This PDF file contains the front matter associated with SPIE Proceedings Volume 12492, including the Title Page, Copyright information, Table of Contents, and Conference Committee Page.

The development of highly intelligent power grid and the construction of new power system bring the demand of massive data access. In recent years, with the continuous updating and iteration of communication technology, the power system has formed the ubiquitous communication network which is mainly based on optical transmission technology and supplemented by a variety of wired and wireless access technology. For terminal access communication network, using wired communication can improve communication reliability and security, but optical transmission network deployment will bring huge investment. Trusted WLAN is a wireless coverage extension of the power fiber private network. It adopts the WAPI authentication standard and has higher security than the US-led Wi-Fi security technology. It provides secure and reliable wireless access for power terminals and prevents unauthorized access, forgery, and intrusion. On the other hand, the current trusted WLAN construction scheme needs to be based on the power fiber private network for wireless access coverage. Therefore, the adaptation of this technology to power services needs to be thoroughly analyzed in terms of communication performance indicators, security, and economy.

In this paper, we first conducted a detailed analysis of the perturbation phenomenon of the sensing matrix, and then, through statistical learning theory, we established a robust optimization and reconstruction mathematical model of compressed sensing based on the perturbation of the sensing matrix. Finally, we carried out a detailed analysis of the optimization of the model. The model is resistant to disturbance uncertainty. As an extension of the standard mean square error boundary constraint, it introduces the distortion parameters caused by sampling and other factors to determine the relationship between the mean square error of the estimated signal and the sparse measure. The robust compressed sensing optimization algorithm based on disturbance established in this paper belongs to the category of convex optimization. Most importantly, the proposed model is proved to be reasonable and mathematical. Finally, we apply the mathematical model of compressed sensing to image similarity comparison, and the experimental results show that the mathematical model has a good matching effect in the field of image similarity, which meets our requirements.

In the hierarchical storage system, the data value needed to be evaluated to formulate rules data for hierarchical storage and migration. To further research the data hierarchical storage system in railway construction period and how to allocate the resources of the storage media with different performances, the researches on evaluation of the data value in railway construction period were pursued in the paper. Based on the business and non-business characteristics of the railway construction period, the evaluation indicator system of data value in railway construction period were designed, which can provide a comprehensive reference for the model. Combined with the quantification standard of evaluation indicators of data value in railway construction period designed in this paper, the complexity of indicator measurement can be reduced. In order to further improve the accuracy of model evaluation and meet data storage requirements, a comprehensive data value evaluation of railway construction period based on analytic hierarchy process (AHP)-grey clustering method was proposed. A simulation was carried out based on the data actually stored in a railway construction period, and the applicability of the model in the hierarchical storage system was proved by the geometric distribution of the results. In order to further verify the accuracy of the data value evaluation model proposed in this paper, its evaluation results were compared with that of the traditional model based on the same group of data. Experimental results shown that the data value model proposed in this paper was more accurate and adaptable than the traditional model in railway construction period, and can better meet the application requirements of the actual railway hierarchical storage system.

Experiment design and implement to detect the possible pedestrian abnormal-behaviors in cross passages of public buildings are more significant to prevent possible crowd accidents than ever before. The further support of abnormal-behavior experiments can be helpful to stability analysis of moving pedestrian crowds. To summarize the experiments on pedestrian abnormal behavior detection based on computer vision technology, this study focuses both on the abnormal behaviors of moving pedestrians in public traffic areas and the computer vision technologies. A 3D scene analysis workflow using computer vision for crowd behavior experiment is designed. The Workflow model of abnormal behavior recognition and stability analysis in crowd movement used in experiment design is proposed based on Lyapunov criterion theory. Finally, a survey table of typical abnormal behaviors in public scenes is figured out.

The solar greenhouse is an infrastructure widely used in agricultural production, and the precise control of its temperature can stabilize yield and production quality. To improve the air temperature control performance without affecting the actual production, a method for simulation modeling of the greenhouse and its air temperature control was proposed based on the thermodynamic principle. Firstly, by combining the following main influencing factors: the heat gained from solar radiation, the heat transferred via ventilation, conduction, long-wave radiation, and the heat storaged by wall, a mathematical model of the greenhouse was constructed. Secondly, a classical proportion-integration-differentiation (cPID) controller was deployed for the air temperature control of the greenhouse based on ventilation. Then an enhancement for cPID, a single neuron proportion-integration-differentiation (SNPID) controller was proposed to improve the control effect, and the former cPID controller acted as a comparison. Next, a MATLAB/Simulink simulation was carried out based on the solar greenhouse model and the PID controller models. As the experimental results show, with sunlight input and Gaussian noise interference in the controlled greenhouse model, the maximum positive and negative temperature offset of SNPID are both smaller than the cPID results, and after interference at the simulated time t=3.33 h, the adjustment time of SNPID is shortened by at least 45.0% compared with cPID. The temperature curve under SNPID control is closer to the target and more stable than the curve of cPID. Thus, the accuracy and robustness of control in this greenhouse are improved.

Model-free adaptive control (MFAC) is based on data and does not depend on the mathematical model of the controlled object. It is simple in structure and easy to implement. At present, there are few methods to determine the parameters of MFAC controller, which brings considerable inconvenience to the research and application. Therefore, an improved sparrow search algorithm (ISSA) is proposed to optimize the parameters of MFAC. The ISSA improves its performance by introducing the piecewise chaotic map operator, and finally realizes the automatic optimization of MFAC. The results show that the ISSA has better searching speed and precision. After using optimized parameters for control, the overshoot is greatly reduced, and the oscillation phenomenon when the desired output changes or the external disturbance occurs is effectively overcome, which makes the system more robust.

The weighing accuracy is an important index of electronic belt scale. It is difficult to achieve accurate measurement due to the influence of environmental temperature and humidity, belt tension and mechanical vibration. The filtering algorithm is introduced into the belt scale data acquisition and processing system. By filtering the collected data can eliminate or reduce the system error and improve the weighing accuracy. By studying the composition structure, weighing principle and error source of belt scale weighing system to analyze and test three filtering algorithms. By comparing the experimental data, an improved adaptive filter algorithm can play a better role and improve the weighing accuracy. It is important to improve the dynamic weighing accuracy of electronic belt scale and reduce the error.

During the operation of power capacitor, leakage, expansion and other abnormal phenomena may occur due to overvoltage, tide, thermal degradation and harmonic overload. Failure to monitor the performance and state of power capacitors in time may lead to power accidents, which will affect the safe operation of power system. Therefore, the monitoring method of performance of power capacitor was studied in this paper. Firstly, the defects and shortcomings of the existing monitoring methods were pointed out based on the analysis of the characteristics of power capacitors. Then the dielectric loss tangent was selected as the characteristic parameter to measure the performance, and the detection principle was proposed. After that, the distributed on-line monitoring system based on WiFi wireless communication was designed against the operating environment of power capacitors in substation, which included the overall scheme, the local signal detection terminal taking DSP as the main control unit and the centralized human-computer interaction designed in LabVIEW. Finally, the design system was tested. The results show that the measurement error of characteristic parameter is less than 10%, and the accuracy of state discrimination is 100%, so the design system have advanced technique and functionality. In addition, higher monitoring efficiency can be achieved because the performance of multiple power capacitors can be monitored at the same time by the distributed monitoring. Therefore, the system has high practical and popularization value.

In this paper, a frequency sweeping signal generator based on FPGA and DAC was described. The signal generator used FPGA and Verilog to design direct digital synthesizer and then realize linear sweeping signal waveform. Several parameters can be adjusted in the software, such as frequency tuning word, delta frequency tuning word, delta frequency ramp rate word, the signal generator will generate linear sweeping waveform. Detailed design of FPGA is presented. Meanwhile, relevant experiments are given, and the experimental result shows that this signal generator can improve the frequency resolution of the direct digital frequency synthesizer effectively and has very good spectral characteristics.

The traditional "smoke detector + sprinkler" can no longer meet the fire safety requirements in large places such as warehouses, shopping malls and airports. Intelligent fire water monitor has been increasingly used in public places. The S7-1200 PLC (Programmable Logic Controller) that developed by Siemens is selected as the controller, the fuzzy PID (Proportional Integral Derivative) control is adopted as the control strategy. By establishing the fuzzy controller in the PortalV15 software, the PID control parameters are imported to achieve the control of water cannon flow. The simulation results showed that the flow control system has high precision and fast adjustment speed, which can be applied to current firefighting.

Under complex weather conditions, the vehicle vision system has low recognition accuracy for traffic signs, and there are problems such as missed detection and false detection. An improved YOLOX-S traffic sign detection model is proposed. Firstly, a detection layer is added to the YOLOX-S network, so that the model can effectively detect the minimum target and improve the prediction ability of the model. Then the attention mechanism module is added to the YOLOX-S feature fusion network to strengthen the feature extraction function of the network. Finally, the data enhancement mechanism is introduced to the model, so that the detection of the model in severe weather such as haze and rain and snow has strong robustness. The detection model was tested on the Chinese Traffic Sign Detection Benchmark dataset (CCTSDB). The results show that the mAP @ 0.5 of the improved algorithm is 2.67 % higher than that of the original YOLOX-S algorithm, and the mAP @ 0.5: 0.95 is 4.58 % higher than that of the original YOLOX-S algorithm. The model meets the requirements of real-time detection.

In this paper, a short-term load forecasting of bi-directional long short-term memory (BiLSTM) neural network based on grey relational analysis (GRA) and attention model (AM) is proposed. Firstly, the GRA is used to analyze the correlation between the load and weather factors, and the optimal feature set affecting load is extracted and selected as the input of the prediction model. Then, the AM is used to tune the BiLSTM neural network model parameters. Finally, the BiLSTM neural network model is used for load prediction and being verified with sample data. Compared with other prediction models, the model proposed in this paper shows higher prediction accuracy.

In this paper, to address the problems of low accuracy and long timeliness of intrusion detection in multi-sensor decision warning technology in security systems, a Fusion_Time_Bagging (FT_Bagging algorithm for short) is proposed, the main feature of which is to participate in multi-sensor decision warning technology by introducing the concept of fused temporal features. The main process of this method is to first discretise the temporal attributes of the training dataset and then assign initial temporal weights; then use a particle swarm optimisation algorithm to optimise the weights of the initial temporal weights to obtain a training dataset with fused temporal features; finally, n sample sets are obtained by self-sampling, and the final prediction model is trained by eliminating the sample sets with temporal weights below the temporal feature threshold. Experiments were conducted on the decision alert dataset and the UCI public dataset, and the results showed that the algorithm outperformed the Bagging and Random Forest algorithms in terms of decision accuracy and timeliness, and outperformed five algorithms including Bagging, Random Forest and XGBoost in terms of performance such as accuracy rate.

Convolutional neural network-based image segmentation has been successfully applied to lane detection. However, existing lane detection method cannot effectively capture the spatial continuity and long-range dependence of the lane line feature. This gives rise to two issues, i.e., discontinuity and structural divergence. In this paper, we propose a generative adversarial network (GAN) framework based on self-attention mechanisms and asymmetric convolution to achieve accurate lane detection. Specifically, in consideration of spatial continuity, we employ asymmetric convolution to capture the continuity relationship from both vertical and horizontal directions. Moreover, we append two types of self-attention mechanisms to enhance the long-range dependency between different channels and different positions of the feature maps. In the discriminator of the GAN, patch loss is used to discriminate the structural authenticity of predicted output. Experiments on the Tusimple, CULane and Caltech datasets show that the proposed method obtains competitive results compared with existing methods. Without any preprocessing or postprocessing, the mean intersection over union (IoU) reaches 53.17%, 59.9% and 73.37% respectively.

As a standard light source, medium and low temperature blackbody is widely used in the field of quantitative infrared remote sensing, so it is urgent to improve its calibration accuracy. This paper studies the calibration method of medium and low temperature blackbody radiation temperature based on standard transfer detector, and designs a dual channel thermal infrared transfer standard radiometer. The radiometer adopts optical path structure with achromatic off-axis reflection, its optical mechanical system is placed in liquid nitrogen refrigeration vacuum Dewar, and an optical chopper is used to improve the systematic noise ratio. Finally, through the correction experiment of commercial blackbody radiation temperature, we found that there is a certain deviation at different temperature points, and its maximum value is 3.22k. The results prove the effectiveness of the design.

A fire detection model with improved YOLOv4 is proposed for mobile devices with limited computing resources and less accurate object localization. Firstly, the network model of YOLOv4 object detection algorithm is modified, and a depthwise separable convolution network is used instead of traditional convolution in the feature extraction network part to realize the lightweight of fire detection model. Then the Loss function is optimized to solve the problem of inaccurate object detection frame localization. The experimental results show that compared with YOLOv4, the improved algorithm reduces the model parameter by 60.7 % and the detection speed increases by 27.9 %. It is more favorable for the model to be equipped on mobile devices.

The expression of fault diagnosis expert knowledge and its reasoning method is the main content of fault diagnosis expert system design and is also an important factor that affects its performance. The traditional expression method of fault diagnosis expert knowledge is more complicated, and its reasoning machine is serial reasoning using the combined rules. When the reasoning logic becomes more complex and the number of rules grows larger, the search efficiency and reasoning speed get lower. This paper defines a new method of knowledge expression and proposes an expert knowledge expression system based on meta-propositional knowledge network model and knowledge matrix. At the same time, a parallel reasoning algorithm based on matrix operation is designed. It is convenient for the expression and storage of expert knowledge in the computer, as well as the detection of rule conflict and dead loop. Using this method can greatly improve the reasoning speed of expert system. And the number of rules grows larger, the effect gets more significant.

The function of the pre-stabilization circuit in the traveling wave tube power supply is to convert the unstable bus voltage on the satellite into a stable DC output. The pre-stabilizing circuit of the traveling wave tube generally adopts the Buck structure. However, there is a suspension driving problem in the Buck circuit, which will make the driving circuit become complicated. At the same time, the efficiency of Buck is generally 90%, and Buck circuit can only be used to reduce voltage. In order to solve these problems and improve circuit efficiency, this paper proposes a new type of pre-stabilization circuit based on quasi-resonant flyback converter. The circuit can realize buck or boost, and the main switch tube has no floating drive problem. In the case of 28V input and 35V output, the efficiency reaches 91%. Therefore, the quasi-resonant flyback converter can replace the Buck for the pre-stabilization circuit.

The proportional-integral synchronization control problem is investigated for a class of complex dynamical networks with coupling delay via a switching event-triggered mechanism. A switching event-triggered mechanism is adopted to determine when a certain control signal should be transmitted, which reduces the updates of the controllers and excludes the Zeno behavior. A synchronization criterion for complex dynamical networks, under a proportional-integral control strategy, is established. In addition, the proportional-integral controller gain matrices and the event-triggered parameters are jointly designed. The validity of the proposed results could be illustrated by a numerical example.

In public places, the fall behaviors of pedestrians possibly lead to the disturbances of the crowd, and even cause stampedes. This study deals with the problem of identifying anomalous pedestrian behavior in an effort to stop potential stampedes in public areas. In order to detect the fall behavior of pedestrians in public places, Baidu AI was introduced in this paper to detect key skeleton points of pedestrians in a single frame sourced from surveillance videos. The ratio of human height to width and cotangent value of the direction angle of the pedestrian minimum peripheral rectangle are selected as feature vectors. Fall behavior detection model based on SVM is proposed. Experiments are designed and implemented to validate the proposed fall behavior detection model in this paper. This study can provide technical support for early warning and prevention of possible stampede accidents in public places.

Trajectory planning is one of the key technologies in the manipulator motion system, and the effectiveness of the manipulator as a whole is directly influenced by this technology. Thus, accurate and effective trajectory planning is crucial. In this study, a reinforcement learning system is used to plan a manipulator's trajectory. A trajectory planning of a manipulator based on the deep deterministic policy gradient (DDPG) method, which is easier to converge on than the Actor-Critic (AC) approach, is proposed in order to address the issue that the AC algorithm is difficult to converge on. A guiding reward function is provided in the reward function area to address the issue that the manipulator can't contact the target rapidly due to blind exploration. This can help the manipulator touch the target more quickly. In this paper, a two-degree of freedom manipulator is selected as the experimental object. The obstacle avoidance reward function based on the AC algorithm, the obstacle avoidance reward function based on the DDPG algorithm, and the obstacle avoidance reward function and guiding reward function based on the DDPG algorithm are compared. The findings demonstrate that the DDPG method converges easier than the AC algorithm, and that the directed reward function can help the manipulator avoid obstacles.

Orbital angular momentum (OAM), a new modulation degree of freedom, has the potential to significantly improve frequency spectrum. Alignment and static are, however, required for the OAM communication system based on uniform circular array (UCA), which is difficult to achieve. A beam selection scheme based on OAM is proposed and matching beam pair is obtained by beam scanning. To eliminate intermodal interference caused by misalignment, a modal selection algorithm is proposed. The simulation results demonstrate that the beam matching scheme can effectively address the issue of system performance degradation caused by misalignment. By selecting transmission modes, intermodal interference can be reduced.

The paper proposes the approach to eliminating the ambiguity of radial velocity measurements in the FMCW automotive radar on millimeter waves. The method involves obtaining the approximate estimate of the radial velocity at the time of trajectory detection and determining the true Nyquist zone with subsequent correction of the measured velocity. The proposed method does not have difficulties associated with alternating fast chirps and the use of multiple frames with different periods of repetition of chirps, both of which require computationally complex algorithms to solve the assignment problem. Additionally, with a low SNR, high reliability of the algorithm is ensured.

Human activity recognition is widely used in security monitoring, medical monitoring, human-computer interaction, and other fields. Radar-based activity identification plays an irreplaceable role in security and equipment independence, receiving extensive attention. This paper mainly studies human activity recognition based on Frequency Modulated Continuous Wave (FMCW). Learning the invariant stability characteristics under different scenes can improve the generalization ability of activity recognition. Inspired by the idea of image style transfer learning, the labeled training data can be transformed into styles of other scenes, to enhance the training data, improve data diversity and network generalization ability.

Traditional person re-identification methods are mostly based on supervised state of manually annotated data sets. Although the recognition accuracy is high, the algorithm completely relies on effective annotation information and lacks practicability. Based on the above problems, an unsupervised person re-identification method is proposed, which refines features to guide multi-label distribution sorting learning. Firstly, a single and multi-label classification network structure is constructed to improve the matching accuracy of tags by learning the characteristics of each person and its k nearest neighbor and sharing the same class. Secondly, the refinement feature extraction module is designed and embedded into the original ResNet50 network, and the potential key areas in the person image are adaptively located. While weakening the appearance differences of the features, multi-information mining is carried out to obtain the refined features of the characters and assist the classification network to improve the performance. Thirdly, reliable multi-category labels are screened by punishing the sorting between positive and negative categories of labels in multi-category labels. Finally, the network is supervised by combining multi-label distribution sorting loss and multi-label classification loss. Experiments show that without any annotation information, the recognition accuracy of Market-1501 and DukeMTMC-reID datasets is 58.4% and 42.3%, respectively, and the mAP of Market-1501 dataset is increased by 12.9%.

Aiming at the problem of coverage search for Unmanned Underwater Vehicle (UUV) in unknown underwater environments, this paper proposes South-First (SF) coverage algorithm based on the scanning line. A template-based patrol marking method is proposed to solve the problem of marking obstacles. For the deadlock situation, a backtracking mechanism is established, and the improved A* algorithm is used to plan the escape path from the dead point to the backtracking point. Simulation experiments show that SF algorithm can complete the coverage search of the unknown area, and the generated path have less steering times, which is suitable for UUV coverage search.

To reduce costs and improve payload capacity, modern solid launch vehicles usually use the depleted shutdown solid motors. Because of the incapacity to control thrust, fuel burning rate and shutdown time, the energy management guidance method is the key technology for this kind of launch vehicles to complete high precision orbit injection and overcome various dispersions in the meantime. To solve this problem, a guidance method with iteration is proposed for launch vehicles with depleted shutdown solid motors by energy management attitude in yaw channel, the method is able to assess the vehicle's orbit injection capability in real-time before the injection stage ignition and implement direct injection. Numerical simulation result showed its adequately high accuracy of orbit injection and considerable value for the depleted shutdown solid launch vehicles.

During the construction of the customer-centered energy Internet marketing service system (hereinafter referred to as "power marketing 2.0"), a unified work order system with the "1233" function was built by referring to the sap work order model, which solved many complex practical problems. These questions include: the work order channel information is not connected, the work order distribution is not intelligent, the work order cost is not automatically collected, and the work order data value mining is not sufficient, etc. It can not only realized unified data aggregation, unified scheduling management, unified cost collection and unified information release, but also achieved the four improvements of "customer satisfaction, professional ability, personnel efficiency and experience benefit".

In view of the large deviation of the pixel value of the generated image caused by the abandonment of the BN layer by the previous deep face super-resolution module, and the inaccuracy of the face prior alignment module, an adaptive modulation super-resolution neural network combining the attention mechanism and face alignment is proposed. Firstly, in order to solve the problem of inaccurate face alignment, a specific attention module is used to extract features with low resolution, and the output feature map is aligned with the key point feature map to increase the accuracy of positioning landmarks. Secondly, aiming at the problem that local pixels have maximum values, an adaptive modulation super-division module is proposed to make the reconstructed image more suitable for visual senses. The experimental results show that compared with face super-resolution algorithms such as end-to-end learning facial prior network (FSRNET), facial landmark attention network (PFSR) and deep iterative collaboration network (DIC), better visual effects and performance indicators are achieved.

The operational status fault detection of the flight control system plays a crucial role in the safe and smooth flight of the overall system. In this paper, a data-driven approach is used to detect several typical faults of flight control system sensors, and a least-squares support vector machine flight control system sensor fault detection method based on the sparrow algorithm is proposed. To address the problem of blind parameter selection in the least squares support vector machine model (LSSVM), the sparrow algorithm is used for parameter search. This paper proposes a sensor fault detection model for flight control system based on KPCA-SSA-LSSVM, and the sparrow algorithm is used to perform parameter search optimization on LSSVM to enhance the fault diagnosis performance of the algorithm. Experiments prove that the diagnosis accuracy of the algorithm reaches 98.23%, much higher than other fault diagnosis algorithms, which proves the rationality of the proposed algorithm in this paper.

At present, the use of robots to carry out transformer internal inspection work has been related research, if the intelligent detection of small targets such as foreign bodies and small discharge traces inside the transformer can be realized, the efficiency of robot internal inspection will be greatly improved. For small target detection, the current popular method in the industry is to improve the detection accuracy by optimizing the structure of the network model, but the disadvantage is that it increases the difficulty of the algorithm design and the computational complexity. In this paper, based on the Faster-RCNN model, small target enhancement and contrast learning methods are proposed for small target detection in the industrial field under the premise of ensuring the detection accuracy of large-scale targets. The experimental results on the transformer internal inspection data set show that our proposed method is superior to the existing methods. It provides a new solution to the problem of improving the recognition effect of small targets.

Under the background of the development of industrial Internet and digital economy era, artificial intelligence is widely used in all businesses of power grid, among which the identity authentication of personnel in substation operation and maintenance area is a common requirement in the business scenario of power grid. This paper proposes a " cloud to edge integration" solution based on offline feature extraction + cloud recognition and authentication, which extracts face features from the end side, compares face information in the cloud, and realizes identity authentication through minimized data transmission. Compared with the traditional non biometric identification means such as key and ID card, fingerprint and iris identification and other biometric identification means, this scheme has the advantages of insensitive identification operation, fast speed, high accuracy and good economic benefits, and has high promotion value.

The main challenges are the intra-class differences of person images and the cross-modal differences between visible and infrared images for cross-modal person re-identification. How to reduce the cross-modal differences becomes the key to cross-modal person re-identification. In this paper, we propose a hybrid learning strategy using Cross-Entropy loss and weighted squared triplet loss as identity (ID) loss to solve the intra-modal and inter-modal person identity classification problem, while supervising the network to extract more effective modal shared features to form specific feature descriptors. Besides, for cross-modal person image attributes, a data augmentation method of channel-swapped random erasure is used to improve the robustness of the model to color changes, simulating different degrees of image occlusion, reducing the risk of overfitting and further enriching the image diversity. Experimental results on the public dataset SYSU-MM01 demonstrate the effectiveness of the proposed method, with an average accuracy mAP of 60.08% even in the most difficult full-search single-shot mode.

Microgrid is an effective way to improve the utilization rate of renewable energy and is an indispensable part of recent power networks. In microgrids, the deployment of energy management systems (EMS) ensures stable operation and maximizes energy efficiency. Due to the uncertainty of non-steerable generation and non-flexible consumption in the microgrid, it is challenging to design an energy management algorithm to schedule the steerable generators and storage. To address this problem, energy management system is modeled as a Markov Decision Process (MDP) with continuous action space in this paper. Then an off-line reinforcement learning algorithm is leveraged to help EMS make scheduling decisions. Compared with other EMS schemes based on deep reinforcement learning, our method can effectively utilize the optimal decision data generated by mathematical programming, i.e., expert knowledge, to improve learning efficiency and decision-making ability. Simulation based on real world data verifies that the proposed algorithm has better performance than other reinforcement learning algorithms.

In public traffic scenarios, the old pedestrians, as the vulnerable and care-needing people, often have serious and irreversible effects on their bodies once accidents such as falls occur in their lives. To recognize the dangerous falling behavior of old pedestrians in public transportation buildings, this paper analyzes the gait frequency characteristics of the old pedestrians, tracks specific old pedestrians, adopts the kinematic stability theory, and obtains the kinematic characteristics of video image information based on computer vision technology. Finally, this study builds a kinematics model for the recognition of the fall behavior of the old pedestrians, which can provide an invisible safety guarantee for the safe travel of the old pedestrians.

There are a large number of air switches in substations, computer rooms and other places. It is often necessary to artificially change the opening and closing states of the air switches for specific tasks. Due to the high labor cost and potential safety hazards, it has great potential to implement edge-computing-based deep learning models in these scenarios. However, these scenarios have complex conditions and many electromagnetic interferences, and the long-term stable operation of deep learning models requires high virtual isolation technology involved in edge computing. In this context, this paper uses virtual isolation technology and deep learning model for air switch detection, uses the trained air switch detection model based on YOLOX deep learning framework to make mirror images, and completes isolation enhancement in the generation of mirror images. Besides, on this basis, a safe and efficient control decision-making scheme based on neural network is proposed.

Taking data collection and recording in special environments as the research object, for the characteristics of confined and narrow spaces, high anti-interference requirements, accurate and reliable data recording, a multi-functional data recording system which is applicable to the underwater unmanned vehicle (UUV) is designed. In this paper, the design of the data recording module, sensor module 1, sensor module 2, and video acquisition module based on CAN bus communication is introduced in detail. In addition, two types of underwater unmanned vehicles with multi-functional data recording system were used to take sailing verification, and the results show that the system has achieved the expected purpose.

Switching Diodes are used in high-frequency circuits, and their capacitance effect cannot be ignored. In this paper, based on the research of the barrier capacitance model of switching diode, the test method and test circuit of barrier capacitance was studied, and the influence of different voltages and working temperatures on barrier capacitance were analyzed, and some suggestions are put forward for device suppliers and users.

The batch of observation data was released to the public. The data was processed using the standard basic astronomical program library of the International Astronomical Association (IAU) and the JPL ephemeris, and the time series of 35 photons arriving at the solar system barycenter was obtained. Based on epoch and known ephemeris parameters, 35 pulse profiles are obtained. The pulse profiles are denoised by kernel regression, and the influence of Bin number on timing residuals when epoch folding is analyzed. The results show that the observation data is valid, and a better pulse profile can be obtained, so that the observation of Crab stars is realized accurately. The kernel regression algorithm effectively improves the influence of the Bin number on the timing residual. The root mean square of the timing residual before fitting is about 40us.

To improve the temperature performance of accelerometer, this paper studied the temperature characteristics of MEMS accelerometer, investigated calibration methods of accelerometer and the error compensation model. We proposed a novel interpolation-based least square fitting compensation method, which firstly uses the interpolation method to predict temperature-zero bias and temperature-scale factor data, then compensating doubled data by the least square method. We carried out a compensation experiment and verified the effectiveness of each step in the proposed method. Testing results show that in the range of - 40°C to 60°C, through the use of the least square fitting method based on interpolation, the extreme difference of the scale factor and zero bias errors are reduced by 83.6%, 93.5% respectively. Meanwhile, the root means square error of the scale factor and zero bias errors are reduced 86.2% and 95.1% respectively. After temperature compensation, the fluctuation range of the upright placed accelerometer is reduced from 0.038g to 0.003g. It is proved that this method can reduce the measurement cost and obtain high precision compensation results, which is suitable for engineering applications.

This paper covers the implementation of predictive maintenance (PdM) predicting the wear of technical components of an industrial thermal system. Object of investigation of the technical components here is the motor of a fan. Basic requirement for the PdM system is the availability of extensive data from operating parameters of the motor, which must be recorded in advance through sensor technology. The sensor used here is a MPU 92.65 motion tracking multi sensor with three-axis acceleration, three-axis yaw rate, three-axis magnetic field and temperature measurement. Goal of this work is to develop an unsupervised Deep Learning (DL) model for anomaly detection of multivariate time series. By using the MPU 92.65 the operating parameters of the fan motor are recorded at specific time intervals ensuring sufficient data quality for the development of the DL model. In selecting a suitable DL algorithm for the anomaly detector a specific Convolutional Neural Network (CNN) called Convolutional Autoencoder (ConvAE) with 2-D convolutions has been selected. CNNs with 2-D convolutions are the typical type of neural networks normally used for images (object detection, segmentation, medical imaging etc.). However, the recorded sensor time series here are in 1-D and must therefore be transformed appropriately. The investigated approach of image transformation is the Gramian Angular Field (GAF), which encodes temporal correlation structures spatially as images. A total of four datasets is used with one representing the baseline for the deployment of the detector. The remaining datasets have been set for testing purposes with their results presented and discussed here.

In the high-density crowd flow places in public buildings, typical mobile obstacles, such as trolley cases, mobile sweeping trolleys, shuttle trolleys, police patrol cars, etc., carried by passengers bring convenience for passengers to travel, and can also act as typical obstacles that hinder the flow of people. It is easy to block the flow of people, cause the crowd to become unstable, and cause overcrowding and even stampede accidents. To study the influence of moving obstacles on crowd stability, this paper analyzes the spatial and moving characteristics of typical moving obstacles and constructs a motion model of moving obstacles. Furthermore, based on smooth particle hydrodynamics (SPH), a coupled macroscopic pedestrian flow model including moving obstacles and pedestrian flow is proposed. In order to verify the effectiveness of this proposed coupled motion model, this study takes trolley luggage as an example to design and implement a moving obstacle experiment in pedestrian flow, exploring the impact of moving obstacles to the pedestrian flow, further to study the stability of pedestrian flow.

The conventional decoherence methods based on the uniform circular array are generally affected by the array model errors. To solve this problem, we proposed a new algorithm based on array rotation and translation method. By rotating and translating the dual antenna array to form multiple virtual UCAs, the proposed algorithm can achieve multi-targets direction estimation of coherent signals. The simulation results show that the new algorithm has better robustness on multi-channel gain-phase errors, and it can realize the direction of arrival estimation of coherent and independent signals at the same time.

In order to improve the synchronization, accuracy and assembly efficiency of the spatial multi-joint array spreading and retracting mechanism, this study takes the spatial multi-joint array spreading and collecting linkage mechanism of a radar as the research object, and carries out detailed design of the assembly method and sequence of the spatial multi-joint array spreading and collecting linkage mechanism, and analyzes and measures the accuracy, deformation and synchronization of the spreading and collecting linkage mechanism after the assembly is completed. The results show that the designed assembly method has higher assembly efficiency, good synchronization, repeatability and reliability, and the root-mean-square value of array accuracy is as high as 1.43mm, which meets the design requirements. This study provides a strong guarantee for the subsequent design and batch production of radar arrays using similar spatial multi-joint spreading and collecting linkage mechanism.