This PDF file contains the front matter associated with SPIE Proceedings Volume 12174, including the Title Page, Copyright Information, and Table of Contents.

This paper introduces a chemical process control system based on the Internet of things, which combines ESP8266WiFi module and multiple sensors. The system includes temperature control, pressure control, concentration control, humidity control, raw material prediction, intelligent adjustment, intelligent judgment, intelligent scheduling. Users can establish a connection with the system by logging in PC or mobile APP to monitor and control the factory's production process in real time. And through the intelligent analysis of product demand, the production line optimization, the optimal configuration of services. Establish a set of self - perception, self - communication, self - learning self - optimization of intelligent control system. The control system of the chemical process is simulated, and the purpose of real-time monitoring, real-time judgment and real-time scheduling is achieved. It shows the feasibility and necessity of intelligent control. Intelligent chemical industry and intelligent manufacturing are the necessary trend of the development of science and technology in the future.

With the development of society, environmental problems are becoming more and more important. We should develop the environment while developing the economy. People's awareness of environmental protection should also continue to improve, and begin to pay more attention to ecology. Cities all over the world should take the road of ecological development and build ecological cities. The construction of ecological city is the symbol of human civilization and the inevitable trend of urban development. This leads to a very important concept in eco-city, that is the plant wall. Based on this, the author designed a simple intelligent planting system based on SCM. The system uses MCU "Arduino Uno R3" as the main control chip. The photosensitive resistance sensor and humidity sensor are used to measure the current information and feed back to the MCU. LED lights and feeders work to realize automatic watering, lighting and feeding fish on time. Finally, the entire system can be controlled through a wireless connection to the phone. In this paper, the hardware and software parts of the system are mainly designed, and the parameters of the system are tested, which basically realize the function and improve the intelligent degree of plant planting.

Timely acquisition of location information of people in distress is an important prerequisite for successful rescue. This paper analyzed the application status of the emerging Internet of Things technology-LoRa in the field search and rescue system, and designed a field rescue system based on LoRa and Beidou. LoRa positioning technology was used as the auxiliary positioning means of satellite positioning technology, which was used to solve the problem of satellite positioning accuracy decline when the line-of-sight condition could not be met. Finally, the IDEF0 function modeling of the system has a certain engineering significance.

Existing energy consumption collection systems cannot support the diverse customer services under the smart grid vision due to issues such as low data collection granularity, lack of perceived information types and limited carrying capacity of communication network. In this regard, based on advanced international standards, we design a new type of smart meter that can surpass many metering applications. In order to leverage its enhanced data measurement and situation awareness capabilities, an IoT (internet of things) based new power consumption measurement and perception system framework is proposed, and successfully pilot-deployed in the Sino-Singapore Tianjin Eco-City Smart Energy Town project. Finally, advanced applications deployed in this solution as refined power consumption data measurement is introduced in detail, which provides significant benefits in improving the user service and system operation and maintenance of utilities.

In the past decade, Electric Vehicles (EVs) have slowly replaced traditional gasoline vehicles. EVs not only are environmental-friendly but also, when used in combination with the smart grid, open up new possibilities and enable the vehicle smart grid ecosystem, commonly referred to as Vehicle-to-Grid (V2G). This not only encourages the switch to eco-friendly EVs/Plug-in Hybrid Electric Vehicles (PHEVs) but also actively supports grid load management and is new to whole entities implicated in such ecosystems providing economic benefits. Nevertheless, the protection of privacy and security are still serious issues for smart grids. The types of equipment used in V2G are small, inexpensive, and limited in resources, so it is vulnerable to multiple attacks. Protocols designed for V2G systems must be safe, lightweight, and protect the privacy of vehicle owners. Physical security is also essential, as EVs and charging stations (CS) are generally not protected by humans. In order to solve these problems, we propose a Secure User Key-Exchange Authentication (SUKA) protocol based on Physical Unclonable Function (PUF) for V2G systems. The proposed protocol uses PUFs to implement two-step mutual authentication between the EV and the grid server. It's lightweight, safe, and privacy protected. The simulation shows that compared to other advanced V2G authentication protocols, our proposed protocol has better performance and provides more security features. Formal security models and analysis are used to demonstrate the security of the proposed protocol.

In order to effectively solve the problem of high reliability of computing units in harsh environments, a design method for PHYTIUM computing units based on OpenVPX is proposed. Firstly, the characteristics of the OpenVPX bus specification are analyzed, Secondly, the functional modules of the PHYTIUM computing unit are divided, and the design process of each module circuit is given, including the parameter calculation process. Thirdly, the PCB layout and routing rules of the computing unit are given. Finally, the feasibility and effectiveness of the design are proved through analysis and experimental verification.

The communication framework of surface unmanned craft is the core of supporting its corresponding navigation algorithm, guidance algorithm and motion control algorithm. In view of the complex component equipment and numerous sensor equipment of the IOT network of surface unmanned craft, MOOS- IvP of the client-server message publishing and subscription framework is used as the core of the overall communication framework. The overall communication framework involves motion controller, navigation and guidance controller, ship energy controller, handle remote controller, image algorithm controller, information query controller and so on. Compared with simple UDP communication, the communication framework with MOOS as the core is more stable and reliable, which can meet the needs of rapid response of unmanned craft. Compared with ROS communication framework, it is more convenient and has the advantages of easy transplantation. It is very convenient to establish the overall software framework of UI.

With the rapid development of 5G, autonomous driving has entered a new stage of development. But the location of the vehicle directly affects the safety of the vehicle. In order to reduce the intelligence degree of vehicles, increase the information sharing ability of roads, and improve the positioning accuracy of vehicles. This paper detects and tracks vehicles based on roadside cameras, and obtains vehicle location information through ranging model and coordinate transformation relationship. At the same time, a weight fusion localization algorithm is proposed by integrating GNSS position information of the vehicle. This method has important theoretical significance in reducing the cost of sensor for vehicle, and has achieved good positioning effect in real vehicle positioning experiment.

Smart home uses remote access and control technologies, but there is a risk that user identities are easily spoofed, user information is easily captured, and home gateways are vulnerable. Based on the blockchain, a smart home remote access and control scheme is proposed. The scheme ensures the secure access and control of the home device by the smart home user and the secure communication of the home gateway by combining the blockchain, group signature and message authentication code technology (MAC). The security analysis shows that the scheme has security features such as non-tamperable, anti-replay attack and anti-DDos attack. Compared with other schemes, the scheme has obvious advantages compared with other schemes.

In this work, an internet of things (IOT) system is designed for wearable health monitoring based on the Harmony operation system (OS). The BearPi-HM_Nano and Arduino Uno development boards are used as the main control units for collecting the signals from multiple sensors, reading the near field communication (NFC) card and excuting interaction through voice broadcasting and LEDs. The Harmony OS is run on the BearPi-HM_Nano boards to read the sensors via APIs and schedule multiple tasks. The WiFi transmission is adopted using the build-in WiFi modules on the BearPi-HM-Nano boards. The web A dedicated server is constructed for the IoT system and run remotely on the Alicloud computing platform. The web front end and wechat applet are desiged to interact with users. As for the functions, the system can be used for patients' remote monitoring of life indicators of themselves and their families, online diagnosis and treatment, intelligent generation of golden health code, disease prediction, generation of electronic health cases, etc.

More and more edge devices are linked with the network along with the digital transformation of society and the rapid development of the Internet of things. A large number of edge devices produce a large amount of data which challenges the IOT network with cloud computing as the core computing power. The edge computing can save network bandwidth and reduce delay as the extension and supplement of cloud computing on the one hand; on the other hand, the characteristics of multiple mobile devices based on edge computing make it very suitable to realize big data fusion with the Federated Learning framework, the privacy and security of users can be greatly improved by edge computing based on Federated Learning, the data silos can be broken and a more intelligent Internet of things can be achieved. This paper summarizes the common algorithms of Federated Learning based on edge computing, analyzes the existing challenges, and summarizes the corresponding algorithms.

On the basis of Petri net analysis of contact network system reliability, the state maintenance theory is applied to contact network system operation analysis, the reliability of contact network in each maintenance cycle is quantitatively calculated, the optimization model of contact network maintenance plan based on "reliability-maintenance cost" is established, and the multi-objective particle swarm optimization algorithm (MMOPSO) with multi-search strategy is used to optimize it. The Pareto optimal solution set is obtained by model validation in MATLAB. By making the contact network system run reliably and safely while reducing maintenance costs, it can provide reference and reference for the actual maintenance work.

In engineering practice, during the gradual development of many equipment from normal state to fault state, they have to experience the three-level working state of "safety state - potential failure - functional failure". From the external performance of potential failure, the first is that the symptom is weak, non-stationary, and does not cause substantial damage to the system. The detected data is zero-failure data, Therefore, the characteristic information of potential failure is relatively small, which belongs to typical small sample information; Secondly, due to the lack of accurate judgment and health analysis of equipment working state, the unnecessary maintenance may aggravate the evolution speed from potential failure to failure; Third, when the equipment operates under overload or harsh conditions, it will promote the deterioration of potential failure to a certain extent. Aiming at the difficulty of reliability evaluation of multi state system, a reliability method based on information fusion is proposed by making rational use of prior information, which provides a technical way for equipment reliability evaluation.

The condition of energized test devices directly affects the quality of power grid condition-based maintenance. At present, the condition of energized test devices is mainly judged by the qualified or unqualified test reports. In this paper, the condition quantities affecting devices quality are divided into four categories: test report, quality information, subjective evaluation and special inspection, and the quantitative scores are given. Taking UHF partial discharge detector and infrared thermal imager as examples, the special detection and evaluation are carried out, and the test result range is refined. The modified Wilson confidence interval ranking algorithm is introduced to realize the objective and fair ranking of energized test devices manufacturer quality.

In order to visualize the maintenance track route in non-gis platforms such as 3D rendering engines including three.js, UE4, U3D, the authors studied concepts of geodetic coordinate system, geocentric rectangular coordinate system and station center tangent plane coordinate system, figure out the method of converting geodetic coordinates into 3D rendering engine coordinates by linear algebra theory, and realized the visualization of the maintenance track route by the UE4 spline curve component and the material component.

Contemporarily, deep Learning has achieved remarkable achievement in the field of artificial intelligence technology. Comparing with traditional machine learning methods, deep learning creates its model by constructing the neural network. This investigation reviews the neural network's development history and describes classical neural network methods, e.g., convolutional neural networks and recurrent neural networks. Besides, the shortcomings and limitations that the neural network is currently facing, including aspects in accuracy, stability, and robustness, are discussed. Meanwhile, the solutions towards these limitations are also mentioned, e.g., capsule network and adversarial attack. These results shed light for the future developments of Neural Network.

The metaverse has become the latest hot spot of research on virtual space in the age of intelligence. In order to better measure the boundary of the metaverse, this paper proposes a measurement method based on big data and cognitive intelligence. Through the quantitative study of the three main characteristics of the metaverse, a method to measure the metaverse and the real world is given to provide a reference for scholars studying this field.

Object detection is an important task in computer vision. There are many practical applications using object detection based on deep learning nowadays. For deployment on FPGA with limited resource and operator support, object detection faces problems such as how to improve speed and reduce power consumption. YOLOX is a high-performance anchor-free YOLO version. To deploy YOLOX network on FPGA, we first replace the Focus layer of YOLOX, adjust the structure of the SPP layer, and change the activation function to meet the operator support constraints. Then we perform sparse training and use scaling factors of BN layer to select out the insignificant channels. The convolutional layer channels are pruned according to the degree of sparseness and pruning ratios. Finally, the network is quantified, compiled via the Vitis AI tool, and deployed on the Xilinx FPGA development board. Comparing the performance with different pruning ratios, the experiments demonstrate that the network runs significantly faster on the FPGA after pruning, and the power consumption is also reduced.

The current data scheduling retrieval model have the problem that the performance of sub flow path at non bottleneck is restrained, which affects the performance of the model. This paper constructs a visual data scheduling and retrieval model for customer service data analysis platform. Using microservice encapsulation, the overall architecture of the platform is designed, and the corresponding services are called to complete business processing. Considering the fairness of bottleneck, the queuing mechanism under path congestion is established, and the transmission path is allocated according to the number of data packets that need to skip the buffer, which is read and sent according to the path sequence number. Combined with round-trip delay and path congestion, a data scheduling retrieval model is established to schedule and retrieve the path with low congestion. The simulation results show that compared with the models based on service differentiation, scheduling priority allocation, cloud computing and evolutionary multi-objective optimization, the total amount of data transmission, throughput and bandwidth utilization of this model are improved. It has certain transmission advantages in case of network congestion, which is conducive to improving the stability of the platform.

In order to identify the risk of the letter and visit stability work, on the basis of summarizing the existing research methods, put forward a scheme for the monitoring and risk identification of the letter and visit information based on the big data of the Internet. Firstly, through real-time network crawler and other means to establish the labor relations field letter information monitoring data set, then based on data mining technology, relying on the network big data to carry out risk identification research on the letter and visit information, and finally based on the letter and visit information monitoring and risk identification results put forward a preventive mechanism.

The cybersecurity of electric vehcles(EVs) charging process has attracted more and more attention. In this paper, through the risk assessment of charging system under charging scenario, we clarify the cybersecurity risks of charging system of EVs, and propose systematic protection measures in multiple dimensions for the main cybersecurity risks in charging process. Further, we conduct experimental verification of EVs charging system based on the cybersecurity protection strategy designed in this paper, and prove the cybersecurity protection strategy can protect EVs charging system security from hardware cybersecurity, software cybersecurity, data cybersecurity and communication cybersecurity.

In the case of frequent extreme weather, in order to improve the accuracy of winter wheat yield forecasts. Based on the data from the Internet of Things, this paper takes Henan Province as the research area and calculates the influence of extreme temperatures on the growth of winter wheat, and takes the sum of growth degree-day (SGDD) and Sum of Extreme Degrees (SEDD)as two major factors; Based on remote sensing technology, this paper calculates the growth area of winter wheat. And the normalized vegetation index (NDVI) of each period; finally use the above three factors, to establish two regression models for winter wheat yield forecast. The model constructed by the two influencing factors of SEDD shows that NDVI is a sensitive factor for yield prediction. The results show that both models in northeastern Henan Province have achieved good results. The model can provide methods for winter wheat production management and decisionmaking in Henan Province support.

The development of big data technology has given birth to big data products based on mobile terminals--intelligent teaching tools. Intelligent teaching tools can establish an objective, diversified, and comprehensive evaluation system based on the analysis of users’ massive data in mobile terminals; they are available to quantitatively analyze the characteristics of users’ behaviors and establish intuitive digital chart models, thereby providing effective decision-making services for users. This paper takes Rain Classroom as an example to introduce the use of big data visualization technology in art teaching.

Oil pump plays an important role in the field of oil transportation, while high-speed rotating bearings, as important parts of oil pump, often suffer various forms of damage, which poses a certain threat to the safe supply of oil resources. Based on the experimental data of bearings from Case Western Reserve University, ten types of damage data were selected for fault identification and analysis. Considering the influence of complex working conditions of the station on the collected signals, Gaussian white noise was added to the experimental data to get close to the collected signals. Based on the energy characteristics of the data obtained by wavelet transform, the probabilistic neural network is used to classify the above ten kinds of feature data. The results show that the accuracy of the classification of the proposed model is 99.76%, which is much higher than the accuracy of the current common models. The research results provide a reference method for on-site fault identification of oil pump and have a certain engineering practical significance.

Merging roadways are one of the main reasons for reducing highway traffic safety and transportation efficiency. This paper proposes a multi-vehicle cooperative driving strategy for on-ramp merging in the environment of connected and automated vehicles (CAVs). Based on the principle of first-in-first-out (FIFO), a new algorithm is designed to optimize the merging sequence of CAVs. The velocity profile of each CAV is planned according to the merging sequence to make CAVs pass through the merging zone smoothly and safely. Compared with the FIFO strategy under different traffic demands, the effectiveness of the proposed strategy is validated.

With the development of information technology, digital sculpture model data in the context of the Internet is being integrated into our daily life through subtle network means such as big data, App and Internet of Things. Sculpture modeling data in the era of intelligent Internet is studied by investigating the shape data of sculpture model in ZBrush software. Researchers combine the development and innovation of data file STL in the era of intelligent Internet to study the interconnection between digital sculpture model data, Internet, human and big data in the era of intelligent Internet, and further analyse the transformation of human emotion.

In recent years, new energy and power businesses and new business formats have developed rapidly, and the energy and power industry is changing from traditional competition to competition among ecosystems. However, the current research on ecosystem construction technology pays more attention to business layout and lacks quantitative summary of the construction practice of advanced enterprises at home and abroad. Therefore, this paper proposes the overall construction technology of Energy Internet Ecosystem in new energy field by systematically summarizing main measures of seven excellent ecosystem, and puts forward that Power Grid enterprises should pay close attention to the changes of users’ needs, consider the overall design of the functional positioning of various platforms and orderly promote the linkage development of businesses in new energy field.

At the present stage, people's requirements for artificial intelligence are gradually improving, providing a guarantee and foundation for the development of the Internet of Things. The Internet of Things technology has penetrated into all aspects of people's production, life, work and learning. The realization of the Internet of Things brings a large number of goods into the Internet, which goes beyond the restrictions of time and space, and uses the network to supervise and control the accessed goods, which provides a lot of convenience in people's lives. In this paper, the classification characteristics and the application of the Internet of Things technology are discussed in detail, and strive to integrate the matters related to the use and development of the Internet of Things, so as to lay the foundation and guarantee for the subsequent development of the Internet of Things technology.

The quality of sleep has become a matter of concern to modern people. The medical field usually uses Electroencephalogram to detect the quality of sleep. The method in machine learning can figure out people's sleep quality by researching Electroencephalogram. However, this method still has its drawbacks, such as being restricted by data. In order to study the relationship between data characteristics and model performance, this paper explored the impact of standardization and data volume on accuracy based on Support Vector Machine. For these investigations, we set up three types of control groups. The first group compares standardized and non-standardized data. The second group compares segmentation at different time intervals, and the third group compares different data volumes. The experiment results show that the accuracy of the standardized data shows a specific upward trend under different segmentation methods. In contrast, the non-standardized data has no apparent phenomenon. Also, in the case of the same amount of data for each interval, taking 25 seconds as an example, the overall accuracy of the sampling situation is lower than that of the entire situation. However, the accuracy of the third class of data is increased. When the same data set uses different amounts of data to experiment, there are no significant test results. In other words, this experiment found that the size of the data and the way the data is processed have affected the accuracy of the results to a certain extent.

With the rapid development of precision agriculture and smart agriculture, the need to build an automatic identification and detection system for diseases and insect pests is increasing. Using computers to correctly label plant diseases and insect pests is an important prerequisite for achieving accurate classification of plant diseases and insect pests and ensuring system performance. In order to improve the accuracy of computer classification of plant pests and diseases, this paper proposes an automatic pest identification method based on the Vision Transformer (ViT). In order to avoid training overfitting, the plant diseases and insect pests data sets are enhanced by methods such as Histogram Equalization, Laplacian, Gamma Transformation, CLAHE, Retinex-SSR, and Retinex-MSR. Then use the enhanced data set to train the constructed ViT neural network, so as to realize the automatic classification of plant diseases and insect pests. The simulation results show that the constructed ViT network has a test recognition accuracy rate of 96.71% on the plant disease and insect pest public data set Plant_Village, which is about 1.00% higher than the Plant disease and pest identification method based on traditional convolutional neural networks such as GoogleNet and EfficentNetV2.

Traditional convolutional neural networks usually consider both resource budget and operation cost. However, if there is a better operating environment and more advanced resources available, a more accurate and faster algorithm to solve some problems of existing networks can be provided. In this paper, the author thoroughly studied the Self Attention mechanism and EfficientNet network that Google has released and combined the advantages of the two networks. Based on this result, the author proposes an improved algorithm of EfficientNet with Self Attention mechanism. Before introducing EfficientNet pre-training model to start training, EfficientNet uses simple but efficient composite coefficients to uniformly scale all dimensions of depth, width and resolution. A simple Self Attention network structure is trained in advance to process data images. This will enable the pre-training network to focus as much as possible on the data and image features with a large amount of information and large gap before entering EfficientNet training. After that, the pre-training network will train the model. Through experiments, it is found that the model has higher quality. The improved algorithm showed extremely high accuracy and fast convergence speed on CIFAR-100, OxFlowers and ImageNet, and three other transfer learning datasets. The accuracy of the improved algorithm network with efficientnet-b4 as the pre-training model on the three data sets is greater than 95%. Compared with the original EfficientNet, the comprehensive improvement is more than 10%, and the running speed is increased by about 30%. Compared with the classic convnet, it can be increased by 20% and the running speed is increased by more than 50%.

AI-assisted grading system saves great amount of time for teachers. Current application focus on grading techniques for Chinese or English composition, where high frequency features contains major feature for the output. In contrast, grading math problems requires sophisticated precision and rigorous logic. This paper illustrated an AI-assisted grading system for math questions. This system learns the mapping from students’ free response in math exams to output score, requiring little effort from teachers. The system extracts equations and reasoning and compares them to output a final score. We proposed a new network, CycleLatex, as an efficient solution to image-to-latex translation problems. It is capable of unsupervised learning and has shown state-of-art performance on handwriting and multi-line equations. The system also implements a parser which is capable of comparing most equations and reasoning at a reasonable speed.

With the booming of the insurance industry, the number of insurance frauds has increased gradually, among which the fraud in the auto insurance industry is the most serious. In order to improve the BP neural network prediction is easy to form local minimal value and not the global optimum, convergence speed is slow and the prediction accuracy is low, etc., we propose an improved particle swarm optimization (PSO) algorithm based on the dynamic linear decreasing adjustment of the weights in the particle swarm algorithm, and then the BP neural network weights and thresholds are optimized to establish the auto insurance fraud prediction based on the improved particle swarm optimization BP neural network model. The 400 sets of sample data of 13 main influencing variables of auto insurance historical claims of an insurance company were selected for PSO-BP and BP neural network model training, and 50 sample data were used for the prediction accuracy evaluation of the optimized model. The results show that based on the improved PSO-BP neural network algorithm its mean square error between the predicted and true values is significantly reduced and the prediction accuracy is greatly improved, which can effectively predict auto insurance fraud.

At present, text similarity calculation technology is widely used in text data mining, text classification, information retrieval, information filtering, machine translation, text checking and other fields, but it is rarely used in project association. Among them, the cluster analysis of electrical projects is a cluster correlation analysis based on the text similarity of the electrical projects in the database. This study studies the related technology of text similarity calculation, and focuses on the problem that the traditional vector space model cannot reflect the special text performance ability of feature projects in different positions in the text similarity calculation, and studies its improved model: text segment vector space model, the calculation efficiency of text similarity of electrical projects is improved, and problems such as duplication of construction are effectively avoided.

The k-means algorithm has been widely used since it was proposed, but the standard k-means algorithm does not perform well in terms of efficiency when dealing with large-scale data. To solve this problem, in this paper, we propose a fast kmeans algorithm based on multiple granularities. First, from the coarse-grained perspective, we use the clustering distribution information to narrow the search range of sample points, which makes the proposed algorithm very advantageous on large k. Second, from the fine-grained perspective, we use the rules of upper and lower bounds to reduce the number of sample points involved in the distance calculation, thus reducing many unnecessary distance calculations. Finally, we evaluate the proposed k-means algorithm on several real-world datasets, and the experimental results show that the proposed algorithm converges hundreds of times faster than standard k-means on average with the accuracy loss controlled at about three percent, and the speedup of the algorithm is more obvious when the dataset size is larger and the dimensionality of the dataset is higher.

K-means algorithm, also known as k-means clustering algorithm, is a clustering algorithm commonly used in image analysis and color extraction. The color of Yi costumes deeply reflects the traditional culture of Yi nationality. In view of the problems that designers often extract and apply the traditional color of Yi costumes through personal experience, it is difficult to truly restore the color image and characteristics, K-means clustering algorithm is applied to the color research of Yi costumes. Based on K-means algorithm, this paper extracts the color of traditional Yi clothing, clusters the color, analyzes the color matching of Yi clothing, and tests the possibility of this method in the color research and application practice of traditional Yi clothing combined with design examples. The results show that k-means algorithm can highly restore the color intention and characteristics of Yi clothing, and can be used as a new method for the innovative application of Yi clothing color.

Repeated hospitalizations have brought a heavy economic burden to the society. Exploring the establishment of a dynamic prediction model for readmission is helpful for early intervention. Based on the machine learning algorithm, a dynamic prediction model of diabetes readmission combined with the A-RES algorithm is proposed. SMOTE oversampling for data imbalance use, encoding processing and One-Hot encoding processing for medical texts are used to train prediction models of random forest, GBDT, and XGBOOST, and the A-RES algorithm is used to combine them for training. The learning results show that the GBDT+RES model has the best performance, with an AUC score of 0.973 points, which is an increase of 0.12 points compared with the original model. The final prediction models all have good model performance.

Network intrusion detection system (NIDS) is a tool that can detect various network attacks by analyzing network traffic. In recent years, traditional machine learning and deep learning methods have been widely used in NIDS. And these works achieved high detection rate and low false positive rate when identifying common attacks (such as DoS). However, the existing models performed poor in detecting rare or unseen network attacks. In this paper, we design a novel network intrusion detection model based on two-phase detection and manually labeling. This two-phase detection model (TPDM) is a multi-classification system, and it can classify network traffic into benign or specific attack type. TPDM gains the overall accuracy 98.35%, and average F1-score 99.23% on UNSW-NB15 dataset. The experiment shows that TPDM performs better than state of the art in detecting rare attacks with few training samples. Besides, it can detect unknown network attacks and then label these attacks.

UHV converter transformer has complex insulation structure and huge volume, and 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 highend 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 in the outgoing line device area is constructed according to the electrical, thermal and mechanical multi physical field simulation model. In fact, it bears the on-line voltage waveform and current waveform, which are monitored by optical sensors, and realizes the interactive linkage between the on-site operation parameters and the loading data of the digital twin model. 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. 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 the converter transformer, especially for the structure design, operation and maintenance of outgoing device area.

Deep learning has a wide range of applications and far-reaching influence in our production and life, but the training of models requires a lot of data. In the real world, data acquisition requires a lot of costs, and the distribution of data is often uneven, with a small number of categories occupying a large number of samples, making the overall data present a long-tailed distribution. This makes convolutional neural network often perform poorly when training data are heavily class-imbalanced. In this work, enhance the feature extraction capabilities of the base model by add attention mechanism, and use the regularization technology mix-up algorithm to enhance the long-tail data. compared several state-of-the-arts techniques on the benchmark datasets imbalanced CIFAR10 and CIFAR100, that our method provides consistent and significant improvements over previous models.

Developing and investigating recently, a new type of model appeared to greatly help the human with generating models. It can be trained with both supervised or unsupervised learning and contain both generative and discriminative models. Style transfer is one of the functions the GAN can be trained to produce, that it can synthesis two images together to get a new result by having one of the pictures as subject and the other one as style. In this paper, the work will introduce GAN and style transfer in detail and the application of style transfer in the real world.

Face morphing attack has become a severe threat to the current face recognition systems. Though there are some methods for detecting face morphing, the performance of these methods is susceptible to noise. Aiming to enhance the performance of resisting noise in face morphing detection, a noise robust convolutional neural network is proposed in this paper. The structure of the network is divided into two parts: facial image adaptive denoising and face morphing detection. Before the face morphing detection, the auto-encoders are first utilized to adaptively denoise the noised facial images, which can effectively reduce the influence of noise on face morphing detection. Then, the pre-trained VGG19 convolution neural network with powerful classification ability is used for face morphing detection with the generated noise-free facial images. Experimental results indicate that the proposed method can effectively reduce the noise influence on face morphing detection, and can achieve better performance compared with some existing methods.

With the development of Internet big data, how people obtain a large amount of news text information and automatically obtain the information they want from the text information is an urgent task. In order to structure and analyze a large amount of Chinese text information on the Internet, this paper proposes an entity extraction method based on the BERT pre-training model and BiLSTM with the Attention Mechanism. Aiming at the problem that the BiLSTM model can only obtain feature information at the sentence context level, but cannot obtain local feature information. In this paper, based on the BiLSTM model, a BERT feature extraction model is added to obtain word vectors containing contextual semantic information, thereby capturing global and local information. At the same time, an Attention Mechanism is added to improve the effect of the model. The model was trained on the 2018 Football World Cup dataset corpus, and it was verified that the precision, F1 value and recall rate of the model have significantly improved performance on the dataset.

Shear wave (S-wave) velocity prediction is important for the evaluation of shale oil and gas reservoir. However, there are some problems with traditional models: the parameters of the petrophysical model are relatively fixed, and the machine learning models do not consider the sequence information of the log data. Therefore, the S-wave velocity prediction model based on Temporal Convolutional Network (TCN) for shale reservoir is proposed. The model can flexibly extract the sequence features by adopting causal convolution and dilation factors and mine the inner relationship between the well logs and the reservoir S-wave velocity to achieve a better prediction performance. Two wells of MY1 and FN4 in shale reservoir in the Permian Fengcheng Formation in Mahu Sag of Junggar Basin, Xinjiang Oilfield are taken as an example. The TCN model achieves optimal results on both MY1 and FN4 with mean relative error (MRE) of 0.84% and 1.39%, respectively, when compared with the results of traditional petrophysical models, machine learning models and conventional deep learning models. This indicates that the TCN model has strong effectiveness and generalization in Swave velocity prediction, which provides a new idea for S-wave velocity prediction in shale reservoir.

With the continuous development of the Internet, more and more information enter people's lives. However, the information is mixed. It is difficult to guarantee the correctness of the text. For errors caused by homophone replacement, an automatic Chinese text local error detection and correction solution based on n-gram model is proposed. A method of local error detection based on the combined model of 2-gram and 3-gram is proposed, and a method of local error correction based on 3-gram model is proposed. Experiments show that the error detection recall rate is 83.1%, the error detection accuracy rate is 41.5%, the F-score is 55.4%; the error correction rate is 78.1%. The method is compared with the 2-gram model and the 3-gram model. The accuracy of error detection is increased by 7.2% and 8.2% respectively. The F-score is increased by 6.3% and 8.2% respectively.

Gesture is a form of non-verbal communication and has many applications, such as sign language communication between deaf and dumb people, robot control, human-computer interaction and medical applications. The commonly used acquisition equipment in gesture recognition is the visible light camera, but illumination has a great impact on the accuracy of the collected data classification processing. The whole project designed a complete end-to-end edge computing system design and deployment, the system can achieve from gesture image acquisition to gesture recognition. A dataset of 3600 thermal images was created, and each gesture had 1200 thermal images with only 4*4 resolution. These images were upsampled by bilinear interpolation and fed into a new lightweight deep learning model combining deep residual learning with ShuffleNet V2 for gesture classification. The system achieved 98.63% accuracy on the test data set. Another advantage is that it is based on thermal imaging, so the accuracy is not affected by background lighting conditions.

By reasoning and analyzing the ranging principle of AN/MPQ-53 radar, the factors affecting the ranging accuracy of AN/MPQ-53 radar are analyzed, the noise error, multi-path error, quantization error, sampling time error, and delay error caused by Doppler frequency shift are distinguished, and the ranging error of AN/MPQ-53 radar is theoretically deduced and analyzed. Considering the irrelevance of error factors, the cumulative ranging error of AN/MPQ-53 radar is obtained. The research results further refine the data structure and model connotation of AN / MPQ-53 radar, which have certain theoretical and practical reference value for the research of electronic countermeasure of AN / MPQ-53 radar and the development of Agent simulation agent.

Ontology is effectively formal, clear and detailed specifications in the form of concepts and relations of a shared conceptualization to a special domain. Ontology construction methods can be classified into manual construction and (semi-)automatic construction. However, manual construction method is usually expensive due to the considerable amount of human efforts it may involve. Therefore, automatic and semi-automatic ontology construction has been a research hotspot in the past decade. A new trend of these approaches is relying on machine learning and automatic language processing technology to extract concepts and ontology relationships from structured or unstructured data (such as database and text). The aim of this paper is to introduce some recent representative technical researches on ontology construction using deep learning model from text.

With the development of society, environmental problems are becoming more and more important. We should develop the environment while developing the economy. People's awareness of environmental protection should also continue to improve, and begin to pay more attention to ecology. Cities all over the world should take the road of ecological development and build ecological cities. The construction of ecological city is the symbol of human civilization and the inevitable trend of urban development. This leads to a very important concept in eco-city, that is the plant wall. Based on this, the author designed a simple intelligent planting system based on SCM. The system uses MCU "Arduino Uno R3" as the main control chip. The photosensitive resistance sensor and humidity sensor are used to measure the current information and feed back to the MCU. LED lights and feeders work to realize automatic watering, lighting and feeding fish on time. Finally, the entire system can be controlled through a wireless connection to the phone. In this paper, the hardware and software parts of the system are mainly designed, and the parameters of the system are tested, which basically realize the function and improve the intelligent degree of plant planting.

Aiming at the problems of low efficiency and poor accuracy in manual identification of tens of thousands of substation inspection images in the power grid, one method and system for identifying defects in substation inspection image data based on Faster R-CNN is proposed. The inspection images are preprocessed through data analysis module, image annotation module, image cleaning module, firstly. A convolutional neural network (CNN) is used to perform feature extraction on the processed images to obtain the target features, secondly. A Region Proposal Network (RPN) is introduced, which generates multiple candidate anchor frames on the target feature images, thirdly. Region of Interest (ROI) pooling is performed on the multiple candidate anchor frames to obtain a feature matrix of a fixed size, fourthly. Then, the frame regression and classification recognition are applied to obtain the defect recognition result and target anchor frame position. Finally, high-precision identification of 25 types of defects of substation inspection images such as oil leakage, insulator damage, mark damage, metal corrosion， is realized. The average accuracy of the method is as high as 93.44%, which is about 27% and 23% higher than the traditional systems with R-CNN and Fast R-CNN algorithms. The time of single image recognition is shortened to (150ms) millisecond level. This method greatly saves the detection time, which can effectively reduce manpower input, improve work efficiency more effectively, and provide a foundation for the digital transformation of the power grid.

In order to solve the problems of low correct recognition rate, high false recognition rate, long recognition time and poor image recognition effect in traditional methods, a multi-type fruit picking image recognition method based on deep learning is proposed. SVM is used to classify the image, the segmentation method based on statistical pattern recognition is used to segment the image, and the image is denoised according to mathematical morphology. According to the results of recognition, segmentation and denoising, the deep convolution neural network in deep learning technology is used to recognize many kinds of fruit picking images. The experimental results show that the error recognition rate of this method is low, the correct recognition rate is high, the recognition time is short, and the recognition effect is good, which fully verifies the application value of this method.

The high-precision classification of built-up areas was extracted by the management departments with the development of social economy. Traditional classification methods cannot discriminate the types of built-up areas very well. The paper proposed a new method for classification of built-up areas with deep learning technique, namely VGG-DeepLab-UNet, which is based on the UNet network. The feature generation part of VGGNet was selected as the encoder of UNet and the upsampling part of DeepLab network was selected as the decoder of UNet. In this paper, Xinjiang was selected as the experimental area, and the new model was used to extract the classification information of built-up areas. The experimental results showed that VGG-Deep-UNet can be used to extract the classification information of built-up areas with high precision. Compared with the manual annotation results, it is pointed out that the classification information of built-up areas can be extracted with higher precision with deep learning technique, which can effectively reduce manual operation and improve the efficiency of extraction.

The word vector representation technology based on neural language model can automatically learn the effective feature representation of words from large-scale unlabeled text data sets, and has made important progress in many natural language processing tasks and research. While achieving the above classification performance improvement, the training time of EMCNN relative to MCNN is reduced by 36% in subjective and objective classification, and in emotion 7 classification A reduction of 33.82%.