracy of ship trajectory, the Transformer model is applied to predict ship trajectory. The pre-processed AIS data are used for the training and testing of Transformer model, and the AIS data from port area of Ningbo Zhoushan are used for validation. The results prove that the Transformer model is suitable for ship trajectory prediction, and the mean absolute error and mean square error of the Transformer model prediction results are smaller than those of other prediction models. The prediction results show that it is feasible to use Transformer model for ship trajectory prediction, and it also provides a kind of data preparation for collision avoidance warning for ships.

It is very important to investigate dynamic interactions of the bridge-moving-load system. There are some difficulties in modeling and solving the dynamic equation of a moving-mass-bridge coupling system because the dynamic coupling behavior usually changes with time.The governing equation of the coupling system between the bridge and moving mass is formulated using the mode superposition technique and Bernoulli-Euler beam theory. Simulink package is used to build simulation model based on MATLAB software. Compared with ABAQUS solutions, simulation accuracy and efficiency are improved by the models presented. The present simulation model has very wide adaptability. It can not only analyze the problem ofa single moving mass but also deal with multiple moving masses.

For improving the behavior of the traditional genetic algorithm (GA) for solving job shop scheduling problems(JSP), a modified genetic algorithm(MGA) is proposed. The reform is reflected in the generation of the initial population and the decoding methods are different. The initial population is produced randomly and the job with the largest remained total processing time is preferred. In decoding, the machine can be selected according to the shortest processing time of the currently available machine and the earliest completion time of the currently available machine. At the same time, considering that dynamic events are inevitable in the process of job shop scheduling, it is more practically in application than static scheduling. Dynamic events are classified according to the degree of disturbance. The change in completion time and the change of processing machine are selected for some operations. Finally, the modified algorithm and the technique of dynamic classified job shop scheduling are proved to be feasible according to the case analysis.

With the development of the industrial big data, research on data-driven industrial equipment fault diagnosis has received more and more attention. However, in the process of data acquisition, the equipment failure frequency is low, and the data set becomes serious imbalance. In order to solve this problem, we propose an unbalanced data generated method based on GAN. And in order to improve the accuracy of equipment fault diagnosis, we propose a fault diagnosis method based on CNN-LSTM, which can effectively utilize the spatial and temporal characteristics of data.

In systems biology, gene regulatory networks can well reveal complex biological systems and dynamic biological processes. Traditional gene regulatory network reconstruction methods ignore dynamic biological processes. A gene regulatory network reconstruction method based on transfer entropy and single-cell dynamic time series data is proposed to solve this problem. And from a theoretical point of view, the possibility and rationality of transfer entropy in the application of big data and causal reasoning are proved in detail. Firstly, the method construct pseudo-time series dynamic gene expression data based on trajectory inference. Secondly, The transfer entropy is used to calculate the directional transfer information between paired genes, and screen the major gene regulatory relationships . Finally, remove the indirect gene regulation relationship according to the data processing inequality, and construct the gene regulation network . This method takes the single cell sequencing data of early mouse embryo blood development as an example, and selects TENET and DynGENIE3 algorithms as comparisons, which proves the feasibility and effectiveness of this method from the theoretical and experimental perspectives. The experimental results not only identify the key cell development regulatory factors and regulatory relationships, but also consume less time. It can not only provide important reference value and assumptions for biological disturbance experiment, but also reduce the time cost and research cost of biological experiment.

Accurate traffic prediction can help residents plan travel routes, relieve traffic congestion, and reduce traffic accidents. At present, most traffic flow prediction models do not fully consider the spatiotemporal dependence of traffic flow data, and cannot explore the deep temporal and spatial correlation of traffic flow. Therefore, this paper proposes a traffic flow prediction model based on ConvLSTM and fuzzy clustering (i.e. ConvLSTM-GK model). The model is structured in feature generation module, ConvLSTM module and GK cluster analysis module. The feature generation module is used for the construction of data features; the ConvLSTM module can fully integrate the spatiotemporal information to make the feature extraction effect better, the one-dimensional convolution in ConvLSTM is used to extract spatial local information, and LSTM is used to extract time series information; according to the similarity between the data to be predicted and the training data of the ConvLSTM module, the GK Cluster Analysis module estimates the error of the data to be predicted based on the similarity between the data to be predicted and the training data of the ConvLSTM module, and compensates for the error of the prediction results of the ConvLSTM module to improve the prediction accuracy of the model. Experimental results show that under the training of PeMS dataset, the prediction accuracy of ConvLSTM-GK model in mean absolute error (MAE) and mean absolute percentage error (MAPE) is 14.26 vehicles/5 minutes and 17.77%, respectively, which is higher than that of LSTM, CNN-LSTM, ConvLSTM and AT-Conv-LSTM models, which proves that the ConvLSTM-GK model has performance advantages in traffic flow prediction tasks.

A method of object detection and ranging based on YOLO for binocular vision is proposed in this paper. Firstly, the object is detected by YOLO, and then the feature points are matched by stereo matching algorithm. The algorithm is proved to be effective by experiments, and it has certain engineering significance to realize stereo ranging by using vision technology.

The current traditional data distribution algorithms lead to poor distribution due to the lack of mapping processing of data. In this regard, the research of data distribution based on AHM attribute hierarchy model is proposed. The AHM structural model is constructed, the data is mapped and processed, and the slice key is specified as the system index basis to realize data slice, and finally the data distribution strategy is proposed by combining the optimization method of virtual node distribution method. In the experiments, the data distribution performance of the proposed method is verified. The analysis of the experimental results shows that the data connection query time is shorter when the proposed method is used for data processing, and it has a more excellent data processing performance.

With the gradual maturity of cloud computing technology, the power data center begins to migrate to the cloud platform, so there are many new operation and maintenance requirements, such as scalability, security, configurability, etc. In order to meet these requirements, this paper proposes an automatic operation and maintenance system of power data center based on cloud platform. This system uses cloud computing platform and virtualization technology to model the hardware resources, data resources, configuration resources, security resources, etc. in the power data center, digitize the resources, and realize the automatic management of resources through the system's data collection, analysis, and transmission, so as to achieve the goal of automatic and efficient operation and maintenance, It can better meet the new demand in the operation and maintenance of power data center.

An improved RRT* algorithm based on dynamic bias angle and path simplification is proposed to address the problem that the traditional RRT* algorithm requires a large computational effort to obtain a better path. The algorithm uses a dynamic bias angle expansion strategy to enhance path planning guidance while avoiding local optimality, and then performs path simplification to eliminate redundant points and reduce path length. Finally, the improved RRT* algorithm is used to smooth the processed paths using cubic B-sample interpolation to extend the life of the robot arm. The simulation results demonstrate that the path generated by the improved RRT* algorithm is of higher quality, and the robot arm completes the grasping and placement of the object with a better path around the obstacle after the smoothing process is added.

Aiming at the problem that the defect detection algorithm of ceramic tile is not much and the performance is not enough to meet the industrial requirements, this paper proposes a ceramic tile defect detection algorithm based on improved yolov5. The improvement scheme proposed in this paper includes the introduction of CNB module and Ghost convolution, and the improvement of spatial pyramid pooling. Through repetitive experiments on the collected tile data set, it can be confirmed that the proposed modules have obvious improvement effects on the accuracy, real-time and lightweight of the model. The mAP of the improved algorithm in this paper reaches 73.1 % under the premise of meeting the industrial demand in terms of speed and memory usage, which is 6.2 % higher than the original yolov5 model.

Some applications require fast queries on big tables whose rows are as large as one billion. If we store table data without any compression on disks, the operation of loading the table data into host memory itself costs quite a long time. To speed up queries, we resort to data compression. Table data is first compressed before being saved to disks. In the query stage, compressed data is loaded into host memory, decompressed and then accessed. To speed up decompressing, we make use of the massively parallel capability of GPU devices. In order to make full use of the GPU computing resources, GPU kernels should avoid divergent execution as much as possible, and should make efficient use of the GPU local memory. Guided by these criteria, we have designed a GPU decompression algorithm. Its basic idea is to decompose the decompression task into a few sequential basic operations, and then accomplish each basic operation parallel by using GPU threads. Experiments showed that the average throughput rate of the decompression algorithm implemented with OpenCL could reach to 13.12 GB/s when using AMD RX 6600. The reduced loading time and decompression time significantly improved the query speed in the query stage.

In order to improve the efficiency of the audit of toll evasion, a toll evasion prediction model is established based on historical ETC(Electronic Toll Collection)portal flow transaction data. Now it has become the most perfect Internet of Things system for expressways, which can be used to strengthen the management of traffic toll supervision. First, the complete trajectory data set of vehicle travel was constructed based on the ETC transaction data, the trajectory data set of vehicle travel were divided into multiple sections for analysis. Second, density clustering was used to partition toll evasion data set. Finally, a random forest algorithm was used to construct a prediction model of toll evasion behaviors. The correct prediction ratio of the model for toll evasion behaviors was 91.3%.To avoid training over-fitting, ETC plate recognition data is used for proving and accurate discovery of toll evasion behaviors validates the feasibility of the method.

Data poisoning attack has been one of the most prominent threats for data-driven machine learning model. Specifically, in the field of recommendation system, an attacker could manipulate recommendation results by injecting some crafted fake data into recommendation model. In this paper, a data poisoning attack method based on dynamic adaptation of data and model is proposed for the recommendation system, referred to as dynamic attack, which solves the problem that the fake data fails to keep aggressive due to difference between models. Experimental results on the two real datasets, MovieLens-100K and MovieLens-1M, show that dynamic attack outperforms existing heuristic-based attacks and the average attack success rate is increased by more than 10 times.

As digital integrated circuit(IC) design techniques and corresponding product manufacturing processes improve, more and more electronic products are moving towards miniaturisation and high concentration, but this is what makes circuit test generation difficult. Since there is a strong link between digital IC test generation and fault diagnosis of digital systems, neural network techniques are applied to fault diagnosis to enable test generation algorithms to accomplish goals such as fault activation and fault propagation. This paper is based on a neural network algorithm to solve circuit fault test sets, generate test vectors and simulate the fault detection process until all fault points have been detected. By comparing the circuit test simulation experiments of the neural network algorithm and the traditional algorithm, it is found that the number of circuit faults increases and the test time of the neural network algorithm is longer than that of the traditional algorithm, but the fault coverage rate is higher than that of the traditional algorithm, indicating that the neural network algorithm can effectively improve the correct fault detection rate although it increases the test time.

In order to efficiently predict and analyze massive traffic big data and improve the intelligent level of road traffic rate and urban traffic, a high-precision parallel convolutional neural network traffic flow big data prediction model based on deep learning is proposed The model first preprocesses the data to obtain an effective data set, converts one-dimensional time series samples and images with regular time intervals into two-dimensional pixel grids with one-dimensional time and one-dimensional location, constructs a parallel convolutional neural network model to predict the traffic flow through a road section, and uses prediction factors to model the traffic flow data The experimental results show that, compared with other models, the model proposed in this paper is superior to the comparison method in terms of average absolute error, average relative error and root mean square error.

Few-shot learning requires fast learning and adaptation during the learning process, which has been seen as a very challenging problem because of its stringent training requirements. With the development of deep learning, meta-learning is increasingly applied to solve few-shot problems. Model-agnostic meta-learning (MAML) is applied to model-agnostic tasks by finding common initialization parameters applicable to the model through both inner and outer loops. However, the generalization performance of MAML is not strong, and its inner loop is not fully functional. Therefore, we add learnable hyperparameters to the inner loop to make the model training results more relevant to the current task and reduce the number of layers of gradient updates without unnecessary computation. Experimental results show that the inner loop hyperparameter learning based on feature reuse (HLFR), finds better initialization parameters compared to MAML.

Adaptive array signal processing is widely used in modern radar anti-jamming design. In this paper, simulation research is carried out on the deficiencies of the current research on anti-noise jamming of radar adaptive array. First, starting from the mathematical model of radar adaptive array based on LMS algorithm, the optimal weight vector of adaptive array based on LMS algorithm is derived and established. On this basis, numerical simulation is carried out to analyze and study the performance of anti noise jamming. The research of this paper has certain reference significance for the design of future radar anti-jamming.

Chinese Spell Check (CSC) aims to detect and correct spelling errors in Chinese text, almost all of which are related to phonetic or visual similarity. Large-scale pre-trained models (PLMs) are currently making substantial progress on the CSC task. However, when correcting errors, PLMs tend to select those words that are semantically sound or expressively fluent, sometimes ignoring pronunciation similarities. Meanwhile, the models lack knowledge of pronunciation differences. To address this problem, we propose a multi-task learning model to help enhance the CSC task. The auxiliary task is to estimate the degree of pronunciation gap between the original input and the corresponding correct text from the granularity of each word. Specifically, we use the edit distance of Pinyin to measure the degree of pronunciation discrepancy. The edit distance scheme we use is modified, due to the specificity of the Pinyin structure. Experiments on a open available benchmark dataset demonstrate the effectiveness of our strategy.

Z-FFR (Z-Pinch drive fusion fission hybrid reactor) is an important base-load energy route with high research significance for early global carbon neutrality. Z-FFR is a nuclear device and its reliability is a central and critical part. This paper attempts to use big data in collaboration with artificial intelligence and industrial software for Z-FFR reliability analysis. Reliability data analysis mining is carried out by artificial intelligence, which is divided into three main levels: foundation platform layer, data management layer and data application layer. The foundation platform is the server layer, including application server, database server, file server and integration interface. The data management layer is the database layer, which provides data management functions for the basic information of Z-FFR, fusion reliability data, blast chamber reliability data, sub-critical reactor reliability data, and reliability data of the whole Z-FFR machine and its direct components; the data application layer is the user layer for data query, reliability data analysis, two-dimensional graph display of reliability data, and three-dimensional graph display of reliability data, which is mainly supported by mainstream This part is mainly supported by mainstream industrial software. The work in this paper provides data support for quality prediction of Z-Pinch and hybrid stack devices, thus improving the quality stability and consistency of the devices, and laying the foundation for the improvement of Z-FFR's refinement management, precision manufacturing, quality assurance and information management.

With the development of natural language processing technologies such as deep learning and large-scale pre-trained language models, intelligent question answering robots have become more and more popular in industry. The algorithms of text classification and text matching are important for robot automatic response. However, for text classification, the number of categories is always fixed. For text matching, the sentence pairs for training are difficult to collect. In this paper, we propose a novel text matching method to solve the above two problems, which can be trained with only the text classification dataset. The proposed model can have better recognition ability for the newly added appropriate problem categories without retraining. On the basis of pre-trained model, we first conduct further pre-training with contrastive learning, and then conduct multi-task fine-tuning (core sentence vector matching and contrastive learning). The finally obtained model can benefit from both the text classification method and the text matching method.

Millimeter wave radar often suffers from unstable operation of internal components, obstruction interference of external metal plates, large vehicles and other obstacles, and uneven impact of road vehicle target echoes, resulting in abnormal mutation or even false invalidity of vehicle targets in one or more frames. Aiming at the problem of abnormal vehicle target detection due to the impacts of various internal and external factors, this paper proposes a vehicle trajectory smoothing method based on Kalman filter. In this paper, using statistical and analytical methods, the measurement error of vehicle target data detected by millimeter wave radar at intersections is quantized in spatial domain, and the function of vehicle target error measured by radar is obtained. This research algorithm combines the above error function of radar detecting vehicle target with the classical Kalman filtering algorithm, so that the vehicle trajectory can more truly reflect the normal vehicle trajectory. The experimental results show that the algorithm weakens the noise interference of radar data obtained due to the impacts of internal and external factors, greatly improves the accuracy, authenticity and stability of millimeter wave radar, and promotes the development of traffic information analysis and processing.

We propose a stable portrait video matting method without any auxiliary inputs like pre-captured background or trimap, but achieves competitive or even better performance compared to existing methods. In this paper, we observe that depth information has significant effect on human matting task since the target is at smaller depth with high probability especially for conference scenes. To this end, we design two decoder branches for coarse depth estimation and matting respectively. Furthermore, we utilize a temporal-spatio consistency module (TSCM) to improve temporal coherence and enforces our network to pay more attention to the foreground at smaller depth. Moreover, existing video matting evaluation metrics are almost derived from image matting, we introduce a simple but efficient algorithm to check flicker clips of the alpha video which reflects its stability.

The connectivity of networked systems is often dependent on a small portion of critical nodes. Network dismantling studies the strategy to identify a subset of nodes the removal of which will maximally destroy the connectivity of a network and fragment it into disconnected components. However, conventional network dismantling approaches focus on simple network which models only pairwise interaction between nodes while groupwise interactions among arbitrary number of nodes are ubiquitous in networked systems like integrated circuits. Groupwise interactions modeled by hypernetwork introduce higher order connectivity patterns, which limits the application of conventional network dismantling methods on hypernetwork. In this brief, we propose HyperCI, a higher order collective influence measure for hypernetwork dismantling. It considers the node co-occurrence characteristics and higher order influence ability both introduced by hyperedges in hypernetwork. We evaluate the effectiveness of our proposed HyperCI on six real world hypernetworks including integrated circuits and citation networks and the results indicate our proposed HyperCI outperforms baseline network dismantling methods for both simple network and hypernetwork.

Publisher's Note: This paper, originally published on 2 May 2023, was withdrawn on 9 May 2023 per organizer and author request.

With the rapid development of the domestic Internet of Things, my country's self-designed SM2 public key cryptographic algorithm has been widely used in the integration of the Internet of Things and blockchain data of blockchain chips. However, the current SM2 public key cryptography algorithm has the problems of low calculation speed and excessive consumption of hardware resources. In view of the above problems, this article applies the Kogge-Stone adder method to solve the problem of low calculation speed; based on the characteristics of the AVR architecture, a method is proposed to optimize the inner loop operation of the Montgomery modular multiplication to reduce the consumption of hardware resources. Experiments show that the improved performance has a 26.2% performance improvement compared to before the improvement. The computing efficiency of the large integer modular multiplication operation is effectively improved, the resource consumption of hardware implementation is reduced, the balance between computing time and hardware overhead is better achieved, widely used in blockchain chips.

Inspired by Transformer, this paper proposes a new attention-based feature fusion network, which effectively combines template features and search region features using attention alone. Specifically, the method includes a contextual enhancement module based on multi-headed self-attention and a cross-feature enhancement module based on crossattention, and finally the two features are combined using the residual structure to effectively enhance the features. Experiments show that our tracker achieves very good results on the GOT-10k benchmark. It runs at approximately 45fps on the GPU, which achieves the real-time requirement.

Sparse signal recovery algorithms can be used to improve image quality of target with sparse property in spatial domain. In order to improve the imaging quality of MIMO Radar, an improved smoothed L0 norm (SL0) sparse signal recovery algorithm is proposed in the paper. Firstly, the sparse imaging model of MIMO radar is established, the imaging problem is transformed into the optimization problem of minimum L0 norm; Then a negative exponential function is proposed as smoothed function to approximate the L0 norm. Finally, a comparison correction step is added to ensure that the search direction follows the fastest descent direction and the gradient projection algorithm is used to obtain target image. Simulation results show that MIMO radar imaging quality obtained by the proposed algorithm has great improvement compared to several popular algorithms.

Transformer bushings are one of the important insulation components of transformers and directly affect the operating conditions of transformers. For fault diagnosis of transformer bushings, historical fault data are collected as the original samples and the fault samples are expanded using SMOTE technique for the unbalanced sample set. The weights and thresholds in the BP neural network (BPNN) are optimised based on the cuckoo Search (CS) algorithm to avoid the randomness of the selected parameters. To demonstrate the effectiveness of the algorithm in the paper, data samples processed by the Synthetic Minority Oversampling Technique (SMOTE) are used as the training set of the CS_BP model for training and comparison with BPNN, genetic algorithm optimised BPNN (GA_BP) and particle swarm algorithm optimised BPNN (CS_BP) to analyse the accuracy, time required evaluation indicators for fault diagnosis under different models. The results show that the SMOTE_CS_BP diagnostic model can reach convergence quickly and has the highest accuracy compared with other models, and the model proposed in this paper provides an effective method for determining the insulation performance of transformer oil-paper bushings.

In order to solve the problem of multi-scale in a single image gathering crowd counting, a new crowd counting network based on the fusion of dilating convolution pyramid and context attention mechanism (DCPCANet) is proposed. With the first ten convolutional layers of VGG16 as the front-end network, an dilated convolutional pyramid fusion attention mechanism module (AMP) is proposed, which is introduced into the three-level upsampling feature fusion module to extract fused multi-scale features, and the AMP module stack is used as the back-end network to capture and fuse multiscale features, The context attention module (CAM) is used to generate the feature map with weight, and high-quality crowd density map is output at the same time. Three mainstream public data sets are adopted, ShanghaiTech PartA,ShanghaiTech PartB,UCF_CC_50. Compared with the previous algorithm, the MAE of the UCF_CC_50 dataset is reduced by 11%, which preliminarily verifies the accuracy and robustness of the model.

To address the problem of inaccurate estimation of model parameters by non-end-to-end image defogging algorithms based on deep learning, and the problem of image spatial information retention by current end-to-end image defogging algorithms based on deep learning, this paper proposes a single-image defogging model DeHRNet based on High-Resolution Net (HRNet). DeHRNet is divided into branches with different resolutions, and branches with different resolutions are connected in parallel and multi-scale fusion is performed at the end of each stage. This paper adds a new stage to the original network to make it better for image defogging work. The addition of a new stage to collect feature map representations of all branches of the network by up sampling to enhance the high resolutions representation, rather than using only feature maps of the high resolutions branches, makes the recovered fog-free images more natural and significant. The experimental results show that DeHRNet has a significant de-fogging effect on fogged images.

Remote sensing (RS) technology plays an increasingly dominant role in Earth observation. However, cloud contamination is a serious hinder in analysis of RS images. Aiming at the problems that the present cloud removal methods remain cloud residues and lose ground scene details in the restored image, We propose a 3D-attention and residual dense based generative adversarial network (3DA-RDGAN) to remove the cloud. We first introduce the residual dense block (RDB) into the generator, so it learns plentiful local characteristics via dense connection and integrats different levels of features via residual learning to restore ground object information. secondly, a 3D attention module (3DAM) is inserted to each RDB to infer the 3-D attention weights for the feature maps without adding parameters of the original network. Under the guidance of attention loss, 3DAM effectively helps the network pay more attention to the cloud areas and discover the difference between cloud and ground scenes. The proposed 3DA-RDGAN is tested on the open source RICE dataset, and its effect is compared with several other existing deep learning methods. The results indicate the superiority of 3DA-RDGAN in cloud removal for RS images.

In this paper, a new anomaly detection architecture, TAGAN, is proposed. By combining the reconstruction approach with the prediction approach, TAGAN is used for anomaly detection over multivariate time series. A new loss function based on Wasserstein distance with gradient penalty is introduced in the reconstruction branch, and attention mechanism is introduced in the prediction branch. The performances of the proposed algorithm are tested over four real-world datasets (MSL, SMAP, SMD, and SWaT). Numerical experiments show that the proposed algorithm performs better than that of six anomaly detection algorithms.

Aiming at the uncertainty of data collected in multi-sensor networks, a multi-sensor fusion technique based on the PSO algorithm is suggested to optimize the RBF neural network with the aim of reducing the uncertainty of data gathered in multi-sensor networks. The RBF neural network's weight and threshold parameters are modeled as moving particles, with vectors used to describe their positions. The PSO algorithm chooses the proper values for the parameters. The ideal parameter values of the RBF neural network are ultimately established following iterative training. It has been demonstrated that the PSO algorithm-based RBF neural network multi-sensor data fusion algorithm has higher fusion accuracy and shorter running times.

Aiming at school road condition monitoring, a semantic segmentation method based on lightweight network is proposed. The original DeeplabV3+ model is used to train the self-made data set to obtain MIoU values of various types, and then different trunk networks are replaced for comparative experiments. Then the original DeeplabV3+ model is adjusted and upgraded and optimized to conduct comparative training with the original model. Finally, the trunk network of the optimized DeeplabV3+ model was replaced for comparative experiments. The results show that the upgraded and optimized DeeplabV3+ model can accurately segment each category with MIoU of 88.7% and original MIoU of 77.8%, and the optimized model has better effect. MobileNetV2 performs better than Xception in any of the different trunk networks.

In order to accurately analyze the impact of different backgrounds on radar detection performance in radar system simulation design, this paper combines the average backscattering coefficient and directional propagation factor of ground clutter and sea clutter, establishes a model of the relationship between signal to clutter ratio and grazing angle, and obtains the clutter echo signal power. Finally, the correctness of the model is verified by calculating the influence of clutter on target detection probability.

This method proposes to first detect the features of moving objects driving on foggy roads, then compare and screen the detected feature data with the sample feature set through the Kalman filter algorithm, and then use the fuzzy k-means algorithm to classify and mark the cleaned vehicles, and at the same time obtain a more accurate feature vector interval, update the feature sample set in time, and finally send the analysis results to the control end and start the warning light for warning: in this method, the laser distance sensor detects the occlusion duration of the vehicle's tires, the occlusion interval time of adjacent tire groups, the vehicle running speed and the distance between the vehicle and the detector to judge the vehicle. This method can timely inform the driver of the information of the vehicle in front of the foggy road in advance and reduce the incidence of traffic accidents.

The study aims to improve the safety of the ADAS (Advanced Driver Assistance System) system under the challenging scenarios of single vehicle perception long tail or complex extreme driving conditions. In multiple vehicles following driving scenarios, when the remote target vehicle is blocked by the nearby target vehicle, the single-vehicle intelligence will be limited by its perception range and FOV (Field of View) and cannot respond in time, probably leading to collision accident or emergency braking thereby reducing the riding comfort. To solve these issues, we developed a V2X (Vehicle to Everything)-based ADAS system and designed a decision-making algorithm considering the interaction risk between multiple target vehicles by V2V (Vehicle to Vehicle) communication. Simulation and vehicle tests showed that compared with traditional single-vehicle intelligence schemes, the V2X-based ADAS system and the decision-making algorithm could provide about 1.75 s earlier time for braking, significantly improving driving safety. Reducing the braking force by about 30% can significantly improve ride comfort at the same time. This study can effectively improve vehicle driving safety under extreme conditions, which has theoretical value for the V2X application in mass-production vehicles.

This paper proposes a shared-taxi scheduling algorithm based on the insertion heuristic for online taxi-hailing services. First, service quality and operation cost are considered in the optimization objective of the shared-taxi scheduling system. There are two scheduling strategies for request insertion. One is the minimum waiting and detouring time, and the other is the minimum detouring and idling costs. The second strategy also considers a weight factor between detouring and idling costs. Then, the framework of the shared-taxi scheduling algorithm is built based on the classic cheapest insertion heuristic. Finally, based on the artificial data of a large-scale road network and requests, three groups of experiments correspond to different numbers of taxis under the two scheduling strategies.

In order to meet the air traffic demand and capacity balance in the pre-tactical management stage. It is necessary to scientifically adjust the flight pre-flight plan according to the change of airport capacity to avoid waste of flight slot resources and reduce flight delays. Firstly, this paper proposes a multi-level full-time zone conflict resolution algorithm design for the problem of demand imbalance in pre-flight planning. Then, within the scope of the underlying design, the time conflict resolution strategy and resource use conflict resolution strategy of the 5-minute time slice within the specified time period are studied respectively. Finally, the underlying strategy of the algorithm for flight advance planning optimization is validated using real airport operational data as an example, and the strategy is compared by calculating a relative deviation index (RDI) for a specific target value. The results show that the different strategies designed in this paper show different range of results in the process of multi-objective optimization, and the optimization effect is more balanced, which provides a reference for the underlying design of a variety of strategies for the realization of pre-flight plan conflict resolution and demand balance.

To solve the issue of standard traffic sign identification algorithms' poor detection accuracy, an improved YOLOv5 algorithm for traffic sign recognition is proposed. To begin, the YOLOv5 algorithm's backbone network is enhanced, and the original YOLOv5 network's backbone feature extraction network is replaced with the ultra-lightweight convolutional neural network MobileOne. To increase the model's focus on additional locations, the Coordinate Attention module's introduction, it incorporates location information into channel attention and performs multi-scale processing and feature fusion. Experiments demonstrate that the new lightweight network model is only 76% the size of the original YOLOv5 model, and the mAP on the dataset reaches 96.2%. This strategy significantly decreases the amount of model parameters and procedures required to ensure detection accuracy, while also improving detection speed and accuracy.

In view of the problems of complex modeling process, nonlinear and poor multi parameter coupling expression in the mathematical mechanism modeling of hydro generators, and the complex problems of wind and solar energy access and grid interconnection faced by large hydro generators, the advantages and disadvantages of mechanism modeling and data-driven are compared and analyzed, and the feasibility of deep belief network algorithm on hydro generator modeling is studied, A large-scale hydro generator model based on the data-driven method of deep belief network is established. A total of 129600 sets of actual data of a unit are used for model training and verification, and simulation tests are carried out under the no-load frequency disturbance of the unit to verify the effectiveness and stability of the model.

To study the effect of multi-mode combat part destruction, in this paper, we use AUTODYN simulation software to simulate the effects of dual-mode combatants on the shaping of the destruction element under different structures of the charge cover, charge height, and detonation method, and determine the optimization scheme of dual-mode combatants through orthogonal optimization. This study has a certain reference value for the structural design of multi-mode warheads and the study of damage effects.

By making deep analysis on the big data information of historical product, the technology of structured data classification based on deep learning can form knowledge for part capability type recognition and judgment, build a model of capability type recognition on the basis of deep neural network, and also develop a software system of capability type automatic recognition ,for assisting the recognition of part capability type. For part newly imported into the system, combined with its attached attribute information, this technology can automatically identify and recommend the capacity type, and forms a set of knowledge-based method for the recognition and judgment of part capacity type.

The blockchain technology based on digital signature provides a non-repudiation and non-tampering mechanism for digital storage and communication, which has been applied in various sub-fields of Information technology. The biggest problem of the current method is that the public-private key pair that identifies an entity's digital identity cannot be effectively mapped to its physical identity, which reduces the reliability of the signature. We propose Offline Blockchain embedded into material objects to achieve strong correlation between physical and digital identities, which ensures the full-process credibility and traceability for process tracing, especially for supply chain management and logistics.

With the development of electric power industries, the number of standards has grown rapidly. However, the contents of standard clauses extracted from various fields are often inconsistent. It is difficult for the staff to choose the proper standard clauses in their work. Therefore, it is significant to provide staff with consistent electric power standard clauses. This paper takes the standard documents in the electric power field as the data source, and focuses on how to find out the related but inconstant clauses in the documents. We take advantage of the entity relationships of knowledge graph to get the discrepancies of electric power standard clauses. The experimental results illustrate the good performance of the proposed method in terms of precision and recall. The precision rate is 76.45%, and the recall rate reaches 84.72%. In addition, the proposed approach could also provide a solution to the differential discrimination of standard documents in various industries.

Accurate short-term load forecasting has important guiding significance for power grid planning and operation. Most of the current mainstream prediction models are single prediction models, which are prone to noise interference and lead to poor performance. In order to give full play to the advantages of different models, a combined time series model based on a deep reinforcement learning algorithm is proposed. First, five-time series models are established respectively, and two dominant models, Bi-LSTM and Bi-GRU, are screened out according to performance indicators. Secondly, the DQN algorithm is used to optimize the weight coefficient of the output result of the combined model to improve the prediction accuracy. Finally, this paper uses Australian electrical load data for example analysis. The experimental results show that the prediction performance of the proposed combined model is better than that of the benchmark single model, and it has certain practicability.

Melanoma originates from the malignant transformation of melanocytes, it can gradually spread and metastasize. As the most aggressive and deadly type of skin cancer, melanoma posed a significant threat to patient health, but early diagnosis and intervention can improve patient survival and improve the prognosis of poor patients. Computer-aided diagnosis can help dermatologists to make early diagnoses of melanoma. UNet++ as a more advanced model among existing segmentation algorithms has practical value in the segmentation and diagnosis of melanoma, but after experiments, we found that its segmentation performance still has much room for improvement. In the study, we tried to improve the model performance based on the UNet++ algorithm, and a new convolutional neural network IDUNet++ (Inception Dilated UNet++) for melanoma skin lesion segmentation by introducing Inception block and dilated convolution was proposed. In the segmentation task for the ISIC2016 challenge skin lesion dataset, the model has further improved in segmentation accuracy compared with the original UNet++ model, which obtained 2.88%, 2.66%, 2.66%, 1.03%, 1.03%, and 1.66% in its six evaluation metrics of IoU, Recall, Precision, Accuracy, DICE coefficient and F1-score, respectively.

Accurate tourist flow forecast can provide important support and guarantee for scenic spot staff and tourists to make decisions or plans. Aiming at the current situation that most of the existing forecasting methods are tourist flow forecasting of a single scenic spot and cannot fully mine spatial dependency information between various scenic spots, this paper proposes AMGCN model, a multi-graph convolution-based tourist flow prediction model for multiple scenic spots. First, AMGCN defines different non-Euclidean correlations between attractions, encoded as different graphs respectively, and uses multi-graph convolution to extract spatial correlation features. Then, a channel-wise attention mechanism incorporating global spatial context information is used to improve the performance of a long short-term memory network (LSTM), which is used to learn temporal correlation features. AMGCN is evaluated on a real-world large-scale urban multi-spot tourist flow historical dataset, the results show that the model outperforms the baseline model on regression evaluation metrics such as MAE, RMSE, MAPE, and SMAPE.

Artificial intelligence technology is a strategic technology leading a new era of the revolution of science and technology along with the industry transformation. However, with the progress of technology and the boosting complexity of model, the problem of lacking of interpretability of artificial intelligence has become extremely acute. From theoretical research to practical application, the obstacle caused by its poor interpretability has become a major weakness of AI at present. This paper proposes to evaluate the interpretability via robustness and induced metrics. And then, this paper gives some existing image-classification networks a comparison and an evaluation of robustness and interpretability, and get the interpretation of models shown in heatmap.

A multi-task model with high robustness is indispensable to resolve the inter-target dependence of molecular properties. Beyond the newly proposed stacked single target method, this paper aims to introduce a novel multi-task learning framework in a deep regressor stacking approach to produce a higher predictive accuracy. For the two datasets of interest, OE62 and QM9, we compared the performance of the deep regressor stacking to the single-task model. The new model shows 1 to 4% error reductions compared with the independent models, and the prediction accuracies are consistently improved across different tasks. Further studies on the selective deep stacking scheme show an additional enhancement of the prediction accuracies, indicating a great potential of the deep stacked framework in forecasting the correlated properties.

Most problems in the power system are related to the temperature of electrical equipment. The abnormal high temperature of electrical equipment will not only cause damage to equipment itself, but also threaten the safety of people’s life and property. Therefore, patrol inspectors of power system will carry out routine inspection on electric equipment to ensure safety. Semantic segmentation for electrical equipment carries a big weight in the inspection. With the result of semantic segmentation of power equipment, patrol inspectors can quickly judge whether the temperature of power equipment is normal, and then take the corresponding action. For this reason, we propose the EdgeFormer, which is a typical end-to-end network using thermal image to segment electrical equipment. In our method, we employ an edge information extraction network to attain rich edge features to promote the segmentation performance in electrical equipment’s edges and interiors, and we also design the global enhancing module (GEM) to get rich semantic information. Besides, we also propose a feature inserting module (FIM) to fuse the feature maps from different stages together. Lots of experiments have been conducted on the LS-ETS dataset and the results show that our EdgeFormer has achieved the best performance.

In order to clarify the relevant elements, stage characteristics, evolutionary context and frontier trends of diversification strategy research, and further guide enterprises to explore diversification strategies, the bibliometric method was used, and the literature elements including keywords were analysed by COOC13.4, VOSviewer1.6.13 and other software. Combined with existing papers, the visualized knowledge graph was drawing, and the hotspots and trends were analysed. The research content of diversification strategy presents the characteristics of multi-dimensional and staged. Scholars have conducted research on the motivation and influencing factors from multiple dimensions such as the internal and external environments of enterprises, the classification of diversification strategies, and the classification of different industries. Through visual analysis, the research process of the diversification strategy in the past 20 years was divided into the following three stages, namely "research boom", "gradual saturation" and "rich development". The main trend of future diversification strategy research is that, combining personalized internal environment and changing external environment, constantly enriching its influence way.

Aiming at the problem that the existing air target tactical intention recognition methods cannot use the sequential characteristic and the multiple dimension of sensor data effectively, an air target tactical intention recognition method based on the fusion deep learning network model is proposed. According to the specific combat task, we select appropriate target features and intentions to construct the feature space and intention space of the target. Then we obtain the real-time target state data as time feature sequences. We assign weights to different features through attention mechanism, use convolutional neural network (CNN) to obtain local trend features of the time series and shorten the time series, use time convolutional network (TCN) to extract short-term local features in the data, and use bidirectional gating recurrent unit (BiGRU) to extract long-term dependence on data and considerate the future information to achieve combat intent recognition of air targets. Through contrast experiments with other models, the effectiveness and advantages of the proposed method are demonstrated.

More and more researchers have paid attention to the tracking of visible-thermal infrared (RGB-T). How to fully exploit the complementary features of visible and thermal infrared images and fully integrate them is a key issue. After extracting image features, many researchers simply fuse the features by adding, connecting operations or designing fusion modules. However, these methods ignore the effects of different levels of fusion features on target modeling and specific feature extraction. In this work, we propose a RGB-T tracking network (MRLRNet) based on feature mutual reinforcement learning and resampling. Specifically, we design a feature mutual reinforcement learning module, which combines different layers of features to achieve progressive fusion. After each layer feature is extracted, the aggregation features are used to enhance specific modal features to achieve better specific feature representation and reduce noise and redundancy features. At the same time, we design a resampling module, which calculates the offset of two adjacent frames by phase correlation operation, and recalculates the Gaussian sample points to solve the problem of ground target loss caused by sudden camera movement. A large number of experiments on three RGB-T tracking datasets, GTOT, RGBT234 and LasHeR, demonstrate the effectiveness of this method.

Deep and convolutional neural networks have performed well in capturing speaker characteristics, while the ECAPA-TDNN model has demonstrated outstanding performance in both the fields of speaker validation and speaker diarization. Within this essay, during the speech segmentation stage, we uniformly redivide the speech on multiple time scales based on the oracle voice activity detection. Meanwhile, we fine-tune the ECAPA-TDNN architecture by adding a RepVGG module to extract more abundant features, then aggregate all of the outputs. Finally, we use DOVER-Lap to integrate the results obtained after the clustering of multiple schemes in a way to obtain the final temporal labeling. The best results achieves 1.91% of the diarization error rate on the classical AMI conference corpus.

Objective This study analyzes, summarizes and summarizes the hotspots and trends in the research literature on technological innovation efficiency, and provides a theoretical basis for future scientific research. Methods We searched the papers on "technological innovation efficiency" from 1999 to 2022 through the database of China Knowledge Network (CNKI), and used COOC13.4 software VOSviewer1.6.13 to analyze the key words, research institutions and authors in this field. The visual scientometric analysis was conducted using COOC13.4 software VOSviewer1.6.13 to visualize key words, research institutions and authors in this field. Results A total of 1,228 papers were included. The author found that the overall trend of the number of publications was on the rise through the knowledge mapping analysis, and the cooperation between the publishing institutions was not close. Research hotspots include technological innovation efficiency, technological innovation, data envelopment analysis (DEA), high-tech industry, enterprise scale, innovation environment, and influencing factors. Conclusion The research on technological innovation efficiency in recent years is divided into three main development stages respectively, the budding exploration stage, the rapid development stage, and the stable improvement stage. Future research needs to strengthen the cooperation between different institutions and authors. This paper predicts that this research field will be broadened in the future and the research on technological innovation efficiency will develop toward a diversified trend.

Intent recognition is the first part which needs to be accurately recognized in the conversation between ai and the user. Due to the diversity of user intent, it is difficult to manually define all categories of intent in advance during the training phase. The inability to distinguish undefined intent categories can cause the ai to keep replying to the user with seriously wrong answers, which can greatly reduce the user's sense of experience. Therefore, it is very necessary to realize the recognition of seen intention categories and the distinction of unseen intention categories. In this paper, a model with Automatic probability threshold which based on the bert model is used to ensure accurate recognition of seen intentions while distinguishing unseen intentions. Due to the inadequacy of training samples, the automatic probability threshold model in the paper uses text vectors from two bert models containing different dropout parameters as the training set, which can improve accuracy of the model.

To address the problem of inaccurate identification of entities with fuzzy fertilizer knowledge areas boundaries. In this paper, we propose a named entity recognition model based on BERT with adversarial training. Disturbance factors generation by introducing FreeLB adversarial training method, combined with the vectors of the word embedding layer in BERT to form the adversarial sample. Improving model recognition of boundary ambiguous entities by adversarial training. Experimental results show that, the model proposed in this paper achieves 89.77% accuracy, 93.72% recall and 91.70% F1 on the dataset. compared with other models, the accuracy, recall and F1 values are improved by 1.75%, 0.53% and 1.17% respectively.

Point cloud is the popular researches in the industry, and the research on partial point cloud is favored by many researchers. In the acquisition process of original 3D point cloud data, there are partial or sparse problems due to occlusion, lighting and other reasons, which will lead to deviation in downstream tasks. We propose a network MDPCN using multiple decoders, which can make the partial point cloud complete. The network uses deep learning to get multiple features of the partial point cloud, and uses multiple identical decoders to decode. Each decoder uses the self-attention module and the Folding-Net to generate smooth point clouds from the features, and finally integrates the point clouds generated by each decoder to obtain dense point cloud. Compared with other mainstream point cloud completion network methods and ablation experiments, it can be proved that this network model can generate more accurate point cloud and effectively complete the 3D reconstruction task.

In this chapter we provide the implementation of the proposed model which is based on hybrid BERT model using Asian languages for MNMT. Due to its benefits in streamlining the training process, lowering the cost of online maintenance, and boosting low-resource and zero-shot translation, multilingual neural machine translation (NMT) (Ni et al., 2022), which translates many languages using a single model, is of significant practical significance. It is exceedingly laborious to manage them all in a single model or use a new model for each language pair given that there are thousands of languages in the world, some of which are very diverse. As a result, multilingual NMT relies on the ability to choose which languages should be supported by a single model given a set resource budget, such as the number of models. Unfortunately, this issue has not been addressed by prior work (Singh et al., 2020). In this study, we create a framework for grouping languages into various categories and training a single multilingual model for each category. Moreover, we also provides two language clustering techniques: (1) language family-based language clustering utilizing prior information, and (2) language-based language embedding, where each language is represented by an embedding vector and grouped in the embedding space. In specifically, we train a universal neural machine translation model to extract the embedding vectors of every language (S. Yang et al., 2020). Our studies on 23 languages reveal that the first clustering approach is simple and understandable but results in subpar translation accuracy, whereas the second approach adequately captures the relationship between languages and enhances translation accuracy for almost all languages (Araújo et al., 2020). (Yang et al., 2021) proposed incorporation of many-to-one statistical machine translation is something else we have done (SMT). When compared to the outcome of the standard SMT, this novel technique gave results in terms of translation accuracy that were equivalent (Kituku et al., 2016).

A one-stage object detector based on SSD, named FIFENet (Feature Integration and Feature Enhancement Network), is proposed in this paper to settle the deficiency of SSD in small objects detection. Two blocks are designed in FIFENet: a feature integration block and a feature enhancement block. Feature integration block fuses the feature map in shallow layers to improve the performance on small objects. Feature enhancement block adopts the residual network (Res2Net) and attention mechanism to enhance feature integration. Experimental result shows that the mean average precision (mAP) on PASCAL VOC2007 data set is 3.1% higher than vanilla SSD, and the accuracy improvement on birds, bottles, chairs, and plants are 3.6%, 9.5%, 5.4%, and 5.5% separately. Results demonstrate that the FIFENet can achieve high detection accuracy while maintaining real-time performance.

The fault of inverters seriously affects individuals' lives, and researchers have become increasingly interested in its method of fault identification. In view of the similarity of three-phase current data of inverter open circuit fault and the low classification accuracy caused by the slight change of three-phase current data measured for distinct open circuit faults, the FDA-KNN method is proposed. Moreover, the method extracts fault characteristics from the three-phase current value of the inverter prior to label discrimination. In the first place, fisher linear discrimination (FDA) is adopted to separate normal data from fault data to extract fault data features, and the subsequently K-nearest neighbor algorithm (KNN) is employed to label faults and identify the open circuit fault type of inverter. The results demonstrate that this method is straightforward to implement and comprehend. It solves the issue of fault classification for similar inverter three-phase current data with greater precision than conventional methods.

Threat Intelligence is the knowledge set and operational advice of a series of evidences including vulnerabilities, threats, characteristics and behaviors obtained through big data, distributed system or other specific collection methods. It can restore the network attacks that have happened and predict the future possible attacks, and provide reference for users to make decisions. Help users avoid or minimize losses caused by network attacks. However, the existing technologies cannot respond in time and defend in advance to threat behaviors in the network environment as a whole, and cannot simultaneously take into account the prediction efficiency and accuracy of threat prediction. Aiming at the deficiency of existing technologies, this paper builds a security defense model of industrial control system based on threat intelligence. Firstly, credible threat intelligence is extracted through the quality assessment model of deep neural network algorithm. Secondly, high-quality threat intelligence is extracted through the self-defined matching principle, and contextual data is extracted to analyze the attack intention and predict the attack behavior. Finally, by constructing an attack and defense game model based on the attacker and the defender, the mixed strategy Nash equilibrium is used to predict the attack behavior based on non-high quality threat intelligence. Through a series of experiments, the model has a good predictive effect in the industrial control system.

Over the years, microservices have become increasingly popular and are being adopted more and more, microservice architecture is gradually replacing monolithic applications as the mainstream architecture. Anomaly detection for microservices is a hot topic of current research, which is important to ensure the security and performance of the application. This paper proposes a deep learning based anomaly detection method for microservices, which first serializes trace data, then trains a model combining AE and LSTM to detect anomalies, and also detects performance anomalies by executing time series. We evaluated our approach on a widely adopted microservice system and showed that our approach outperformed current trace-based approaches.

For higher accuracy of color image registration, this study proposes an SVM-based fast retinal lesion vascular classification method. Firstly, the retinal image is pre-processed by denoising and contrast enhancement, and then the LBP feature vector of vessels is extracted to reduce the dimensionality. Finally, we input the feature vectors into the SVM classifier, choosing the RBF kernel function and defining suitable penalty factors for training. The algorithm performs well in terms of classification with limited small sample labeled data, as demonstrated by experiments, and can recognize retinal vascular occlusion with a recognition rate of 85.29%. In addition, we can make different pre-processing processes for different vessel conditions, which can effectively help improve the fusion accuracy of retinal color images and have practical clinical value.

Scientific studies have shown that regular breathing of fresh air can speed up human metabolism and improve human immunity. In order to effectively realize the update of indoor air and ensure good indoor air quality, this paper uses sensor technology and infinite communication technology, through circuit design and program simulation, designs a sensor-motor-window hybrid system that is in line with the actual situation, and uses low-power MCU STM32 as the main control center. Assisted by a variety of sensors to monitor the indoor and outdoor environment in real time and send signals, according to the data of several dimensions, a set of algorithms for analyzing whether the current environmental factors should open/close the window are written, and finally the automatic switch control of the window is realized by controlling the stepper motor. In addition, this design provides a visual applet to realize the intelligence from air quality monitoring to switching Windows. The design of this paper can well adapt to the needs of the market for smart home in the ordinary people's homes, and has a wide range of market practice value.

The effective traceability of fire products is an important guarantee to prevent potential safety hazards. Aiming at the problems of unverifiable authenticity of fire product information and insecure data storage, this paper constructs a fire product traceability model and implements a traceability system. The traceability model uses the Internet of Things technology to ensure the credibility of the data source, and stores fire data in combination with blockchain and IPFS (Inter Planetary File System). Meanwhile, the traceability system is designed based on the Hyperledger Fabric architecture to achieve efficient information sharing and traceability management of fire products. The experimental results show that compared with the traditional traceability system, the traceability system has low design and maintenance cost and improved security.

URLs (Uniform Resource Locators) are widely used on the Internet, but they are often used maliciously to carry out cyberattacks, causing significant losses to many enterprises and individuals. Therefore, it is crucial to spot those URLs to ensure network security. In this study, we propose a method for detecting malicious URLs that uses bidirectional LSTM and deformable convolutional network. First, the character vector corresponding to the URL is obtained by applying the embedding method, the bidirectional LSTM network is then fed the character vector to extract the global information in the URL, and the parallel deformable convolutional network receives the extracted data as input to learn multiple types of local area features. Finally, the fused local features are output to the FC network for URL classification. In this study, we conducted comparison experiments of different methods on different datasets. From the experimental results, on the three sampled datasets, the method's accuracy was 96.96%, 99.85%, and 96.43%, respectively, comparing with other research methods, the accuracy of the method proposed in this study for malicious URL detection was significantly improved.

In recent years, with the rapid development of the Internet and cloud computing technology, enterprises in all sectors of society and in all fields have entered the wave of digital transformation. The traditional manual programming-based application development model gradually fails to meet the growing business needs of enterprises. In this context, low-code development technology has been developing rapidly, and major Internet companies at home and abroad have invested in the research of low-code technology, and there are currently many mature low-code development platforms that play an important role in fulfilling the business needs of enterprises. However, the current mainstream low-code platform often has low scalability and covers limited business scenarios. This paper takes this as the main entry point, proposes a new solution and designs an intelligent low-code development platform technology. The overall architecture process of the platform, the platform visualisation building panel, and the automatic code generation are explained in more detail, and an intelligent low-code development platform with good scalability is implemented on this basis.

Elastic scaling is one of the important features of cloud native. In the actual production environment, because the business requirements and application resource load are in a dynamic change, the deployment of business and allocation of resources can be adjusted in real time through elastic scaling to ensure the overall quality of service of the system. As the current mainstream container orchestration technology, Kubernetes' built-in elastic scaling strategy HPA (Horizontal Pod Autoscaler) obtains the corresponding indicators by monitoring the component metrics and calculates the expected value of the replica by comparing it with the user-defined threshold, thus realizing the elastic scaling function. Although this scaling strategy can solve the problem of dynamic scaling, there are problems of response delay and scaling jitter, which makes the quality of service of the system unable to be guaranteed in a specific scaling period. In view of the above problems, the predictive elastic scaling strategy is improved, and the fuzzy AR (p) time series model is built to predict the specified load indicators, so as to achieve predictive elastic scaling. The experimental results show that this strategy can solve the response delay problem well and reduce the unnecessary jitter scaling.

Multi-object tracking (MOT) is to track multiple objects simultaneously in a video. The main application scenarios of MOT are security monitoring and automatic driving, etc. In these scenarios, we often need to track and identify many targets at the same time. In this work, we have presented a multi-camera person tracking and re-identification system based on edge computing which comprises of communication solution, a FairMOT based tracking system and a MoT based re-identification system. We apply our system in real scene online, the results show that our system is efficient for multi-camera person tracking and re-identification tasks and can run with frame rate at real-time.

Human action recognition is the process of automatically recognizing human activities in digital video sequences. It is an important research topic in computer vision, particularly in the field of video surveillance. Conventional approaches for skeleton-based action recognition mainly focus on modeling the temporal evolution of the skeleton data, while not taking the contextual information of the scene into consideration. In this paper, we propose a novel human action recognition framework with context awareness: YOLOv5 is used for context recognition first, then the obtained information is integrated with ST-GCN for skeleton-based behavior recognition. Experimental results show the superiority of our algorithm over ST-GCN alone in industrial scenes.

Accurate inbound and outbound passenger flow prediction is an essential task of urban rail transit intelligence, which guarantees the operation scheduling, and planning of stations and stations. Considering multiple complex relationships and time-varying characteristics among stations, the urban rail transit passenger flow prediction model based on multigraph convolution and Transformer model is proposed. Three graph relations, namely the connectivity graph, association graph, and interaction graph, are proposed from the passenger flow characteristics and topology between rail transit stations. The multi-graph convolution constructed based on the three graph relations and Gated Recurrent Units (GRU) are combined to form a multi-graph convolution unit, effectively simultaneously capturing local spatio-temporal dependencies. At the same time, the output state of each multigraph convolutional unit is stitched into the Transformer model to mine the global features of the long-range time dimension. The model's validity is verified after finishing with the AFC data of Hangzhou Metro 2019 total stations, and the results are obtained after repeated experiments. The proposed passenger flow prediction model has better prediction ability as the values of three error evaluation indexes are smaller than other traditional benchmark models.

The LEO satellite constellation communication system integrating with the existing ground 5G system is an extremely complex system engineering. Therefore, it is necessary to build a simulation and evaluation system to provide technology supports for multiple stages of the engineering construction, including design demonstration stage, operation management stage and on-orbit application stage. One of the most important tasks of building a simulation system is modeling the key components of the constellation system. Based on the unique properties of LEO satellite constellation and some important factors affecting communication performance, the satellite, gateway, core network, system control center, terminal and radio channel were modeling with a proper simulation granularity. We focus on the model architecture and functional compositions supporting payload service simulation, and detail the modeling process of the inter-satellite laser link affecting the entire satellite network performance, as well as giving the modeling methods of laser capture and spot alignment. The influences of satellite orbit and attitude on the optical inter-satellite link were put into considerations, which is consistent with the actual working status when satellites are on orbit. The modeling methods are efficient for improving the credibility of the simulation and evaluation results of the LEO satellite constellation system.

Due to the rapid development of digital power grids, 3D visualization has gradually become the key to digital 3D design and 3D review of power grid engineering. Therefore, this paper has carried out research on the design of the 3D visualization system of digital power grid engineering. Firstly, the proposed hierarchical model of 3D visualization system for digital power grid engineering is described; then its key technology and development environment are introduced; finally, the effectiveness of the system in digital power grid engineering application is verified through application examples and effectiveness. Application examples show that the proposed system can effectively help power grid engineering to realize 3D visualization and improve the efficiency and intelligence of power grid engineering construction.

With the development of computer technology, virtual reality has gradually penetrated and changed every industrial field. The final form of Metaverse must be decentralized, and the current network ecology cannot fully meet the needs of Metaverse's decentralization. Some people believe that the coming Web 3.0 era is inseparable from the Web ecology required by the meta circle. Web 3.0 could be an important step towards a metaverse world. Different from traditional online interaction methods, pictures and videos are mainly used, which does not affect the actual production. There is no clear understanding of the overall layout of the screen, and the interaction efficiency is low. Considering the above shortcomings, this paper adopts the octree scene division method and the LOD (Levels of Detail) model technology to save storage space, speed up the rendering speed, and achieve fast rendering of virtual scenes. Introduce artificial intelligence into collision detection of 3D models, create OBB (Oriented Bounding Box) bounding boxes of model objects, and use particle swarm search algorithm to quickly find pairs of collision objects to improve collision detection efficiency. Experimental results show that a more stable average search length is realized and provides better node balance. At the same time, the multi-level detail technology LOD corresponding to the structure of the eight tree is designed to simplify the fine representation of objects in the scene layer by layer, to reduce the time of scene rendering, reduce the complexity of time and space, and improve the visualization speed.

To solve the safety hazards caused by easy tilting and manual decoupling of the lifting platform, such as large weight, unstable force, and uncertain stiffness and strength in the overall device, according to the development status of the existing lifting spreader, combined with its mechanical structure and the balance of carrying, the finite element analysis of the lifting platform is carried out with the help of ANSYS Workbench software to obtain the deformation influence and force value. After calculating the maximum stress value and maximum pressure point of the platform, the parameters of the lifting platform are determined in combination with the needs of the factory work. As a result of the finite element analysis, the maximum elastic strain of the platform is 1.112×10^-3 mm, the maximum stress is 222.42 MPa, and its safety factor value is within a reasonable range. Analysis and simulation are conducive to optimizing the accuracy of platform structural details, making the design more reasonable, and preventing serious consequences caused by platform bending.