This PDF file contains the front matter associated with SPIE Proceedings Volume 12176, including the Title Page, Copyright information, Table of Contents, and Conference Committee listings.

According to the different physiological indexes of medical patients, aiming at the data of blood oxygen saturation level in continuous monitoring period, combined with the booming trend of artificial intelligence and its potential in predicting blood oxygen saturation, this paper quantitatively analyzes the prediction results of different physiological indexes on blood oxygen saturation. First of all, in order not to lose useful information, all the characteristics of patients are taken as the characteristics of the input model. Secondly, in order to find out the internal relationship between different patients' oxygen saturation, we add the previous patients' oxygen saturation data to the input features. Finally, multiple regression model, RNN model, GRU model and LSTM model were established to predict the mean value of blood oxygen saturation. The results show that the deep learning model has faster convergence speed in the training set and better prediction effect in the test set. Compared with other models, LSTM model has the best prediction effect, RMSE = 0.021, MAPE = 0.017%, R2 = 0.99.

Aiming at the problem of communication mode selection in battlefield environment, a military communication mode optimization algorithm based on fuzzy multi-objective decision theory is proposed. This method firstly based on the principles of index system design and military communication to build their own specificity effectiveness evaluation index system, secondly combining the fuzzy theory in this paper, a multiple communication mode decision algorithm, and communication were determined by cross iteration weighted index and the target weight of meteorological index, and combining the optimal target value calculation of each communication mode. Through simulation comparison and analysis with the existing methods, the effectiveness and practicability of this algorithm in the optimization of military communication mode under meteorological environment are proved.

Aiming at the problems of low optimization accuracy, slow convergence speed and easy to fall into local optimization of sparrow algorithm, an improved sparrow search algorithm combining adaptive mutation and proportion adjustment strategy(AISSA) is proposed. Firstly, the nonlinear decreasing inertia weight is used to adjust the global and local search ability of SSA. Secondly, the producer-scrounger proportion adjustment strategy is introduced to adjust the proportion of producers and scrounger in the iterative process, so as to improve the optimization accuracy and convergence speed of the algorithm. Finally, Tent chaotic mutation and adaptive t-distribution mutation are carried out for individuals with different fitness values to improve the local optimization ability of SSA and speed up the convergence. The final results show that AISSA has higher optimization accuracy and better stability than SSA.

After the outbreak of COVID in Wuhan, it has had an impact on all aspects of tourism industry. Tourists' sentiment is an important factor for people to make tourism decisions. The implementation of tourism decisions affects the development of tourism to a certain extent. In order to explore the impact of the COVID-19 on the tourism industry from the micro level of tourist sentiment. Firstly, the text mining algorithm is used to analyze the emotion of tourism microblog text, and the tourism emotion index TSI is constructed. Then combined with the tourism heat index THI, the tourist sentiment TS comprehensive index is constructed. The temporal and spatial differences of the impact of the epidemic on tourists' emotion are analyzed by comparing the tourists' emotion and epidemic data in different regions and stages. From the temporal and spatial distribution of tourist sentiment and epidemic situation, they are not completely parallel related, and there is spatial heterogeneity. Tourist sentiment is affected by multiple factors such as economic level and geographical location. The change of tourists' mood does not only depend on the change of epidemic data, but also related to many factors such as economic level and geographical location.

In order to overcome the trend influence of novel Coronavirus epidemic in the future, this paper proposes the panel data modeling method based on big data crawler technology, which is based on Python crawler technology to obtain a more effective estimation model from the dynamic perspective of time and cross section. The results showed that the fixed effect error rate established by the development of COVID-19 in China, Japan, South Korea, Germany and Italy was about 3%, and there is a positive correlation between cured cases and confirmed cases of COVID-19. The predicted confirmed cases of COVID-19 in week 63 will be 69, 11,908, 3156, 112293 and 147,545, respectively.

In recent years, neural networks have demonstrated attractive performance in a variety of computer vision tasks. This has led to the deployment of CNNs into real-world applications. However, most existing CNNs require a large amount of storage space and computational resources, which hinders their deployment on mobile devices. In order to solve this problem, the existing research shows that quantization and pruning can greatly reduce the memory occupied by the neural network in the device without reducing the accuracy of image recognition, which plays a great role in solving the deployment problem of CNNs. Therefore, this paper classifies and analyzes the quantitative algorithm model of neural network according to the parameters that need to be quantified, summarizes the common defects of the current quantitative algorithm in terms of accuracy, and looks forward to its future development trend.

Multi depot school bus routing problem (MDSBRP) refers to the school bus routing problem (SBRP) with multiple depots, which effectively reduces the service cost by adjusting the order of school buses from depot to station. In this paper, a heuristic algorithm based on simulated annealing is proposed to solve the multi depot school bus routing problem. The problem considers the constraints of school time window and school bus capacity. Through the operation adjustment of depots and stations, the number and driving distance of school buses can be effectively reduced. The existing benchmark case set and modified test cases are used to test the algorithm. The experimental results show that the use of multi depot mode can effectively reduce the number of school buses and save operating costs, which shows the effectiveness of the algorithm.

The identification of disease and diagnosis of pain are the forefront of the application of machine learning research in medicine. Machine learning is widely used in medical and other aspects, and it is also a hot issue in diabetes prevention. To provide a rapid and accurate reference for diabetes prevention and diagnosis, the research uses machine learning modeling and classification, and it analyzes the Pima Indian diabetes data set to study the model of diabetes and classification problems. The study was completed to identify diabetes based on the correct classification. Through various algorithm models, this research adjusts various parameters and selects important features to optimize the classification results of the KNN and the decision tree. Through this kind of research, certain theories and thoughts can be provided for diabetes and other diseases in the future, and then diseases can be discovered and treated early, which has very practical significance to help diagnose diseases. Through this study, the accuracy rate of the KNN model and the random forest model is 76% and 80% respectively, which is helpful for the establishment of prediction model.

Lung cancer has one of the highest incidence rate and mortality rates in the world. Therefore, the recognition and diagnosis of pulmonary nodules based on CT images is an essential index for the diagnosis of lung cancer. Therefore, the detection and segmentation of pulmonary nodules are of great significance in assisting doctors in detecting lung diseases. In this paper, the luna16 data set in the LIDC-IDRI (the lung image database consortium) initiated by the National Cancer Institute is used for research. Given a large amount of data and the diversity of types and sizes of pulmonary nodules in this data set, the detection, and segmentation of various pulmonary nodules are realized based on the improved deep neural network 3D VNet. The deeper the network, the stronger the expression ability and the better the performance. However, with the increase of network depth, there will be problems such as gradient dissipation and gradient explosion. The improved 3D VNet network improves this problem. More than 200000 lung CT images in the LUNA16 data set are preprocessed, such as image denoising and interpolation sampling, and then coarse segmentation images and mask images are generated. They are trained many times in the improved 3D VNet model, and the results with high detection accuracy are obtained. The experimental results show that the dice similarity coefficient (DSC) and IOU of the improved 3D Vnet network can reach 88.29% and 87.25%, which has a good effect on the current detection methods of pulmonary nodules.

There have been two major obstacles in the process of recommender systems. One is the cold start problem, and the other is the data sparse problem. How to solve these two obstacles is the main direction that needs to be improved and resolved in the field of recommendation systems. The paper designs a network model based on UIEncoder-DAMLPARL and proposes a deep learning recommendation method that combines user reviews and ratings with bilateral personalized attention mechanism. It is proposed that learn the characteristics of user reviews on the basis of convolutional neural networks, so as to enhance the interpretability in recommendations. Then the scoring features and comment features are integrated into a unified neural network model through the DAML model, and learn the representation of the user or item from the comments through the user/item encoder, so as to enhance the personalized recommendation. Finally, the high-order nonlinear interaction between the features is realized through the neural factorization machine, so as to achieve the final score prediction. In addition, to solve the problem of data sparseness, the PARL model, combined with our model, assists in modeling when there are few user reviews and product reviews. It is verified in the experiments that our method can effectively improve the performance of the recommender system on Amazon data sets.

This paper proposes an improved Faster R-DRN (Dense Residual Network, DRN) algorithm, which is based on Faster R-CNN, using densely connected residual network DRNet to replace VGG network. This algorithm is suitable for special scenes of building recognition. It has a residual network and a deep convolution residual network structure, which can efficiently perform image detection, classification and recognition. This design optimizes the problem of algorithm overfitting due to the increase of network depth. In this paper, a comprehensive sample data set for various landmark buildings is established, and samples with different weather, different lighting, and different angles are taken to effectively improve the resistance of the training model. Combined with the optimization of the network structure and the training of targeted data sets, the final feature block diagram generated by DRNet not only does not lose the lowlevel edge texture information, but also reuses the low-level feature block diagrams in the deep convolutional network to make the fused feature block Richer feature information effectively improves the model's recognition rate for photos taken in complex environments. The experimental results show that the accuracy of this method for predicting landmark buildings can reach 82.0% of mAP, and the recognition performance of images taken in complex environments is excellent.

At present, many scholars have conducted extensive research on parameter design in particle swarm optimization algorithms and have achieved many results. The particle swarm algorithm is easy to fall into the local optimum, so the paper modifies the particle swarm parameters to improve the algorithm's performance. To further improve the algorithm, this paper proposes a combination of flower pollination and particle swarm algorithm. The comparison of the optimization accuracy and the convergence speed on the four test functions proves the superiority of the flower pollination and particle swarm algorithm.

At present, China's energy storage potential is huge, and the energy storage capacity needs to be gradually improved. Zhejiang Province has issued a number of energy storage subsidy policies, which is very important for the evaluation of energy storage cost before and after subsidy. Based on the full life cycle kwh cost algorithm, this paper calculates the kWh cost before and after the subsidy of three energy storage modes in Zhejiang Province. The results show that the subsidy in Zhejiang Province is effective, which reduces the kWh cost of energy storage. Calculate the payback period and return on investment without considering the subsidy of charging electricity price, so as to provide reference for the government and enterprises.

The firework algorithm is a new type of swarm intelligence algorithm, which is inspired by the process of fireworks exploding to produce sparks and continuously split and explode. This algorithm has a strong ability to solve complex optimization problems and is widely used in many practical problems. This article focuses on summarizing the current research progress of the firework algorithm and its improved algorithm, as well as prospects for its future research directions.

With the development and popularization of the Internet and the rapid development of e-commerce platforms, resulting in a huge amount of commodity short text. Research on short text classification of commodities is of great significance to commodity circulation, consumer behavior research and advertising push. This paper takes the commodity short text data set as the object, and the Convolutional Neural Network (CNN) algorithm based on deep learning is used to construct the classification model. Experimental results show that CNN algorithm is better than traditional machine learning algorithms (Naive Bayes(NB), K Nearest Neighbor(KNN) and Support Vector Machine(SVM)) in commodity short text classification.

Through the energy big data platform, realize the intelligent management of energy, improve energy utilization efficiency, improve energy production system and develop renewable energy, and finally build the energy Internet, promote the application of renewable energy and complete the intelligent peak shaving of energy. The energy big data platform realizes intelligent energy management, builds a control platform that can manage and ensure the efficient operation of the center. The author compares the actual energy data from 2018 to 2021 with the decision tree algorithm and achieves good results.

Since the equipment in the factory is hard to move and the layout is difficult to change, rational planning of the factory becomes the goal of the enterprise, so a factory layout planning method based on Systematic Layout Planning (SLP) and genetic algorithm is proposed. Combined with SLP qualitative and quantitative analysis methods, this paper considers the relationship between all work units in the factory, establishes plant planning constraints, and considers multiple objectives for processing costs, non-logistic relations and regional utilization. SLP scheme is used as the initial solution, and a genetic algorithm is used to solve a better planning scheme for the factory. Taking an engine manufacturing plant as an example, the method is applied to design a layout scheme for the factory. Compared with the method obtained by SLP, the result shows that the planning method is more feasible and can provide a new method for factory planning and design.

With the gradual elimination of the environmental toxicity of halon fire extinguishing agents, water mist fire extinguishing technology has been widely used for its environmental protection and high efficiency, and plays an increasingly important role in fire extinguishing. In this study, computational fluid dynamics model and fire dynamics simulator (FDS) were used to analyze the influence of droplet size and spray flow rate, two main factors affecting the filling volume of water mist space, on the fire extinguishing effect. The results show that when the limit water mist space filling volume is 6.57 × 10, it can effectively extinguish the 200Kw room fire. As the droplet size increases, the fire extinguishing time becomes longer, and the smaller the droplet size, the stronger the ability to isolate oxygen; as the flow rate of the nozzle increases, the fire extinguishing time becomes shorter. By reducing the droplet diameter or increasing the flow rate, the extinguishing time can be reduced.

With the rapid construction of high-speed railway network and the rise of the urban agglomeration, the passenger volume of large-scale railway passenger transport hubs continuous increases. However, due to the limitation of the through train operation scheme, only part of Origin-Destination(OD) pairs can be served, resulting in a large number of transfer passenger flows in the railway network. Adjusting the train arrival and departure schedule in large passenger transport hubs can safely and effectively organize the transfer passenger flow at the station and significantly improve the passenger satisfaction. Therefore, a 0-1 mixed integer linear high-speed rail hub train connection optimization model is established in this article which make the overall invalid passenger waiting time minimum as the optimization function, considering maintenance time constraints, station operation time constraints, reception-departure track capacity constraints, and transfer time constraints, and using the Benders decomposition algorithm for solving. The results show that this technique can obtain the exact optimal solution of the model, and has significant competitiveness in terms of solution quality and solution efficiency compared with other optimization algorithms of the same problem.

The rapid development of science and technology has brought convenience for photographing and sharing images. However, with the rapid increase of image data, security problems such as leakage and tampering also occur frequently. The application of image encryption technology is imminent, especially the encryption of color images is in urgent need of improvement and development. The traditional encryption technology is mainly aimed at data stream encryption, which has low efficiency, large amount of calculation and certain limitations. Based on transform domain encryption, the image is transformed from spatial domain to frequency domain for encryption, and then transformed to spatial domain, which belongs to a kind of lossy encryption. Encryption based on chaos has large key space, simple implementation and fast encryption speed. Generally, multiple chaotic systems need to be used to enhance the security of encryption. Therefore, a simple and secure spatial encryption algorithm for RGB three channels of color image is designed in this paper. Firstly, a safer chaotic sequence is generated by logistic sine cosine mapping, and then four rounds of scrambling and diffusion pixels are designed by chaotic sequence. Finally, In a series of security analysis experiments, the security and effectiveness of color image encryption algorithm based on logistic sine cosine mapping are verified.

Intrusion detection based on machine learning plays an important role in the security protection of the network environment. In order to improve the accuracy of network intrusion detection, intrusion detection algorithm based on dilated convolution and a two-way gating unit is proposed. Firstly, dilated convolution (DCNN) is used to increase the receptive field of information and extract spatial features. Secondly, the bidirectional gated cyclic unit (bigru) model is used to capture the temporal features of data, and the attention mechanism is introduced to increase the calculation of feature importance. Compared with the classical machine learning classifier, this model has a high detection rate. The multi-classification experiments on the famous UNSW-NB15 data set show that the model has leading performance, and its classification accuracy can reach 97.14%.

Focusing the problem that the traditional fuzzy c-means clustering (FCM), which is a local search algorithm, it uses iterative hill climbing technology, which is sensitive to initial values and easy to fall into local minimums. An FCM method based on improved sparrow algorithm is proposed. Improve some problems that appear in Fuzzy c-means clustering (FCM) at this stage, and then combine them with association rules. First, through Logistic chaotic mapping, the initialization of the population can traverse the entire solution space and obtain the initial optimal position of the sparrow, so that it has a richer population and solves the problem of the randomness of the initialization population. Secondly, the adaptive t-distribution mutation operator is introduced into the sparrow algorithm, and the number of iterations of the algorithm is used as the degree of freedom parameter of the t-distribution to enhance the diversity of the population and greatly avoid the probability of the algorithm falling into local optimum. Then combine with the FCM algorithm to cluster the data. The innovation of this algorithm is that the sparrow algorithm and FCM are combined for the first time, combining the advantages of the two algorithms, the effect of data clustering is better, and the time of clustering is also accelerated. Compared with the proposed algorithm FCM algorithm, SSA-FCM algorithm and GA-FCM algorithm, it has stronger optimization ability and better clustering effect.

Based on the traditional SSD(Single Shot multibox Detection) object detection algorithm, this paper replaces the largescale VGG feed-forward network with the lightweight depthwise separable mixed convolutional neural network Mixmobilenet ,proposed in this paper, which makes the overall object detection model parameters volume decreased. Then add multi-scale feature fusion method FPN (Feature Pyramid Network) and focal loss function FL (Focal Loss). Based on the above, this paper innovatively proposes a lightweight depthwise separable convolutional object detection algorithm Mixmobilenet-FFSSD. Experiments show that the accuracy rate on the PASCAL VOC dataset is 72.5m AP, and the parameter quantity is only 6.6M. The object detection algorithm has good performance in terms of parameter quantity, accuracy and real-time performance, and is suitable for the efficient object detection requirements of embedded smart devices.

For the diagnosis of cancer and tumors, medical pathological image analysis is critical. Whereas, due to the diversity and complexity of pathological data, the segmentation task is faced with challenges such as blurred edges, less training data, difficulty in feature extraction and case segmentation. The advancement of deep learning technologies has resulted in breakthrough achievements in medical image analysis by its powerful feature learning, flexible design, and other characteristics, and it has widely applied. In recent years, many scholars have improved the classic segmentation method. By combining various segmentation methods, segmentation efficiency has effectively improved, and the improved algorithm makes up for the defects of the original segmentation method. In this review, we evaluated and examined the recent research accomplishments in medical picture segmentation using various deep learning approaches, as well as the future research directions for medical image segmentation using deep learning.

In recent years, thanks to the development of GPU (Graphic Processing Unit) computing power, deep neural network (DNN) has drawn extensive attention from academia and industry and has been widely used in various fields of artificial intelligence. However, the mainstream deep learning models have some problems such as large storage space consumption and high computational complexity, which makes it difficult for the deep neural network model to be deployed to mobile devices with limited computing resources or applications with strict delay requirements. Therefore, on the premise of minimizing the loss of model accuracy, the algorithm researches of deep neural network compression and acceleration which can lighten the model has been carried out. In recent years, model compression methods such as pruning, quantification and knowledge distillation have emerged. This paper focuses on neural network pruning algorithm. Firstly, this paper expounds the basic methods and classification of neural network pruning algorithms. Then according to the classification of pruning granularity, a more detailed classification, analysis and comparison of specific models are carried out from the algorithm level. Then based on cifar-10 data set, validity tests of some neural network pruning algorithms and contrastive analyses of the experimental data from different models such as the parameters are carried out. Finally, in combination of the recent results of neural network pruning algorithm, we put forward research dilemmas and problems to be solved. Then the future research trends and methods of neural network pruning algorithm are prospected.

A high speed moving target detection algorithm based on multi-stage velocity compensation for modulated stepped-frequency pulse-Doppler (PD) radar is proposed in the paper. The range migration of high speed target is corrected utilizing scaling processing along stepped-frequency signal bursts. Then range-Doppler coupling of chirp sub-pulse, range-Doppler coupling and range-profile distortion of stepped-frequency signal are compensated by Doppler channel division method. Computer simulations verify the effectiveness of the algorithm in this paper.

There are many practical problems in real life, which are finally attributed to solving large sparse linear equations. In order to solve sparse linear equations in parallel, this paper first analyzes the potential parallel process of solving tridiagonal sparse linear equations by Gaussian elimination method and matrix splitting method, and designs and implements these two parallel algorithms to solve sparse linear equations, Through different process tests and performance analysis, it shows that the Gaussian elimination method has poor time performance, while the matrix splitting method has good efficiency in both space and time.

Crow search algorithm is a kind of swarm intelligence optimization algorithm, which simulates the behavior of crows tracking each other. In order to improve the convergence accuracy of the algorithm and better balance the local and global search strategies, this paper proposes the crow search algorithm based on dynamic progressive strategy (DPCSA). The DPCSA uses a reverse search mechanism to improve the convergence speed, introduces the dynamic progressive strategy into the key parameters: awareness probability and flight length, adds the global optimal position information to the position updating dynamically. These improved strategies make the simulation of optimization process more reasonable. In order to evaluate the effectiveness of DPCSA, a series of test functions are used for experiments and the results show that the improved algorithm has good performance.

Whale optimization algorithm (WOA) has the advantages of simple implementation, few adjustment parameters, strong optimization ability, and fast convergence speed, and has been widely used. Currently, it is mainly used in image segmentation, feature selection, model prediction, path planning, production scheduling and other fields. But there are also shortcomings such as easy to fall into local optimum and low convergence accuracy. Therefore, research on the improved whale algorithm has important theoretical significance and practical value. This article describes the basic principles, algorithm flow and characteristics of the whale optimization algorithm, analyzes the range of algorithm parameters and their impact on algorithm performance, and reviews the algorithms in recent years from three aspects: algorithm hybridization, operator improvement, and parameter optimization. The improvement ideas and development trends of the algorithm point out the next research direction of the algorithm.

In the increasingly complex network environment, various attacks emerge one after another. Being able to accurately detect elephant flow and mouse flow plays an extremely important role in defending against large-scale network attacks. Aiming at some shortcomings of current methods for detecting elephant flow and mouse flow, this paper proposes a detection method based on information entropy and improved random forest. After preprocessing the data, first calculate the data feature score with information entropy, screen out the truly valuable features according to the score, then put them into the improved random forest classifier, and finally get the detection results. In this paper, the grid search algorithm is used to optimize the tree and depth of random forest tree, so that the detection results can be obtained quickly and accurately. Experiments show that due to the significant difference between the data characteristics of elephant flow and mouse flow, this method can effectively identify elephant flow and mouse flow. The accuracy rate is 97.93%, the precision rate is 99.99%, the recall rate is 97.91%, and the F1-score is 98.94%, which is improved compared with other algorithms.

Comparison Network (CN) is a kind of special structure for parallel sorting. There are two important problems in CN: Verifying the correctness of a given CN and finding the minimum cost CN. For the two problems, this paper establishes a verification model similar to Petri Net for CN based on [0,1] principle, and provides two algorithms for them. The deduplication and pruning strategies are introduced in the algorithms which make better execution efficiency compared with results in the existing literature.

The traditional matching algorithm is easy to lose the detail information in the complex background, which is disadvantageous to the 3D reconstruction. To solve these problems, a quasi-dense algorithm based on KAZE feature detection is proposed in this paper, which is helpful to more accurate 3D reconstruction. Firstly, the method uses the KAZE Algorithm to extract the feature points, constructs the feature vector to match, and uses the random sample consistency algorithm to remove the mismatched points. Then the region growth is realized by the polar distance constraint criterion based on the adaptive window diffusion method. The experimental results show that the proposed algorithm can get satisfactory quasi- dense matching results and good disparity map, and can reflect the real scene.

Reasonable cruise catering supply plays a vital role in the reputation and economic benefits of the cruise industry. The key to cruise catering supply lies in the balance between catering procurement and sales, and the difficulty lies in scientifically grasping many productions, procurement, and sales. First, this paper establishes a classic retailer inventory management system, then studies the procurement, processing, distribution, and safety inventory issues related to catering issues, and determines the various costs of cruise ship catering. Finally, this paper uses Simulink simulation software to establish cruise ship-related catering inventory management systems to study the relationship between catering food production methods, purchasing decisions, tourist demand, and cruise ship profitability. Compared with classic retailer inventory management, this catering inventory management system adopts more economical production methods and superior purchasing decisions, considering the particularity of the marine environment and tourist groups. And it obtains a more superior cruise catering supply plan.

Uav remote sensing images have the characteristics of complex background and multiple noise interference. If significance detection is carried out before rural homestead identification, the target rural homestead can be located quickly and accurately, providing location information for the detection of rural homestead, thus improving the detection efficiency of rural homestead. In this paper, we studied four commonly used algorithms of FT, HC, LC and AC with good saliency detection effect. The experimental results show that the detection time of LC algorithm is at least 0.63 ms, and the longest time of FT algorithm is 24.46 ms. The LC algorithm is more suitable for the saliency detection of rural homestead images in this paper.

Text intent classification technology, most of them are based on training semantic word vectors to represent words to encode text. The word vectors considers semantics, grammar, etc. However, the global position information of the sentence is relatively lacking. Although the LSTM and other sequence models have short-term memory, the ability to extract global position information is still slightly insufficient. This article proposes to add global position information to the word vector to make the word vector feature it is stronger, and we have verified the text classification effect of FastText, TextCNN, TextRCNN, LSTM-Attention, GRU-Attention and other models in the medical consultation question and answer data. After adding global location information, the FastText classification effect has increased by 1.92%, and the classification effect of the TextCNN model has improved by 2.19%, the classification effect of the TextRCNN model has increased by 0.72%, the classification effect of the LSTM_Attention model has increased by 0.62%, and the classification ability of the GRU_Attention model has increased by 0.34%. From the experimental results of embedding location information into different models, we can note that the global location Compared with the traditional method, the position encoding embedding word vector method has higher classification accuracy and efficiency.

This paper introduces the current application status and main security issues of IC card in the field of urban public transport internationally and domestically, and proposes to strengthen the application security of IC card by using dual algorithms. Dual algorithms mean the two most popular types of encryption algorithms currently used: algorithms 3DES and SM4 in terms of symmetric keys; RSA algorithm and SM2 algorithm based on ECC in terms of asymmetric keys. The two types of encryption algorithms have their own characteristics: algorithms 3DES and RSA, as internationally agreed cryptographic algorithms, have the advantages of good applicability and compatibility. However, the technological progresses pose a risk of being decoded constantly. If we enhance the decoding difficulty by increasing the encryption length, the computing speed will become lower and resources consumption higher. Algorithms SM2 and SM4 which are the cryptographic algorithms independently researched and developed in China, are superior to the algorithms 3DES and RSA in terms of encryption and decryption speed, resource consumption and security, but they don’t bring enough applicability. In view of the features of the two types of cryptographic algorithms, this Paper proposes to integrate the two types of algorithms into one IC card, and to switch the algorithm by the algorithm indicator or the algorithm switching instruction, so as to achieve the adaptive execution of appropriate algorithm based on practical conditions. Therefore, it can enhance the application security of IC card while ensure the wide application range.

Short-term power load forecasting is an important foundation to safe dispatch and economic operation of power system, and with the marketization of power system, the accuracy of load forecasting plays a direct role in the economy and stability of power system operation; however, the instability of load sequence increases the difficulty of forecasting. Traditional load forecasting methods cannot achieve satisfactory forecasting accuracy, but the rise of synthetic intelligence technology, especially the rapid development of deep learning and big data technology, has laid a good foundation for further improving the accuracy of load forecasting. Among them, Least Squares Support Vector Machine (LSSVM) not only maintains the advantages of Support Vector Machine (SVM), but also lowers the evaluation complexity and accelerates the solution speed, which provides a new research direction for short-term power load forecasting. In this paper, LSSVM is used for short-term power load forecasting. This paper applies LSSVM to shortterm power load forecasting and establishes a short-term power load forecasting model based on LSSVM. Moreover, an Improved Particle Swarm Optimization (IPSO) algorithm, which provides guidance for initial particle selection and avoids premature aggregation of particles based on population diversity information, is used to optimize the penalty factor and kernel parameters of LSSVM, and is validated with historical load data and meteorological data of a region. The experimental results display that the IPSO-LSSVM model used a higher prediction accuracy in this paper.

In recent years, with the development of artificial intelligence chip, brain-like computing chip has become one of the hotspots in the field of artificial intelligence chip. Brain-like computing chip is a micro nano photoelectric device that can simulate the functions of neurons and synapses by physical and physiological deconstruction of human brain or animal brain. Brain- like chip is the basis of other high-level artificial intelligence tasks. Because of its excellent performance and the ability to subvert the "memory wall" problem existing in von Neumann architecture, it has extremely important research significance in the fields of artificial intelligence, computer science and so on. With the increasing need of the hardware hashrate and the invention of advanced methods, all kinds of brain-like smart chips have developed rapidly. Thus, This paper aims to evaluate the developments and applications of brain-like computing chips with a more thorough view. First, the basic structure and development of class chip is described; Second, the principles, methods, process, structures and overall frameworks are introduced in detail based on different brain-like computing chip models, such as Loihi, TrueNorth, and “Darwin NPU 2”. Besides, experiments are carried out according to the Tianjic chip model. The accuracy, memory, time-consuming and other data of Tianjic chip (“Tianjic Ⅰ” and “Tianjic Ⅱ”) and original model tested with full connected ANNs and small-scale CNNs are analyzed and compared, trying to figure out Complete and accurate deficiencies and advantages of the Tianjic chip. Finally, combined with the current research progress of brain-like computing chip, the obstacles existing in the research of it are summarized, and the future research trend and main goals of brain-like computing chip technology are prospected.

The insulation structure of the outgoing device area of UHV(Ultra High Voltage) converter transformer is complex and huge. It bears the effect of high current and high voltage for a long time. It will cost a lot to carry out on-site operation and maintenance test. At present, digital twin technology is becoming more and more perfect, and gradually introduced into the power industry from automobile, aviation and other manufacturing industries. Based on this, this paper introduces the digital twin technology into the high-end power equipment of UHV converter transformer. Considering the complex structure of converter transformer and its outgoing line device and converter bushing area are important typical accessories, this paper focuses on the construction of digital twin 3D model in this area, and carries out on-site operation and maintenance simulation test and high current functional response analysis under high voltage load. The digital twin model of the outlet device area is constructed according to the electro-thermal coupling physical field simulation model. The electro-thermal sensor is used to monitor the on-line voltage and current waveform in real time, and the interaction between the on-site operating parameters and the loading data of the digital twin model is realized. On the other hand, the intelligent extraction and identification of material area in the digital twin model is realized, and the material parameter performance can be changed according to the physical field environment, so as to adapt to the structural design and performance evaluation of different operating environments. Further, in order to analyze the relationship between the internal transient temperature of the converter bushing and its short-term current carrying capacity and long-term aging performance, the cyclic transient variation law of the hottest spot temperature of the high-voltage bushing is determined by fitting the diurnal cyclic load through the step load curve and combined with the thermal network. The electrical and thermal stresses are obtained by using the regional digital twin model of the outgoing device of the converter transformer. The long-term aging performance of the converter bushing is analyzed according to the classical ZHURKOV and CRINE electro-thermal combined aging model. This paper focuses on the 3D construction of digital twin model in the outgoing area of converter transformer. Its research method can be extended to key components such as converter body winding structure, oil paper insulation area and on load switch. The research results of this paper can provide theoretical guidance and technical reference for the insulation structure design of converter transformer body, especially for the structural design and operation maintenance of outlet device area, and can provide some theoretical guidance for on-line analysis of short-term current carrying capacity and long-term aging performance of converter transformer outlet device area.

Density peaks clustering (DPC) is a new algorithm based on density in clustering analysis. The algorithm considers local density and relative distance to draw the decision graph, quickly identify cluster centers, and complete clustering. The DPC has a unique input parameter and requires no prior knowledge or iteration. Since it was proposed in 2014, DPC has aroused great interest among scholars and developed rapidly. The paper first introduces the advantages and disadvantages of DPC and its basic theory. Secondly, we analyze the DPC optimization method from four perspectives. Then, five improved algorithms of DPC are introduced, and experiments on DPC and these improved algorithms are carried out in synthetic datasets and real-world datasets to evaluate the performance of these algorithms.

We consider an online scheduling problem on three parallel and identical machines with two grades of service (GoS) levels. Given are three machines and n independent jobs, each job and machine assigned the GoS levels of 1 or 2. The processing rule is that the job is allowed to be processed on some particular machines only when the GoS level of the job is no less than that of the machine, the jobs arrive over time and the three machines are idle at the same time only when all jobs arrived are completed. The objective is to minimize the makespan. In this paper, we show that any online approximation algorithm for the problem under consideration achieves the competitive ratio is no better than 4 / 3 , even for the case where the GoS levels of machines are 1, 2, 2, respectively. Moreover, we present an approximation algorithm with a competitive ratio of 5 / 3which is better than the best known algorithm whose the competitive ratio is 9 / 5 .

In the context of carbon peaking and carbon neutralization, developing new energy such as wind generation greatly is an effective way to deal with global climate change and energy crisis. However, due to the intermittence and volatility of wind generation, there are many influencing factors affecting it, and the influence mechanism between them is complex, so it is difficult to predict the wind power with a high-precision. In order to improve the accuracy of wind power prediction, this paper proposes a forecasting method based on LassoLarIC and Long Short-Term Memory (LSTM) for wind power from the perspective of variable selection and feature extraction, so as to extract the best feature set of wind power from different influencing factors, different historical data and different leading steps. Firstly, the optimal leading steps are selected by LassoLarIC, so the leading steps with the highest correlation are found from the historical data; Secondly, LassoLarIC is used again to extract the features from the selected leading steps, so the impact of different influencing factors in different leading steps are evaluated effectively, and the optimal feature set of wind power are obtained accurately; Finally, based on the optimal feature set extracted by LassoLarIC, LSTM is used to predict the wind power in the next hour. The prediction and comparison results show that the effect of feature extraction is obvious, the prediction accuracy before and after feature extraction is improved by more than 25%. The conclusions of this study can provide reference for variable selection and feature extraction of wind power prediction research

Surface Electromyographic(sEMG) is a kind of complex bioelectrical signal, and it is very important to select an appropriate feature extraction method. In this paper, the time domain analysis method, frequency domain analysis method and time-frequency domain analysis method are compared through experimental data, and the results of feature extraction of time-frequency domain analysis method are more representative, higher separation degree, and greatly reduce the one-sidedness of feature extraction. Finally, the energy eigenvalue of wavelet packet coefficient in timefrequency domain method is selected as the feature vector of signal pattern recognition to provide theoretical basis for real-time and accuracy of multi-motion pattern recognition.

The RLC AM mode is an important protocol in the 5G/NR air interface. Two problems found in the RLC/AM implementation are described in detail, which are the processing mechanism when exceeding the maximum retransmission times, and the choice of re-transmission PDU when the t-PollRetransmit expires under double-no condition. For these problems, the corresponding optimization scheme is given and implemented. The actual end-to-end test results between base station/terminals show the effectiveness of the proposed optimization scheme.

With the development of network technology, information security faces many threats. Malware shows an explosive growth trend, malware variants also emerge in endlessly, and malware detection is currently one of the necessary means to ensure information security. This paper conducts a detailed investigation on the malware detection mechanism based on behavior analysis, and introduces three types of malicious software detection technology, including methods based on statistics, methods based on system call graph, and methods based on structures. Generally, the goal of network intrusion and system intrusion is to steal sensitive information, such as file modification, or even controlling the host to perform remote attacks. Most of these operations require functions provided by the kernel layer, so it is inevitable to use multiple system calls. In addition, malicious behavior can be detected to a large extent by monitoring the execution of system calls. This paper introduces the detection technology based on system call graph, including function call graph method, control flow graph method, and data flow graph method. The paper also discusses the features of advanced detection technology based on system call graph. Moreover, the paper analyzes the data set, modeling language and evaluation performance index used for system call graph based detection technology, which will help security researchers evaluate the applicability and limitations of detection algorithms to identify malicious behaviors, so as to design more efficient detection strategies. Finally, this paper clarifies the role of features in malicious behavior identification and the characteristics of various detection methods, which can help security researchers to break through the limitations of traditional detection technology.

With the development of deep learning, the performance of vehicle detection algorithms based on deep learning is constantly improved, which plays an important role in the construction of intelligent transportation. Single-stage target detection model is widely used in vehicle real-time detection because of its advantages of detection speed. In view of the low detection rate of small objects in images, this article proposes a vehicle object detection method based on the improved YOLOv4 algorithm, using k-means clustering algorithm to re-create a anchors suitable for the UA-Detrac dataset and improve the PANet. Compared with other target detection methods, the improved algorithm can effectively detect small targets and improve the Precision, Recall and mAP of vehicle targets.

The load prediction of electric vehicle charging stations is of great significance to the planning and scheduling of charging stations. In this paper, the charging load prediction of electric vehicle charging station is analyzed and studied, and the charging load of the future hour level is predicted by the short and longtime memory neural network. First of historical load data interpolation processing complete data sets, and then build LSTM neural network model of data for training, by comparing different training set and the proportion of the data set under the condition of MSE is worth out of prediction scheme, escape the cycle of neural network (RNN) step after a long time the problem of dissipate, intuitive experimental results.

By using the national secret SCTO module of theFT-2000/4 development board, based on the three algorithms of SM2, SM3 and SM4 to realize the packaging, encryption and file point-to-point transmission of local random files. The above functions are integrated into a complete set of CLI tools; at the same time, a distributed file sharing system is built, and the encryption system on the hardware side of the development board is transplanted to the computer software side by using .NET. The user node can connect to a single or multiple servers to build a file index directory. Relying on the index server, it still uses point-to-point encrypted transmission to realize multi-user file sharing service users; in order to enhance the interactive integrity of the system, this article uses the UI graphics library of the .NET platform, and the statement statement XAML to write an independently rendered graphical interface. The operating system designs a set of cross-platform, portable UI login interface and file download and upload management interface, and uses .NET Runtime data transmission to achieve portability compatibility with mainstream APP software. National innovation and entrepreneurship training program for college students 413.

Attackers can exploit vulnerabilities in web applications to commit malicious acts such as corrupting application functionality and Trojan horse implantation. For injection vulnerabilities in Web applications, existing methods are limited by the variety of programming languages and the difficulty in extracting semantic information to detect complex vulnerabilities. The paper proposes a pattern matching-based method for identifying injection vulnerabilities in Web applications, which transforms vulnerability identification into path matching in graphical databases by modifying the code property graph to enable it to handle more complex inter-functional relationships in Web applications. In this paper, we designed and implemented a prototype system, VulnFinder. By collecting 100 randomly selected Github high star open-source projects as the dataset for performance testing, we found 262 real vulnerabilities and conducted comparison experiments with static scanning tools RIPS and Cobra, VulnFinder far exceeded the comparison tools in terms of vulnerability determination accuracy, and in terms of the dataset VulnFinder was 94% accurate in determining vulnerabilities in the dataset. In scanning large projects, VulnFinder was approximately 21% more efficient than the comparable methodology tool RIPS.

Cell phones have become an essential part of daily life, and users have a strong demand and dependence on them, while the online and offline synergy also provides a good platform and environment for teaching and research activities. Therefore, the online real-time database system in the field of education has been improved, and its main function is to collect and manage data such as student information and teachers' teaching situation. Based on the computer algorithm, this project develops an easy-to-use and convenient APP based on cell phone terminals, and describes in detail the development method of the communication module of the APP. The APP designed in this paper can help teachers not to worry about the time they spend on uploading data after work, and at the same time, it can keep students updated with the latest information and school status, and improve the stickiness and satisfaction of teachers and students.

For the wireless sensor network of airborne test system in flight test, in order to obtain better communication performance and improve the signal interference to noise ratio, the sensor nodes in the network will increase the transmission power, which will lead to the aggravation of the interference between the nodes, and bring negative impact to the wireless sensor network, finally cause the decline of overall network performance, and energy waste of sensor nodes. In order to solve this problem, this paper introduces the balance factor and designs the utility function considering a distance factor based on the Nash bargaining solution cooperative game theory. At the same time, a new power control algorithm is derived. The simulation results show that the proposed algorithm is simple to convergence, which could effectively improve the power consumption and the overall performance of wireless sensor network system, in addition, prolong the service life of sensor and achieve higher network utility.

Speech conversion has significant applications in medical, robotics, and other industries. With the rise of deep learning, CycleGAN is widely used in speech conversion technology. However, the existing CycleGAN-based methods do not consider the speech signal’s temporal and spatial features. In addition, the training of CycleGAN is difficult to converge due to the gradient disappearance problem of the generator. We propose SSCGAN, whose generator is a U-shaped encoder-decoder network that extracts the temporal and spatial features by using 1DCNN and 2DCNN in parallel. A feature fusion module based on multi-scale mixed convolution is embedded between encoder and decoder to achieve high-level fusion of spatial features and temporal features. To make the network training more stable and easier to converge, SSCGAN uses Wasserstein distance instead of the original Jensen–Shannon divergence to calculate the distance of the probability distribution, which can alleviate the gradient extinction problem for generators. In addition, SSCGAN utilizes the PatchGAN structure in the discriminator, which considers the samples’ local details by dividing them into different patches. It can improve the discriminative ability of SSCGAN. The experiment results in the nonparallel corpus database VCC 2018 show that SSCGAN is superior to existing methods such as CycleGAN-VC, StarGan-VC. In inter-gender speech conversion, the MSD of SSCGAN is decreased by 0.162 on average compared to other methods, and in intra-gender speech conversion, the MSD is decreased by 0.118 on average. In subjective evaluation, participants also think SSCGAN is the best.

BeiDou Navigation Satellite System is an important national space-time infrastructure and an important tool for realizing social informatization. BeiDou satellite navigation system has been widely used in transportation, hydrology, meteorological monitoring, and other fields since it provides services. At the same time, the BeiDou satellite navigation system also needs to be further applied in ground navigation. Given the domestic situation, the best practice place is the high-speed field, such as the wide plain, highways, and so on, and the focus is to design a low noise, low power consumption, broadband navigation satellite receiver frequency synthesizer. Considering the selection of frequency points and loop bandwidth of the navigation satellite, this paper selects the phase-locked fractional frequency synthesizer. The linear model of the system is established. After introducing noise, the phase noise model of the system can be obtained. In this paper, the 65 nm CMOS process is used to design and realize dual-mode broadband VCO, decimal frequency divider, frequency discriminator, charge pump, automatic frequency band selection circuit, and loop filter. After simulation verification, the design index is completed.

The quality of teaching is an important indicator to measure whether the computer network education system is perfect, whether it can achieve the expected results, and to what extent it has developed. Under the traditional mode of China's colleges and universities, the main method of teaching inspection is manual. Due to the information construction of university teaching, the establishment of a computer network-based teaching testing system is the right thing to do. The teaching testing system studied in this paper is an open platform which inputs all teachers into it and then analyzes the course content according to different users to derive the learning effect, knowledge mastery and ability level of each student and finally evaluate them. The system described in this thesis is developed by computer network technology, which has the following functional modules: login management; data acquisition; a series of operations such as generating education class grade information reports after the initial processing of the system is completed through the WEB front-end page.

The analysis of security protocols is of great significance to ensure the safe communication of computer networks. Using formal methods to analyze security protocols has been a research hotspot in this field. Based on the run modes method and Petri net, this paper presents a specific method for analyzing security protocols. Firstly, the concepts of basic Petri net model (BPN) and attack Petri net model (APN) of protocol run modes are given, then the algorithm for constructing APN through BPN is proposed. Finally, the effectiveness of this method is illustrated by analyzing Helsinki protocol.

For problems that are difficult to extract characteristics of misalignment fault signal of harmonic reducer of rotating unit of satellite antenna operating in space environment, a fault diagnosis method of optimized whale optimization algorithm based on whale optimization algorithm and Teager energy operator envelope demodulation was proposed. Firstly, the envelope entropy which is easy to the characteristics of fault signal, it is often used as fitness function and objective function of whale optimization algorithm (WOA). Secondly, whale optimization algorithm has the advantages of simple operation, less adjustment parameters, and strong ability to jump out of local optimization. It is suitable for whale optimization algorithm optimization. Finally, the characteristic modal component (IMFC) is processed by Teager energy operator (Teo). This is used to analyze the signals collected from both ends of the input and output shaft of the harmonic reducer. We could find in results that Fourier transform method, this method improves the decomposition efficiency, and has good performance in the decomposition principle and accuracy of fault signal extraction parameters fast compared with the traditional. It could realized the accurate diagnosis of the misalignment fault signal of the harmonic reducer, and had certain research significance and engineering practical value.

With massive data, complex information, and the increasingly expanding huge stock pool, it is always difficult for investors to quickly and accurately choose stocks with good inherent value and predictable constant growth. Therefore, stock analysis before investment is vital. In this paper, by applying classical K-means clustering algorithm in data mining, 100 US stocks randomly selected from Yahoo Finance are first classified into different performance groups based on their historical trading data in the recent one year, then detailed analysis of the typical stocks in each performance group is given from the perspective of technical analysis, attempting to find some important features for simple and rapid evaluation of stock investment value, in addition, corresponding investment advice is also proposed based on some statistical analysis index.

Under the development of people's living environment, more and more people are willing to use their money to invest in financial projects such as stocks and insurance. Nowadays, science and technology are widely applied in people's life. Machine learning is one of them. Machine learning is particularly important to apply to stock forecasting to better meet the requirements of people who want to gain more benefits. The purpose of this work is to compare using GRU, LSTM, and bidirectional LSTM's MAE and RMSE on the closing price. The method of this experiment is to compare with root mean squared error (RMSE) and mean absolute error (MSE) after the input variables of the past 63 trading days passing through those three models. The results of the experiment indicate that MAE of GRU model is lowest. Still, only nine of fifteen experiments show that RMSE of GRU model is lowest, and five of fifteen experiments show that RMSE of LSTM is lowest. One of fifteen experiments expresses that RMSE of bidirectional LSTM has the lowest RMSE. Thus, GRU is considered to be the best model for stock price regression.

The operation of urban rail transit will surely have effect on the original conventional bus system. To realize the development system of “Integration of the two networks, coexistence and common growth” between the rail transit and the conventional bus lines, this article aims to further optimize the layout of the conventional bus lines in connection with the urban rail transit. First of all, this paper analyzes the connection between the original conventional bus stations and the existed rail transit stations, and determines the generating steps of the candidate lines on this basis. Later, this article constructs an optimization model for bus network connected to rail transit, and tries to improve the traditional genetic algorithm by introducing simulated annealing operator. Finally, this article takes Xiguan Cross Station of Lanzhou Rail Transit Line 1 as the example, and applies the improved genetic algorithm to clarify the example, which is meant to verify the effectiveness of the algorithm and to provide the reference for the development of public transportation in Lanzhou City.

The online toxic comments cause enormous harm to the society, where toxicity is defined as anything rude, disrespectful or otherwise likely to make someone leave a discussion. To have a safer, more collaborative internet, grateful contributions are made by a main area of focus on machine learning models to identify toxicity in English, whereas part of misinformation disseminates in other languages. Over the past year, pretraining multilingual language models give rise to impressive gains for cross lingual toxicity classification. This paper presents an approach to build toxicity models applying the Jigsaw Multilingual Toxic Comment Classification dataset provided by Kaggle. We set our ensemble model in three parts based on Besides, we implement subsample, Pseudo-labeling with open-subtitles, translating non-English languages to English language, and Post Processing to improve the classification accuracy indispensably. Our final model achieved an AUC of 0.9469 for the training set and 0.9485 for the validation set, demonstrating the effectiveness of performance under cross-lingual toxicity detectors.

Video summarization technology extracts key frames or video clips that can most effectively express video content from the original long video, so that the summarized video clips contain the content information that users are most interested in, which is convenient for users to quickly browse and retrieve the video. In this paper, we regard video summarization as a sequence annotation problem. Different from the existing methods using recurrent models, this paper proposes a fully convolutional neural network model combining spatial attention mechanism to solve the video summarization problem. Firstly, the pre-training model is used to extract the cube features of the input video frame, then the cube features are aggregated into the attention vector by the attention mechanism, which is input into the fully convolutional neural network model for binary classification, and then the video summary is generated. Extensive experiments and analysis on two benchmark datasets demonstrate the effectiveness of the proposed method.

Embedding-based methods have currently become the hit of the entity alignment, among which Graph Convolutional Networks model(GCNs) is receiving widespread attention because of its powerful capability of identifying isomorphic structures. Note that the aligned entities in real-world knowledge graphs always have non-isomorphic neighborhood structures, which may cause the GCN-based method to generate different representations for them. In addition, most embedding-based entity alignment methods ignore the relation information and semantic information of entities, which can cause inferior performance because of the lack of necessary information. This paper proposes a novel KG alignment framework named MGN(Multi-information aggregation Graph neural Networks), which aims at capturing the multiaspects features to learn the representations of entities at the same time. Our method can make up for the lacking of feature information and ease the non-isomorphism of structure effectively. The experiments on three real-world crosslingual datasets demonstrate that the MGN can get better results over the state-of-the-art alignment methods.

This paper focuses on integrating sensor network into traditional garden irrigation. Through different terminal sensors, soil temperature and humidity, trace elements, PH value and other parameters through the Internet of Things technology to achieve reliable information transmission, through the regional controller using LORA protocol, wireless unidirectional or bidirectional 485 communication, to achieve intelligent control.The resolution of soil temperature is 0.1℃and the accuracy is 0.5℃.The water resolution is up to 0.1%, the accuracy depends on the water content, but not more than 3%, high measurement accuracy, fast response. Sensor to obtain soil moisture, when the occurrence of diseases and insect pests, the system will feature extraction of the picture, identify the disease and insect information in the picture, in order to help users to solve problems. Through the complementation of APP, PC and terminal sensor network irrigation equipment, integrated services are created to realize the intelligent system management of sprinkler irrigation equipment. The system has been proved to have real time, practicability, memory and reliability.

User behavior contains a large amount of value information, and assessing user behavior can be a good guide for enterprises to develop service policies. The current power user behavior assessment model mainly relies on the past evaluation theories, and the selected indicators are not comprehensive enough, leading to unsatisfactory evaluation results. In order to optimize the above problems, the construction of the power user behavior assessment model based on 95598 user portraits will be studied. After text conversion, data cleaning and variable attribute analysis of 95598 grid customer service phone data, 95598 user portrait is established. The logistic regression model is combined with decision tree algorithm to complete the construction of the electric power user behavior evaluation model. The results of the model feasibility study show that the correct rate of the constructed electric power user behavior assessment model is higher than 95%, and the model has a high accuracy of predicting the abnormal behavior of electric power users, which has good practical application effect.

Since drawing a complete set of high-quality Chinese calligraphy fonts by hand is time-consuming and labor-intensive, in order to realize the automatic drawing of Chinese calligraphy fonts, this paper proposes an encoder-decoder network model based on residual structure. The model realizes the automatic generation of Chinese calligraphy fonts through the adversarial training of the generator and the discriminator. Experiments show that our method can not only generate realistic calligraphy fonts, but also has the ability to generate a complete font library.

This paper firstly researches the few-shot learning technology and the capsule routing theories. The iterative consistency mechanism of dynamic capsule routing is combined with the prototype network to design a better generalization performance. And a few-shot image classification model with better robustness for the posture of the sample in the test is proposed. In view of the defects in it, a multi-marginal cosine loss function is proposed for the model. Finally, the CosGCN model is proposed to ensure recognition accuracy and improve the computational efficiency of the model training. Experiments are carried out on the small sample dataset Omniglot and miniImagenet datasets. The results show that the model has better posture robustness.

Nowadays, automobile intelligence is the general trend, but new technology brings dividends at the same time, but also bear risks; Intelligent connected vehicles are quite different from traditional vehicles in terms of product structure and function realization. Therefore, while bringing convenience to users, they also have security problems such as unauthorized personal information and important data collection and utilization, network attacks, network intrusion and other network security problems. And driving automation system random failure, insufficient functions caused by road traffic safety problems. This paper will study the Internet of vehicles system from domestic standards, Internet of vehicles asset security risks, and security testing.

In the era of big data and artificial intelligence, the Smart Classroom, a new type of classroom, is becoming popular in primary and secondary schools. Meanwhile, the evaluation of its effects has gradually attracted much research interest. Through two years of follow-up and data mining, this study evaluated three consecutive semesters' final Chinese grades of 195 junior school students who participated in the experimental teaching of Smart Classroom with algorithm called latent growth model(LGM). The results showed that despite the significant individual differences in starting grades, Smart Classroom produced significant positive potential growth rates which averagely increased 5.536 points per semester. This study conducted a more accurate algorithm compared traditional analysis of means and variances of educational dates, as well providing a more precise empirical evidence literature of effect of smart technologies on Chinese language educational in junior high school.

With more and more uncertain factors affecting enterprise operation in market economy, enterprises are facing with more and more diversified risks. This requires accurate prediction of the future financial situation of the current enterprises in the market as far as possible, and take preventive measures in advance for the companies with hidden financial crisis. Based on the dynamic CBR model, this paper considers the influence of the year of the training sample in the current year, that is, the distance of the sample from the current year on the model's early warning performance, and builds a dynamic financial crisis early warning model with CBR as the basic classifier considering both concept drift and sample time weight. In this paper, the samples of the cloud financial sharing center of the current year are given a higher weight, while the samples of the distant year are given a lower weight. The results show that the prediction accuracy of THE CBR model considering the time weight of the sample has a stable improvement compared with the CBR model without considering the time weight of the sample.

Four rotor unmanned aerial vehicle (UAV) control of the bottom of society, this article is based on the cascade PID algorithm with the improved design and double closed-loop position controller and attitude controller, the outer ring of position and attitude controller adopts hyperbolic tangent function as the position loop, position controller of the PID as the inner ring speed ring, inner ring USES the intelligent PID control algorithm as angular velocity loop. In this paper, detailed steps of UAV modelling are given, and experiments are carried out on ROS, PX4 and Prometheus. In the experiments, the algorithm is verified and compared with traditional PID and cascade PID control algorithms. The experimental results show that the double closed-loop intelligent controller designed in this paper can ensure the flight stability of the quadrotor UAV and the accuracy of the trajectory tracking, and has a certain anti-interference ability.

Existing Spatio-temporal graph models consider traffic flow prediction as a "node-level" task, and it is difficult to make decisions based on the results of point prediction alone. The modeling idea is to explore the hidden graph structure that better reflects the spatial dependency of the road network from the traffic flow "signal" to replace the less accurate graph structure that is calculated directly based on the geographical location. Here, we propose the Dilated SpatioTemporal Graph Convolutional Network (DSTGCN) model and the Adaptive Graph Recurrent Neural Network (AGRNN) model for traffic flow prediction. The main work is to propose a self-learning transfer matrix for directed graphs that can automatically discover potential graph structures from the historical traffic of each node, improve the gating mechanism in temporal convolution, and introduce probabilistic prediction into the traffic flow prediction problem to enhance the dimensionality of information used for decision making. DSTGCN and AGRNN are evaluated on two publicly available traffic datasets. METRLA and PEMS-BAY were evaluated and achieved the fastest computational speed (compared to other Spatio-temporal graph models) and the best prediction results, respectively.

As an important branch of modern cryptography, secret sharing can diversify security risks and effectively ensure the security of information. The exchange of secret shares is an important part of secret sharing. How to ensure that the secret shares submitted by each participant does not affect fairness is an important topic to be studied in the field of secret sharing. This scheme proposes a fair (t, r, n) threshold secret sharing scheme, in which honest participants can recover the plaintext, while cheaters cannot recover the plaintext, and can resist collusion attacks by participants to ensure the fairness of the scheme.

Aiming at the problem that pedestrian trajectory prediction network based on Encoder-Decoder structure is easy to lose part of trajectory information in the process of long sequence encoding and decoding, Generative Adversarial Network based on Temporal Attention (TA) is proposed in this paper. It is used to assign influence weights to the trajectory information of the encoding and decoding layer, so that the model can make full use of the trajectory information useful for predicting the trajectory in the future and reduce the influence of redundant information. This paper adds a TA to the encoding and decoding layers of the Generative Adversarial Network, and trains it on the ETH and UCY datasets. Experimental results show that the proposed network has better prediction accuracy than existing methods.

In order to avoid the occurrence of Automated Guided Vehicle (AGV) collisions in the intelligent warehouse system and improve the cooperative efficiency of multiple AGVs in the intelligent warehouse system, an improved A* path planning algorithm is proposed considering the possible occurrence of collisions in the intersection of multiple AGVs. This article improves the traditional A* algorithm, dynamic mechanism and conflict resolution strategy are introduced, and the path and mode of AGV are optimized to improve operation efficiency. The experimental results show that the improved A* path planning algorithm can effectively resolve conflicts and shorten operation time.

The goal of PReID (Person re-identification, PReID) is to maximize the function of distinguishing pedestrians in order to deal with the influence of uncertain background. The multi-layer learning method based on the neural network is applied to the background of extracting the feature information of the image, according to the composition method of the pedestrian structure, the problem of pedestrian rematching is considered in detail. Designed and proposed a multi-flow pedestrian re-identification algorithm, based on the neural network framework of global and local feature extraction, combined with multiple loss functions to optimize the pedestrian re-identification process, and achieve the purpose of pedestrian re-matching. Use long and short-term memory (LSTM) networks to model pedestrian images, focusing on the combination of information between global and local features. Using triple network multi-channel feature extraction method to expand the breadth of image extraction. In addition, the introduction of multiple losses to optimize the fusion of local information reduces the adverse effects of this part on network discrimination, realizes the complementarity of local body features and global features, and the complementarity of recognition tasks and ranking tasks to achieve higher feature learning accuracy. Experimental results show that the proposed method is much better than the latest method on three challenging data sets (including Market-1501, CUHK03 and DukeMTMC-reID).

With the rapid development of virtual reality and convolutional network technology in the field of computer vision, enabling deep learning technology to 3D reconstruction has become a research hotspot. 3D reconstruction based on virtual reality and convolutional networks is widely used in the field of 3D reconstruction of ancient buildings. This paper studies the 3D reconstruction of ancient buildings based on virtual reality and convolutional network technology. Firstly, the principle of 3D reconstruction and the method framework for fast reconstruction of ancient models are introduced. Then, on the basis of semantic segmentation of 2D images, the segmented image blocks of indoor scene elements are extracted, and the 3D model matching model of convolutional neural network is analyzed. 3D technology plays a very important role in the reconstruction of ancient buildings.

Aiming at solving the difficulties of modeling factor processing and forecast modeling caused by time correlation and non-linear variation of low-temperature disastrous weather processes during winter, this paper takes the winter low-temperature cold damage index calculated from the daily temperature and precipitation data in Guangxi as the forecast factor. Based on the NCEP/NCAR reanalysis data and forecast field data, a large number of characteristic factors are extracted by using the random forest algorithm (RFS). Combined with the long-short-term memory neural network (LSTM NN) in the nonlinear time series memory characteristics and nonlinear timing dynamic system control advantages, we established a model for low-temperature cold hazard forecast with 72 hours based on the random forest and LSTM_NN. Experimental results indicate that the current method has high prediction accuracy, good stability, and displays better applicability to nonlinear low-temperature cold hazard forecasting.

Studies on prevention and control practices show that the mobility of personnel is the main reason for the spread of the epidemic. The epidemic prevention and control work is gradually becoming normalized. By giving full play to scientific and technological advantages, mining the intrinsic value of information and analyzing and utilizing trajectory data, transmission can be effectively avoided from the source, which is of great significance for epidemic prevention and control. This paper realizes the analysis of personal trajectory data through the research of point aggregation algorithm and visual field following algorithm, processes the track big data by using big data technologies such as Kafka and Spark, displays the track data according to WeChat applet and visualization large screen, and designs the individual trajectory and track big data visualization end system based on the trajectory data. The experimental results show that the system can display the whereabouts in real time, which has good practicability and versatility, and has positive reference significance for helping residents to travel and supporting the government's decision to fight the epidemic.

Wireless sensor network (WSN) is a wireless communication self-organizing network composed of multiple sensor nodes, which can be used to obtain the information of the monitored objects in the target area. However, due to the characteristics of the nodes that make up the wireless sensor network, the network has the characteristics of large scale, low energy consumption, limited computing and storage capacity. These characteristics play a decisive role in the application of wireless sensors. In this paper, a weight algorithm based on the adaptive sleep mechanism in wireless sensor networks is proposed to improve the network performance. The algorithm mainly aims at the adjustment of the weight, and then to evaluate the importance of the data. By applying the weight algorithm, the original algorithm based on the sleep scheduling mechanism is improved. The performance of the whole network is significantly improved compared with the network optimized by LEACH algorithm and GAF algorithm. The important data can be transmitted to the sink node as soon as possible, which effectively improves the link quality. The energy consumption of the nodes in the network is balanced through the adaptive sleep scheduling algorithm without location information, which improves the network performance and prolongs the network lifetime as much as possible. The simulation results show that the proposed network can perform better than the fully connected network in most cases. In the application and promotion of wireless sensors, there are more in-depth research and wider applications.

The sharing economy model based on blockchain technology can hand over user data and market data to users and communities. Based on obtaining the consent of users, anyone can use the shared data to create useful products and services to build a new business model and continuously upgrade the user experience in the competition. Due to the imperfection of the trading system, the non-standard trading mode, the uncontrolled trading process, and the digital economic value sold are easy to be damaged. The quality of the digital economy purchased is difficult to guarantee. This paper adopts the decentralized transaction mechanism of asymmetric encryption algorithm, the two-way anonymity mechanism of hash encryption algorithm, the Ownership Authentication Mechanism Based on digital signature algorithm, the data confidentiality mechanism of data encryption technology and interface communication, and the fraud control mechanism based on qualification authentication to realize the point-to-point transaction of the digital economy. It provides a safe and reliable transaction environment for both parties of the digital economy transactions as well as a security guarantee for realizing the value of the digital economy.

In order to facilitate government departments to assess security risks and prevent infiltration, it’s necessary to recognize the IoT device from open data by the method of Named entity recognition (NER). In this study, a hybrid method of chinese NER which contains neural networks and templates is proposed to extract IoT device information from open web. The model first initiates characters of input through BERT, which can create character embedding while reduce dependency with external datasets. Then, the initiated vectors are fed into a multi-layers BiLSTM model to encode the contextual representation of each character, and a linear CRF layer after BiLSTM is used to assign the scores to every character for entity annotation. In addition, a dictionary and rule base is constructed based on the characteristics of IoT devices to correct the annotation results. We experimented the methods on the dataset of IoT, the results prove the proposed model achieves better recognition effect than other models, and the F-scores value is 88.9%.

With the wide application of blockchain, the access control of on-chain data and privacy protection are two urgent concerns to be solved. Although many access control systems based on attribute-based encryption (ABE) have emerged, they are not satisfactory for achieving the privacy protection of data owners’ attributes and the revocation of data consumers’ attribute sets. An anonymous ciphertext policy ABE with attribute revocation (AR-ACP-ABE) was unveiled by adopting the idea of attribute revocation based on broadcast encryption. Then, a new privacy-preserving attributebased access control system was constructed. It is an extension of the AR-ACP-ABE scheme by adopting the technique of blockchain and outsourced computation. The new system supports completely hidden access policies and allows data owners to revoke data consumers’ attribute sets. In addition, data consumers only need to perform one exponent operation on the target group during the access phase. Security analysis shows that the new system meets the properties of correctness, confidentiality, and integrity.

This paper cites the carbon neutralization index, calculates the carbon absorption and carbon emissions of 34 provincial administrative regions in China, ranks the carbon neutralization index of each province and city, comprehensively considers the area and industrial layout of each province and city, selects the six provinces and cities with the lowest index, Shanghai, Tianjin, Shanxi, Shandong, Ningxia, Jiangsu cluster analysis of the region, determines the latitude and longitude coordinate area of establishing ecological reserves, and promotes the process of achieving the carbon neutralization goal in China.

In recent years, the development of science and technology has driven the progress of Internet technology. With the progress of various new technologies, new data analysis technologies have also been derived. 4G network can no longer meet people's requirements for network speed. The emergence of 5g provides a strong basic support for online education platform. Online live education platform is more and more valued and recognized by people. This paper describes the function development of online education platform. Finally, this paper describes the main application advantages of the live broadcast platform.

In this paper, a hybrid model based on BERT-BiLSTM-CBAM is proposed to classify the sentiment of online reviews more accurately. Firstly, the BERT model is used to pre-train the text information to obtain feature vectors; then the feature vectors obtained from Bert pre-training are stitched and reorganized through a bi-directional long and short-term memory network (BiLSTM) and CBAM mechanism to obtain new feature vectors. Finally, these new feature vectors are input to the fully connected layer, and the sentiment category of the text is calculated by the SoftMax function. Experiments on the Amazon reviews and Yelp reviews datasets show that the method is more accurate and reliable on both datasets.

Neuromorphic vision sensor is a new type of visual sensor inspired by biology, which can process image information by imitating the operation of biological retina. It is different from the traditional camera that captures the image frame at a fixed rate, but asynchronously outputs the event flow related to brightness change information, with the characteristics of low delay, high time resolution and high dynamic range. Pedestrian detection based on neuromorphic vision sensor is one of the research directions in the field of computer vision, which can be applied to automatic driving, video surveillance and other fields. This paper introduces the principle of neuromorphic vision sensor, and introduces the pedestrian detection method based on the sensor, analyzes the existing detection methods and related data sets, and finally prospects the development trend of this direction.

With the development of intelligent driving technology, the progress of hardware technology. Visual perception technology based on deep learning has been applied more and more in the field of intelligent driving. Vision, as the main part of information acquisition, is the core of automatic assisted driving technology. Based on the intelligent vehicle as the research platform, the use of ROS combined the technology of deep learning design implements a lane detection algorithm, in the car on the road ahead uninterrupted detection, extraction of the road lane information, to guide the car driving direction, effectively improve the security and intelligent vehicle driving process, to achieve the function of unmanned vehicle automated driving laid the foundation.

With the rapid development of tourism, more and more people choose to travel by air, and the price fluctuation of air tickets is large and the law is not significant, which makes it necessary to use some big data technical means to predict the change law of prices. The research and analysis of the characteristics of air tickets are helpful to analyze the internal value of tourism. Based on the web crawler technology, this paper obtains a large number of flight information from Shanghai to Guangzhou. Through data exploration and preprocessing, Random Forest classification model is established to predict the purchase of air tickets. The prediction accuracy of whether to buy or not is 96.96%. Then, ARIMA model is established to predict the ticket price trend. The deviation rate of the model is less than 18%. The future ticket price can be effectively predicted through Time Series. The machine learning model established in this paper provides model support for the study of the characteristics of air tickets. The machine learning model established in this paper provides model support for the study of the characteristics of air tickets and technical support for the specified price strategy of the aviation industry, so as to promote the development of tourism.

With the arrival of the era of big data, scientific and rational use of historical data has become one of the convenient ways to earn information. Search index records hundreds of millions of search attention and demand, for enterprises to grasp the market to provide the necessary database. This analysis takes the search index of Huawei brand on the Baidu Index website as an example. Based on R-STUDIO, the monthly search index of the brand is extracted as historical data, and the time series is used for analysis and prediction.

Aiming at the multimodal characteristics of multi-objective optimization problems, which need to provide decision-makers with a variety of choices or various information, this paper proposes a multi-modal multi-objective particle swarm optimization algorithm using ring topology and neighborhood disturbance strategy. Using the system topology, the population can form multiple independent search environments without specifying any parameters. To better global exploration and local development, an automatic conversion mechanism of global search and local search is proposed. The stagnation detection strategy is introduced to disturb the neighborhood optimal particles to improve the diversity of particle swarm optimization. Prevent the algorithm from premature convergence to a pareot optimal solution set. Experimental analysis verifies the feasibility of the proposed algorithm. Experiments are carried out with the comparison method. The experimental results show that the proposed algorithm can better ensure the diversity and convergence in the target space, which is to find more complete and evenly distributed optimal solution sets.

In recent years, the society has paid more and more attention to the problem of excessive greenhouse gas emissions, and the carbon emissions in the transportation industry have accounted for one-fifth of the global carbon emissions. Therefore, reducing the carbon dioxide emissions in the transportation industry can help The whole society has made a huge contribution to energy conservation and emission reduction. In order to improve transportation efficiency while reducing carbon dioxide emissions, transportation methods and transportation routes should be selected reasonably. This article is based on these aspects as a starting point to study the selection of transportation routes and methods in multimodal transportation. Based on time constraints, this paper comprehensively considers carbon emissions and transportation cost factors, constructs a low-carbon multimodal transportation route optimization model that targets the carbon emission cost, transportation cost and transshipment cost of cargo transportation, and uses genetic algorithms to obtain the optimal multimodal transportation. The genetic algorithm is used to obtain the optimal multimodal transport combination mode and route, and the feasibility of the model is verified through case analysis. The results show that the low-carbon multimodal transport mode can effectively reduce carbon emissions by integrating the technical and economic advantages of multiple modes of transportation, has a significant effect on energy conservation and emission reduction, and is conducive to promoting low-carbon sustainable economic and social development.