Abstract:At present, the research on video semantic segmentation is mainly divided into two aspects. The first one is how to improve the accuracy of image segmentation by using timing information between video frames, while the second one is how to use the similarity between the frames to determine the key frame, reduce the amount of calculation, and improve the running speed of the model. In terms of improving segmentation accuracy, new modules are generally designed and combined with existing CNNs. In terms of reducing computation load, the low-level feature correlation of frame sequence is used to select the key frame, which reduces computation load and operation time at the same time. Firstly, this paper introduces the development background and operation datasets Cityscapes and CamVid of video semantic segmentation. Secondly, the existing video semantic segmentation methods are introduced. Finally, it summarizes the current development of video semantic segmentation, and gives some prospects and suggestions for future development.

Abstract:In order to realize rapid discovery and scientific rescue of life after earthquake disaster, multi-source data is needed. Remote sensing data, scene environment data, distribution data of dangerous facilities, life signs data acquired by life detection equipment, status data of life detection equipment, and historical data constitute multi-source heterogeneous data sets in emergency rescue scenarios. Aiming at the application requirements of unified data supervision and multi-dimensional analysis in the field of emergency rescue, this work deeply studies the three-dimensional spatial fusion analysis technology of multi-source heterogeneous data, proposes a plug-in-free three-dimensional spatial fusion scheme based on WebGL rendering technology, and develops a visualization system of emergency rescue data fusion. Fusion and expression of multi-source heterogeneous data needed in emergency rescue scenarios can be realized under the unified space-time framework. It can assist operators to conduct joint dynamic potential analysis, help commanders to make command decisions, and greatly improve the efficiency of emergency rescue field applications.

Abstract:Internet identifier is the key resource for identifying and managing goods, information and machines in the Internet. It is an important basis for the operation and development of the Internet. In the process of correspondingly identifying personal information and basic resources, there are also major security risks of personal information leakage. At present, the introduction of privacy protection laws such as the GDPR and the Cyber Security Law of the People’s Republic of China imposes stricter requirements on the protection of personal information. This study first analyzes the privacy risk of the Internet identifier and discusses the challenges of personal privacy protection; then uses the typical Internet identifier-domain name as an example to deeply analyze the privacy risk in the Internet identifier service process. Combining the different privacy protection needs of the data life cycle, the Internet identifier privacy protection technology framework for the publication, storage, mining and use is proposed. Finally, specific privacy protection technology solutions are proposed for each kinds of Internet identifier data life cycle.

Abstract:To solve the problem of missing structural information and other meta-path semantic information in heterogeneous network representation based on single meta-path, this study proposes a representation learning method of heterogeneous network based on fusion meta-path weight. This method learns from the set of meta-paths in heterogeneous information networks, and then the low-dimensional representations of different meta-paths are fused with appropriate weights. The representation of heterogeneous networks with semantic information of different meta-paths are obtained. Experiments show that the heterogeneous network representation learning based on fusion meta-path weights has sound representation learning ability and can be effectively applied to data mining.

Abstract:In view of the traditional BP neural network model used in Photo Voltaic (PV) power generation prediction, there are shortcomings of low prediction accuracy and slow convergence speed. In this study, an improved BP neural network model is proposed to improve the defects of traditional BP model by adding momentum term and adaptively selecting the best hidden layer. Firstly, the correlation of meteorological factors for PV power output is analyzed, and six meteorological factors that can affect the PV power are extracted as input of the network model. Then, an improved BP network model is established, combined with the historical data of PV power output, to directly predict the generation of data. Finally, according to the prediction results under different climate types, the feasibility and effectiveness of the model for power prediction are analyzed and verified.

Abstract:Gas load forecasting is affected by various complex factors such as social economy, weather factors, date types, and the combination of multiple factors, and it will inevitably lead to a large randomness and a certain degree of complexity in the trend of gas load sequence changes. In order to effectively improve the accuracy of gas load forecasting, a new integrated deep learning algorithms is proposed to predict the gas load in multiple steps. Firstly, the non-stationary nonlinear load sequence is decomposed into several steady-state and linear IMF components and residuals by the set of EEMD algorithm, which effectively avoids the modal aliasing problem caused by the traditional EMD. Then, each subsequence obtained by EEMD decomposition is composed of a training matrix different from the feature sequence extracted by AutoEncoder. After that, each subsequence obtained by EEMD decomposition is composed of a training matrix different from the feature sequence extracted by AutoEncoder. Finally, the corresponding Long Short Term Memory (LSTM) prediction model is established for the training matrices corresponding to different subsequences, and the component prediction values are reconstructed to obtain the final prediction result. In order to verify the effectiveness and prediction performance of the proposed algorithm, the Shanghai gas data was used to simulate the above model. The results show that the prediction accuracy is significantly improved compared with the comparison method.

Abstract:With the continuous operation of high performance computing environment, computing requirements of accounts increased rapidly and demand for resources continually rise. The environment needs to extend resources to supply more powerful computing power and meet needs of different fields. Meanwhile, expansibility, easiness, and reliability also need to be improved. Platform for high performance computing environment is designed according to requirements of accounts to achieve the goal of supplying quality service. The study extends the following content around the platform for high performance computing environment, highlighting structure of the platform, related important technologies, main functions, and implementation process. Through deploying and testing, results show that the platform can meet requirements of existing accounts. In the future, it will extend user group of high performance computing environment and promote the development of high performance computing in more areas.

Abstract:In response to the characteristics of multi-source, heterogeneous and massive data in meteorological research activities, the China Meteorological Administration has issued a unified data interface service based on China Integrated Meteorological Information Sharing System to provide data support for meteorological activities, there are certain technical barriers to data access in practice. In order to eliminate this technical barrier, the meteorological data visualization scheduling retrieval system developed based on hybrid architecture combines the task scheduling model, visual interaction and data format adaptation technology, and allows researchers to quickly access data in a visual and zero-programming manner. It meets individual needs for data formats. The application proves that the implementation of the model improves the data organization efficiency of the meteorological research business, assists the users with weak programming ability to use the data access interface, the ideas about the scheduling model and the business of data sharing can be used for reference in other fields.

Abstract:In order to improve the intelligence of smart city lighting monitoring system, the association rules are used to mine the intrinsic connection among control commands. Considering that the traditional Apriori algorithm uses a self-connected method to generate a large number of candidate sets, the execution efficiency is low, so the improved Apriori algorithm is adopted. The frequent set is generated by traversing the weighted graph form, the algorithm execution efficiency is improved, and the control mode association rule is generated. The application of the rule and the control module of the smart city lighting monitoring system realizes the intelligent association of the control mode, reduces the dependence of the system on the management personnel experience, improves the work efficiency, and has certain social application and promotion value.

Abstract:Equipment failures can cause serious production accidents and pose a serious threat to business, society, and personal safety. Therefore, it is important to analyze the state of the IoT device and reasonablely process. Aiming at the large and complex data of IoT devices, this study proposes a massive data processing architecture for IoT devices. At the same time, combined with Dask distributed computing framework, a distributed device state analysis and processing system for IoT based on Hadoop environment is designed. The system mainly includes three modules of data service, data analysis, and data storage, and through reasonable node scheduling scheme, the efficient operation of the algorithm and the stability of distributed computing are guaranteed. The system operation shows that it can effectively process large quantities of data and accurately predict the status of the equipment in real time to meet the practical application in the industrial intelligent manufacturing process.

Abstract:College students’ extracurricular activities are rich and varied, with the increasing number of activities day by day, the activities management is cumbersome and the task is challenging, but there is still lack of one-stop solution. To meet the needs of college students to build an activity management platform, we build user front-end of activity management platform based on Wechat mini program, back end of activity management platform based on Node.js, express, and MongoDB framework. Therefore, activities and user management are realized integrally. Its features include mailbox verification, two-dimensional code check-in, reading and writing Excel, etc. Encrypted front-end and back-end communication is realized by HTTPS RESTful API, perfectly meets the needs of students and ensures the safety of students' information, which provides convenience for the management of college students’ activities.

Abstract:In order to warn and prevent haze effectively, an regional haze warning system for heterogeneous networks is designed in this study. The system consists of collection module, coordinator, gateway, server, and display terminal. It collects sensor data by deploying three kinds of heterogeneous networks, i.e., LoRa, NB-IoT, and ZigBee. The gateway is designed to achieve data uploading. As the brain of the whole system, the server achieves the classification pre-warning by providing analysis and processing of the environmental information, triggers the corresponding mechanism response, and takes corresponding measures. Firstly, this paper introduces the overall design of the system and the working mechanism of the classification pre-warning. Then, the classification pre-warning mechanism is improved by the support vector regression model which is optimized by genetic algorithm. Finally, the system is proved to forecast haze stably and effectively by analysis of the case.

Abstract:This study develops 3D visualization design verification platform by virtual reality technology, to optimize the design, intuitively analyze of product manufacturability, accessibility, disassembly and maintenance, provides a visualized platform for construction, commissioning, operation and maintenance to rehearsal and feedback. The study focuses on the selection method of the virtual reality software, and solves the problem of interface transformation between virtual reality software and 3D design software.

Abstract:Now, space field lacks 3D simulation softwares that are verified and can be used on all-domestic platform reliably. We introduce some design strategies, such as differentially loading Earth image and elevation data, optimized build of scene graph, filtering of trajectory data, and so on. The proposed software can solve the reliability and rendering efficiency problems in 3D simulation on all-domestic platform. Some simulation experiments have been conducted to valid reliability and rendering efficiency.

Abstract:The campus virtual roaming system is the important content of college information system development strategy, but universities do not widely use it currently. By reducing the cost and technical difficulty, taking a university as the virtual space, the system based on Minecraft platform is constructed by using SketchUp, MCEdit, and WorldPainter to build the scene, and taking command block instruction and Game Mod to implement the function of interaction. Through the upload the map to the server, the system is established. This system has perfect scene and strong dynamic interaction, which is convenient for users to develop and access the system. Experiment proves that Minecraft is a very useful tool for designing a virtual platform.

Abstract:In view of the waste of energy in using air-conditioner, a smart air-conditioner control system was put forward and implemented in this study. This system, which is based on OneNET, monitors temperature and humiture in real time and controlls the air-conditioner remotely by a smart air-conditioner control device module. It includes mobile phone app, virtual devices on OneNET, and a smart air-conditioner control device module, which can be applied to various brands of air-conditioners. The smart air-conditioner control device module based on STM32 collects temperature and humiture in real time and transmits data to OneNET. Users could check data by mobile phone app and transmits instructions to the smart air-conditioner control device module by OneNET. The smart air-conditioner control device module communicates with air-conditioner by infrared. The results show that this system could monitor temperature and humiture in real time and air-conditioner could be used economically, which creates a comfortable living and working environment.

Abstract:Motor is the most widely used power equipment in the industrial field and fusion field, and it is also an important key equipment in the Industrial Internet Of Things (IIOT). The traditional way of status monitoring is to monitor a single motor, and the collected data is not stored effectively, so that the data cannot be analyzed and processed and the value of the data is not fully utilized. With the development of the technology of cloud computing and IIOT, the manufacturing process has many advantages, such as massive data storage, remote control, and so on. At present, professional cloud services have reduced the threshold of the traditional industry with cloud service and improved the safety and diversity of cloud services, which provide effective technical guarantee for the product of related intelligent motor enterprises. Web technology and real-time data from acquisition system are used to realize real-time status monitoring for motor in this study. The users can be informed the status of motor, and analyze the collected data at any time when they login the system. It is necessary to prevent or prejudge the fault of motor equipment in advance, and avoid major accidents as much as possible. It provides valuable experience and basis for routine maintenance and management of production equipment, and eliminates problems such as “over maintenance” or “under maintenance” during traditional periodic maintenance. The system has been developed based on Apache/MySQL/PHP framework, cloud platform has been used to store the data collected. The design details will be given in the paper.

Abstract:In the production and life of human beings, water resources are essential substances. Nowadays, environmental issues have received much attention. People are particularly concerned about water quality safety issues. The safety and quality of domestic water has also received wide attention from all walks of life. In order to make people more intuitive and easy to know the water quality status of domestic water, this study provides a household water quality detection system based on narrow band Internet of Things technology, and gives the overall design scheme, hardware design, software design, and test results of the system. This system can detect the pH value, temperature value, and Total Dissolved Solids (TDS value) of household water in real time, and synchronize to the cloud platform through NB-IoT communication technology in real time. The system has clear functions, simple operation, and easy observation. It has a good user experience and commercial application value.

Abstract:This study mainly focuses on the traditional Itti visual attention calculation model, introduces edge feature information, and optimizes the visual attention calculation model. In the process of introducing edge feature information, this study improves the Canny edge detection algorithm: replace the Gaussian filter with an improved bilateral filtering algorithm to better maintain the edge, use the Sobel operator to calculate the gradient magnitude from four directions instead of two directions in the original method, the improved OTSU algorithm is used to select the double threshold instead of manually setting the threshold, thus reducing false detection and missed detection during image segmentation. In the experiment, we found that the proposed method is much better than the Itti visual attention calculation model based on the ordinary Canny algorithm.

Abstract:Accurate Time Delay Estimation (TDE) is a prerequisite for sound source localization technology based on Time Difference Of Arrival (TDOA). Compared with other TDE algorithms, the Generalized Cross Correlation (GCC) algorithm has been widely used due to its low computational complexity and implementation simplicity. GCC time delay estimation algorithm uses different weighting functions to suppress noise and other interference in different noise conditions. This work first presents microphone array model and GCC algorithm. Then, an improved algorithm proposed in the study is elaborated in view of the disadvantages of GCC algorithm. Furthermore, GCC algorithm to partial weighting functions is simulated using MATLAB under different Signal-to-Noise Ratio (SNR) conditions. The advantages and disadvantages of these weighting functions are analyzed by comparing the performance of time delay estimation and the accuracy of sound source localization.

Abstract:The Chimera grid methods have been widely used in the computation of flow over complex configurations or unsteady moving boundary process. In this study, the Cartesian auxiliary grid is introduced into the multi-block structured grid system. The background mesh is used to replace the spatial structural grid, and the mixed mesh system of the Cartesian auxiliary mesh and the multi-block structured mesh is established. By means of background grid, the occupation of computing resources is reduced, and the fast establishment of the boundary and interpolation relationship between meshes are realized. On this basis, the parallel overlapping process of hybrid mesh is realized. Through the definition and application of overlapping weights of overlapping grids, the parallel distribution pattern and load balancing model of hybrid grids are established, which can effectively reduce the communication of overlapping interpolation information among different processes, and realize the uniform distribution of computing and communication load among processors.

Abstract:The dynamic characteristics of Internet of Things related equipment require that the routing protocol not only possess the characteristics of low energy consumption, but also adapt to the characteristics of dynamic network environment. In this study, a dynamic routing algorithm based on LF-GFG is proposed based on the right hand rule of traditional location routing protocol and the easy realization of virtual coordinate system. The results of the NS-2 experiment show that the algorithm reduces the maintenance cost, improves the processing speed, and maintains a higher fault tolerance than other routing protocols.

Abstract:Football video lasts for a long time, and many video content is not the interest of audience. Therefore, football video scene classification has become an important research topic in recent decades, and many machine learning methods have also been applied to this topic. In this study, a soccer video scene classification algorithm based on 3D (three-dimensional) convolution neural network is proposed. The 3D convolution is applied to the field of soccer video, and the feasibility of this algorithm is verified by experiments. The flow of this experiment is as follows. Firstly, football video scene switching is detected based on frame difference method and logo detection method, and shot segmentation is realized. On this basis, the semantic features of shot segmentation are extracted and tagged, and then football events are classified by C3D. In this study, football videos are divided into seven categories: long shot, medium shot, close-up shot, playback shot, audience shot, opening shot, and VAR (Video Assistant Referee) shot. The experimental results show that the classification accuracy of the model is 96% on football video datasets.

Abstract:In the shallow seismic source location, the positioning model is the key to achieve high-precision positioning. However, due to the complex structure of the shallow underground medium, the extraction of characteristic parameters is difficult, and the number of sources is small, and the single-shot vibration data is limited, resulting in the traditional travel time positioning model is not accurate in the shallow microseismic positioning area. Aiming at the above problems, based on the travel time positioning model, combined with deep polarization information, and improved the traditional particle swarm optimization algorithm, this study proposes a high-precision source localization method based on its fast convergence speed and high positioning accuracy. The experimental simulation results show that the population optimization and cross-mutation PSO algorithm can effectively reduce the risk of the algorithm falling into the local extremum when solving the hybrid positioning model, and verify the accuracy of the algorithm, which can effectively improve the microseismic positioning.

Abstract:At the present stage, fires occur frequently, and fire detection and identification are required automatically. Although there are fire detection methods such as temperature and smoke sensors, the real-time detection is not guaranteed. To solve this problem, a method based on improved YOLOv3 fire detection and identification is proposed. Firstly, a multi-scenario large-scale fire target detection database was constructed to mark the categories and locations of the flame and smoke areas, and the problem of insufficient performance of YOLOv3 small target recognition was improved. Combined with the feature extraction ability of deep network, the fire detection and recognition were formalized into multi-class recognition and coordinate regression problems. The detection and recognition models of flame and smoke were obtained under different scenarios. Experiments show that the improved YOLOv3 algorithm proposed in this study can achieve ideal results for flame and smoke detection under different shooting angles and different illumination conditions, and also meets the real-time detection requirements in terms of detection speed.

Abstract:In order to solve the problem of urban traffic congestion, the state advocates travelling by public transportation, and the use of bus smart cards has become more common. At present, for the data generated by intercity bus smart card travel, there is very little research on the time between bus stations. Therefore, the power-law distribution analysis of running time between bus stations based on machine learning technology is proposed. The station algorithm is used to divide the bus stations in the city, and the running time of the bus in the two adjacent stations is obtained. The time interval data were fitted linearly. Using two data sets from a city in South China and a city in North China, the results show that the bus running time interval is in accordance with the power exponential distribution; the time interval of bus operation is in line with human behavior dynamics.

Abstract:In order to optimize the current Particle Swarm Optimization (PSO) algorithm, which is easy to fall into local optimum, slow convergence and another faults, this study proposes an improved inertia weight parameter method to optimize the algorithm. Combining the operation of mutation operator in Differential Evolution (DE) algorithm to improve the self-adaptation of the algorithm and limit the speed and search space of the algorithm to prevent particles from jumping out of the prescribed search space. Choose the corresponding test function and compare the improved algorithm with the other two algorithms by using Matlab software. The results show that the improved algorithm has a certain improvement in the convergence speed and the stability of fitness value.

Abstract:Aiming at the problems of small space and low complexity of low-dimensional chaotic system and single DNA encryption scheme, a color image encryption algorithm based on multi-chaotic mapping and DNA is proposed. Firstly, Arnold transform is used to scramble the position of each component of the image. Logistic-sine chaotic map is used to generate random matrices of the same size as the plaintext image and block them. Then DNA rule operation is performed. The operation mode is determined dynamically by the chaotic sequence generated by Chen hyperchaotic system. The simulation results show that the algorithm has good encryption and recovery effect, can effectively resist various statistical attacks and differential attacks, and has good security, anti-noise, complex and high encryption performance.

Abstract:In the face of increasing large-scale scheduling problems, the development of new algorithms becomes more and more important. A Q-Learning scheduling algorithm based on reinforcement learning is proposed for permutation flow shop scheduling problem. By introducing state variables and behavior variables, the scheduling problem of combinatorial optimization is transformed into sequential decision-making problem to solve the permutation flow shop scheduling problem. The proposed algorithm is used to test the Flow-shop international standard provided by OR-Library, and compared with some existing algorithms, the results show that the algorithm is effective.

Abstract:With the rapid development of economy, many enterprises are stepping into the era of scientific management. Sales forecasting is an indispensable link in the business activities of enterprises. The accuracy of forecasting is directly related to the success or failure of sales operations. Therefore, an improved BP neural network forecasting model based on the combination of traditional BP neural network and time series forecasting model is proposed. Through the forecasting of this model, the sales volume of enterprises in the future per unit time can be more reliably predicted. The improved neural network makes self-calibration referring to the prediction of synchronous time series, and uses genetic algorithm to achieve self-optimization through calibration, simplifies the network structure and improves the accuracy of prediction.

Abstract:Demand forecasting is the basic link in the organization of POL support for synthetic brigades, which has a relatively important impact on the successful military operations of synthetic brigades. Because of the particularity of the composition structure of synthetic brigade, the traditional forecasting methods have some drawbacks. Therefore, a demand forecasting method for synthetic brigade based on fuzzy clustering and intuitionistic fuzzy reasoning is proposed. Firstly, the fuzzy C-means clustering algorithm is used to realize the preliminary screening of historical cases in order to improve the speed of case retrieval. Then, the subjective and objective comprehensive weight model of case feature attributes and the case retrieval model based on intuitionistic fuzzy sets are constructed to ensure the accuracy of case retrieval. Finally, a POL demand forecasting model for synthetic brigade based on the overall data characteristics is constructed. The feasibility and practicability of the forecasting method are verified by an example analysis, which proves that the proposed method is helpful to improve the retrieval speed and forecasting accuracy.

Abstract:This study is designated to solve the problem of that server system and software’s security configurations can be reset after the server is invaded, and the network security equipment (hardware firewall, etc.) has large granularity. We analyze common network applications of WWW server, such as WWW, DNS, and FTP, summarize of the characteristics of each network application protocol, and according to the principle of dynamic port fixation and dynamic managerment IP fixation, configure the server access switch ACL, then apply each server’s ACL to the server-connected switch port, protect the server specially. When the server firewall rules are disabled, the server access switch ACL can limit the behavior of the server, thus protecting the servers and the intranet network devices. Using the Pktgen tool based on INTEL DPDK (Data Plane Development Tool) to test, ACL in the server access switch can filter the high abnormal traffic from the server and protect the network and equipment.

Abstract:In view of the increasingly serious internal threat behaviors in enterprise information system, especially the behaviors such as pseudonym login and unauthorized operation, based on the technology of user behavior analysis, a layered security model with a mixture of subject and object is adopted to establish a new internal threat detection framework of information system. Malicious insider threat behavior is found by comparing the abnormal behavior of users and the authority of subject and object. Regular expression and mixed encryption algorithm are used to ensure the accuracy of detection and log security. Security detection is carried out from four aspects: identity authentication, access control, operation audit, and behavior threshold technology. The key technologies are introduced in detail. Experiments show that the proposed detection framework can prevent internal personnel from stealing data, provide response and intervention capabilities, and improve the security of information systems. Finally, the development trend of internal threat detection technology is prospected.

Abstract:In recent years, the web content detection mainly focuses on how to extract features from HTML document through semantic analysis or emulation execution, while it is undesirable, because it significantly complicates implementation which requires high computational overhead, and opens up an attack surface within the detector. A deep learning approach to detect malicious web pages is proposed. Firstly, we take advantage of the non-complex regular expression to extract tokens from static HTML document, then capture locality representation at multiple hierarchical spatial scales over the document with neural network model, by which the mode can quickly find tiny fragments of malicious code in any length of web pages. The experimental results show that this approach achieves a detection rate of 96.4% at a false positive rate of 0.1%, much better than the baseline and simplified model at the classification accuracy. The speed and accuracy of proposed approach makes it appropriate for deployment to endpoints, firewalls and web proxies.

Abstract:As a popular data carrier, E-mail is an important tool in campus daily work and study, and account stolen issues are typical threats to the E-mail security. Firstly, a systematic study on the optimal security management of Nanjing Agricultural University self-built E-mail system is carried out in this study, then potential risks, courses, and strategies are discussed. Finally, the comprehensive optimization scheme is given, especially on the discovery and handling of account stolen issues, and a novel account status monitoring method based on automated scripts is proposed in this study. Implementation indicates that the proposed scheme significantly increases both the discovery rate and handling efficiency of account stolen issues, particularly stolen cases and spam mails from domain users are explicitly reduced, meanwhile, the manpower of management is saved and the user experience is improved. It also has a certain reference value in college E-mail system management applications.

Abstract:In the scenario of human-computer interaction by the mouth, the light changes, the complexity of the small target detection, and the detection method of none generality factors under different scenarios have brought great difficulties to detect the mouth. In this study, we take the face images with different scenarios as data source and propose a face recognition algorithm based on Faster R-CNN. In this method, multi-scale feature maps are combined in Faster R-CNN framework for detection. Firstly, we introduce a modified multi-scale feature map to effectively utilize multi-resolution information. Then, feature maps need to share the same size, so that element-wise sum operation can be performed. Features with higher resolution and stronger expression ability can be obtained by up-sampling on the output feature map. The detection performance of the small target is improved. In the training experiment, multi-scale training and increasing the number of anchor points are used to enhance the robustness of the network to detect targets of different sizes. Experiments show that the detection accuracy of the mouth is improved by 8%, and it is more adaptable to the environment compared with the original Faster R-CNN.

Abstract:Because there are many factors affecting coal mine water inrush and they have strong correlation, the prediction accuracy of the model will be affected. Due to the heavy workload and high cost of data collection, how to select features scientifically to improve the accuracy of model prediction has become the focus of this study. At first, this study uses stability selection to select 7 factors which are more important in 22 known influence factors, and then builds three typical machine learning classification forecasting models including random forest, neural network, and support vector machine (SVM) to forecast the data before and after feature selection. The experimental results show that the prediction model is very stable after the feature selection and prediction accuracy can reach 100%, and also decrease the cost of the sample data collection.

Abstract:Beidou RDSS has an advantagement of short-text-message communication. It can cover a large area, especially suitable for emergency communication in special situations such as isolated islands and deep mountains. It can obtain important data for dealing with emergencies. Based on the characteristics of Beidou RDSS short-text-message communication and the randomness of data in nuclear radiation monitoring industry, a method of Beidou emergency communication is proposed to coordinate the frequency of monitoring data and Beidou message frequency, which improves the communication emergency capability and intelligence of Radiation Monitoring System (RMS).