Table of Content

28 January 2021, Volume 0 Issue 01

Previous Issue    Next Issue

Improved GM (1,1) Grey Prediction Model Based on Background Value of Variable Weight Optimization and Its Application 

ZHANG Li-jie, SHA Xiu-yan, YIN Chuan-cun, DUAN Jun-tao, ZHANG Xin-yi, LI Zi-tong, JIANG Fu-lei

2021, 0(01):  1-6. 

Asbtract ( 247 )   PDF (1355KB) ( 158 )  

References | Related Articles | Metrics

In view of the disadvantage that the traditional GM(1,1) grey prediction model adopts equal weight in the background value, which results in low prediction accuracy, this paper proposes a variable weight optimization method for selecting background value. Firstly, the golden section search and parabolic interpolation method are combined to determine the background value of the improved GM (1,1) model. Then the improved background value is brought into the grey prediction algebraic recursive equation, replacing the whitening equation in the traditional GM(1,1) grey prediction model. Finally, we select the exponential sequence to simulate, and carry out a simulation experiment based on the actual statistical data of the number of teachers in a university. The results show that the improved GM (1,1) model reduces the average relative error, improves the prediction accuracy and has certain application value.

Vector Control of Permanent Magnet Synchronous Motor Based on Fuzzy PI

ZHENG Fei, WU Qin-mu

2021, 0(01):  7-11. 

Asbtract ( 215 )   PDF (1290KB) ( 98 )  

References | Related Articles | Metrics

Aiming at the problem that the traditional PID controller in the permanent magnet synchronous motor (PMSM) vector control system is not efficient in the PMSM control performance and is poor in the adaptability to working environment change, firstly, the mathematical model is established by adopting the dual-axis theory in the two-phase rotating coordinate system; Then, the fuzzy control theory is used to improve the traditional ordinary PID controller of speed loop in the PMSM vector control system, and the fuzzy PI controller is obtained from the theoretical level research and analysis. In Matlab/Simulink environment, a PMSM vector control system simulation model based on fuzzy PI controller is established, and the established fuzzy PI controller is introduced into the PMSM vector control system for simulation experiments. The experimental results show that compared with the traditional PID controller,  the fuzzy PI controller has superior dynamic performance and steady-state performance.

Design of RFID Power Tag for Electrical Equipment Against Electromagnetic Interference

DENG Zu-qiang, LIU Chao, ZHOU Jing, ZHANG Jin-luan, DONG Guo-chao,

2021, 0(01):  12-16. 

Asbtract ( 184 )   PDF (1598KB) ( 89 )  

References | Related Articles | Metrics

RFID power tags of traditional power tools cannot work normally in a strong electromagnetic interference environment, which can easily lead to tag damage and information leakage. In response to the above problems, this paper designs an RFID power tag that resists electromagnetic interference. The RFID power tag housing uses anti-interference materials and high insulation and voltage-resistant packaging. The tag uses a STM32L4 microcontroller and a two-stage filtering algorithm based on Kalman filtering is proposed. The data transmission uses UHF, which achieves a good anti-electromagnetic interference and anti-noise effect. In addition, the communication between the tag and the reader uses a physical unclonable function based on the fingerprint of the devices physical characteristics for encryption. Finally, it is verified by experiments that the RFID tags designed in this paper can not only resist strong electromagnetic interference, but also prevent information leakage, and can be well applied in complex power environments.

Multi-objective Optimization Model of  Power Material Storage Service Under Economic Benefits

PAN Zi-chun, CHEN Jia-yu, ZHEN Li-xia, CHU Li, YANG Fei-long

2021, 0(01):  17-21. 

Asbtract ( 97 )   PDF (716KB) ( 67 )  

References | Related Articles | Metrics

In order to achieve the multi-objective optimization of the storage business of electric power materials under the effect of good economic benefits, a multi-objective optimization model of the storage business of electric power materials based on the analysis of the characteristics of game correlation is proposed. Firstly, the electric power material storage business under the economic benefits is extracted by association rules. Then the feature decomposition and optimization extraction of electric power material storage business are carried out by the multi-objective optimization method of fuzzy association rules to complete the centralized multi-objective optimization of the established storage business, and the adaptive score of electric power material storage business under the economic benefits is carried out by the game correlation feature analysis method block matching. Finally, combined with deep learning method, the multi-objective optimization decision-making of electric power material storage business under economic benefits is realized. The experimental results show that after the application of the new optimization model, the multi-objective decision-making and evaluation level of the power material storage business is significantly improved, which fundamentally improves the dispatching ability of the power material storage business under economic benefits.

Performance of 5G-NR LDPC Code in Troposcatter Channel

WANG Bo-yuan, ZHANG Tao, WANG Wei

2021, 0(01):  22-27. 

Asbtract ( 201 )   PDF (1479KB) ( 60 )  

References | Related Articles | Metrics

This paper researches the encoding and decoding performance of 5G-NR LPDC in troposcatter channel， because of the lack of research on fading channel, especially in the nonlinear environment of time-varying channels. The 5G-NR LDPC code is classified as a quasi-cyclic LDPC code. Compared with the traditional coding scheme, it can support a variety of code rates and code lengths through the construction of the basis matrix, which is easy to realize rate-adaptive. This paper proposes a modified layered normalized minimum sum decoding algorithm, by modifying the normalization factor, which is more suitable for scattering channels. The simulation results show that compared with the traditional decoding method, the proposed algorithm improves the decoding speed by 3 times, reduces the number of iterations and reduces the complexity; and the performance of the proposed algorithm is improved by about 1.3 dB and 0.6 dB, respectively, either at low or high bit rates when the bit error rate reaches 10-5 under 16-diversity.

Exponential Weighted Smoothing Prediction Model Based on Abnormal Detection of Box-plot

GU Guo-qing, LI Xiao-hui

2021, 0(01):  28-33. 

Asbtract ( 156 )   PDF (1298KB) ( 176 )  

References | Related Articles | Metrics

Data prediction model is a research hotspot in the field of data transmission in wireless sensor networks in recent years. As the monitoring environment becomes more complex and diversified, the data set collected by the sensor node is often accompanied by abnormal points. Most of the current prediction models do not filter the abnormal points. In order to effectively filter out abnormal points and improve the streamlining of data transmission and the accuracy of prediction, this paper proposes an exponentially weighted smoothing prediction model based on box-plot abnormal detection, and introduces a short-term chain ratio mechanism to determine emergencies. Experiments show that the model can effectively filter out abnormal points and determine emergencies under different data sets, smoothness coefficient changes and different dynamic thresholds. The streamlining of data transmission is increased by 5.8%, and the prediction accuracy is increased by 8.4%. Compared with the existing prediction models, it has better robustness and abnormal point processing ability.

MR Brain Tumor Image Segmentation Based on Rough Set Adaptive Granularity

YAO Chuan-wen, HUANG Dao-bin, YE Ming-quan,

2021, 0(01):  34-37. 

Asbtract ( 145 )   PDF (550KB) ( 81 )  

References | Related Articles | Metrics

Partial volume effect and uneven gray level of magnetic resonance (MR) imaging make the brain tumor image segmentation low accuracy. In order to solve the problem, a new brain tumor image segmentation method is proposed, which is a rough set adaptive granularity method. The method firstly uses the tumor features in the image to adaptively select the optimal granularity. It uses the rough set idea to simulate the upper and lower approximation of the target and background regions. The best threshold for MR brain tumor image segmentation is obtained  by optimizing the roughness of the target and background regions. This method can extract the area of brain tumors. The experimental results show that the method is superior to the traditional rough set segmentation methods. This method has certain practicability and flexibility.





HEp-2 Image Recognition Based on Deep Residual Shrinkage Network

HE Tao, CHEN Jian, WEN Ying-you

2021, 0(01):  38-42. 

Asbtract ( 179 )   PDF (2330KB) ( 76 )  

References | Related Articles | Metrics

Detection of antinuclear antibodies by human epithelial cells (HEp-2) is a common method for the diagnosis of autoimmune diseases. Image recognition of HEp-2 cells is of a great significance for the diagnosis and treatment of many autoimmune diseases. Aiming at the problems of low efficiency and high labor intensity caused by manual evaluation methods, a HEp-2 cell image classification model based on the depth residual shrinkage network is proposed. The model is improved on the basis of the deep residual network, and the residual learning module can train the deeper network by using the identity mapping method. In each residual learning module, a soft threshold nonlinear transformation sub network is embedded. The soft threshold is used to eliminate the noise and redundant information in the data. These thresholds are automatically learned by the sub network. Experiments show that this method has good performance and is superior to other depth neural network methods.

A Method to Identify Diseases of Bridges Based on Transverse Cracks

MA Dong-qun, LI Bao-lin, WANG Qiu-yue, HE Xian-bo

2021, 0(01):  43-49. 

Asbtract ( 113 )   PDF (4875KB) ( 63 )  

References | Related Articles | Metrics

Bridge safety is one of the most important parts of traffic safety. Bridge inspection is the most effective evaluation method for bridge safety, and the identification of diseases is especially important in bridge inspection. According to relevant statistics, more than 80% of the diseases of concrete bridges in China are caused by cracks. Therefore, the health inspection of bridges is mainly the detection and measurement of the apparent cracks of bridges. In order to improve the accuracy and efficiency of detection, and to count the accuracy and convergence of bridge crack disease images, this paper combines Convolutional Neural Networks (CNN) technology and uses bilinear interpolation provided by TensorFlow to preprocess crack images. We preprocess crack images into resolutions of 75 px, 150 px and 300 px; and build a CNN training model in the TensorFlow architecture, use this model to train horizontal crack disease pictures with resolutions of 75 px, 150 px and 300 px respectively, and analyze at different resolutions under the circumstances, the training accuracy, convergence and efficiency of the lateral crack disease. The three different resolutions of the transverse crack disease picture training result model are saved as the test model of the lateral crack disease identification, and these models are used to test the lateral crack picture in the actual engineering application and the lateral crack picture in the test set for analyzing the recognition accuracy, convergence and efficiency of the test set pictures after preprocessing and the transverse crack pictures taken in the actual environment. Experimental results show that compared with the recognition algorithm used by Liu Hong-gong et al, the training accuracy, scalability and real-time performance of the algorithm in this paper are better, and the recognition rate is as high as 99% and above, which lays a good foundation for the recognition of other crack diseases. The foundation can also provide reference data for the detection of bridge cracks.

Smoke Removal Algorithm of Medical Operation Image Based on Conditional Generative Adversarial Network 

MA Yue

2021, 0(01):  50-55. 

Asbtract ( 166 )   PDF (1774KB) ( 81 )  

References | Related Articles | Metrics

The image smoke removal algorithm in medical operation can improve the imaging quality and reduce the harm of image-guided operation, which is a very ideal preprocessing method in many clinical applications. To solve this problem, this paper proposes an image smoke removal network based on conditional generation countermeasure model, which is composed of generator and discriminator subnetworks. Among them, the Tiramisu model is used instead of the traditional U-Net model to get higher parameter efficiency and performance. In addition, it provides a new way to generate a large number of training data sets for such problems by using the computer graphics rendering engine. The experimental results show that this method can effectively reduce the smoke while retaining the important perceptual information of the image, and is superior to the existing image smoke removal algorithms in both qualitative and quantitative analysis, thus providing a better visual field for surgeons.

Internal Attack Detection Based on Shell Command

CHEN Ming-shuai, WU Ke-he

2021, 0(01):  56-60. 

Asbtract ( 160 )   PDF (771KB) ( 73 )  

References | Related Articles | Metrics

Information system not only faces the threat of external attack, but also faces the threat from the internal system. In this paper, aiming at the internal attacks of the system, the internal threats and internal attacks of the information system are briefly described and analyzed. Based on the general rules of user’s operation behavior, this paper proposes several detection models, and finds out a good detection model by comparing the detection results. Based on SEA open data set, feature extraction uses several methods, such as word bag, TF-IDF, vocabulary and N-Gram, and uses different machine learning algorithms to build detection model, including XGBoost algorithm, implicit Markov and multi-layer perceptron (MLP). The results show that the accuracy and recall rate of the test samples using the word bag+N-Gram feature model and XGBoost learning algorithm are high, and the detection effect is the best.

A MQTT Abnormal Traffic Detection Method Based on Random Forest Algorithm

WU Ke-he, ZHANG Ying, CUI Wen-chao, CHENG Rui

2021, 0(01):  61-64. 

Asbtract ( 196 )   PDF (938KB) ( 88 )  

References | Related Articles | Metrics

With the wide application of Internet of things technology, the industrial Internet of things system suffers from increasing network security threats, and information security becomes a major challenge in its development. The MQTT (Message Queuing Telemetry Transport) protocol is the mainstream protocol for Internet of things communication. The research on communication security of Internet of things based on the protocol is a hot topic at present. In order to ensure the communication security of restricted devices in the Internet of things, this paper focuses on the abnormal detection of MQTT traffic. Traditional traffic identification technology such as deep packet inspection cant effectively identify abnormal traffic conforming to packet format, and abnormal traffic identification technology based on machine learning theory shows very good effect. For this, a MQTT abnormal traffic detection method based on random forest algorithm is proposed, which achieves an overall accuracy of more than 90% and gets better recognition effect than other common classification models.

Intrusion Detection Based on Focal Loss and Convolutional Neural Network

YAN Rui-an, ZHANG Li-chen

2021, 0(01):  65-69. 

Asbtract ( 159 )   PDF (952KB) ( 86 )  

References | Related Articles | Metrics

Intrusion detection is an important link in the field of information security protection. With the development of network technology, the active defense against network intrusions becomes more and more important, and intrusion data becomes more massive, complex, and unbalanced, which leads to the detection performance of the traditional intrusion detection technology is relatively low, so how to improve the detection performance of the intrusion detection system for unbalanced data sets is a huge challenge. The traditional CNN model has a good performance for processing complex data, but its effect of dealing with imbalanced data set is not very good. In order to solve this problem, an intrusion detection method based on Focal Loss and convolutional neural network is proposed. Different from traditional convolutional neural network, this model uses the Focal Loss function to solve the data imbalance problem, and in the convolutional layer, a regularization method (DropBlock) is added to improve the generalization ability of the model. Experiments of using KDD 99 data set show that the accuracy and precision of intrusion detection of this model are improved compared with the traditional intrusion detection model.

Weighted Slope One Optimization Combining User Fuzzy Clustering and Similarity

SHI Peng, YAO Wen-ming, WANG Xiang

2021, 0(01):  70-75. 

Asbtract ( 138 )   PDF (822KB) ( 72 )  

References | Related Articles | Metrics

In the case of sparse data sets, the traditional Slope One algorithm has poor recommendation and low accuracy, and the algorithm treats all users equally without considering the similarities and differences between users. At the same time, as the amount of data increases, the real-time performance has gradually deteriorated. In view of the above problems, a weighted Slope One algorithm optimization study is carried out. Firstly, we use fuzzy clustering technology to classify different types of users and reduce the nearest neighbor search range and calculation complexity. Then, we improve the weighted Slope One calculation formula and use the Pearson correlation coefficient to limit the calculation. finally, we use the improved weighted Slope One algorithm to predict user ratings in each cluster, and then generate a recommendation set. Experiments show that the algorithm in this paper effectively improves the accuracy of recommendations and enhances the real-time performance of recommendations.

Ontology Based Knowledge Modeling and Analysis of Military Equipment

LIU Meng-chao, WANG Yu-mei, WU Ya-fei, ZANG Yi-hua, LIANG Jia

2021, 0(01):  76-80. 

Asbtract ( 176 )   PDF (883KB) ( 70 )  

References | Related Articles | Metrics

Aiming at the problem that different audiences have different application requirements in the process of ontology modeling, a demand-based spiral feedback method is proposed by combining the seven-step method of ontology modeling with the spiral model of software development. Firstly, we clarify the domain knowledge category and user needs, and carry out the overall ontology design; secondly, the domain experts evaluate the knowledge system according to the principles of ontology construction; and then, we assemble the knowledge system model, define the ontology attributes, create examples, and complete the detailed ontology design; finally, user’s evaluation results are fed back, a new round of incremental iterations starts until a correct and usable knowledge system is formed. Compared with the seven-step method, the proposed method emphasizes ontology testing and user requirements, and constructs domain knowledge models with small segments. In addition, the military equipment knowledge system realized by this method is oriented to combat needs and provides strong theoretical support for military operations.

Design of Data-Driven Fault Detection Toolbox in MATLAB Environment

GUO Jin-ping, BIAN Ruo-peng

2021, 0(01):  81-86. 

Asbtract ( 209 )   PDF (1181KB) ( 88 )  

References | Related Articles | Metrics

Fault detection is an important technology to ensure the normal operation of modern industry. With the large-scale and complex production, data-driven fault detection has been widely used in the field of fault detection. The basic principle and the implementations of data-driven fault detection including the univariate statistical detection and the multivariate statistical detection are illustrated in this paper. Based on the basic theory of data-driven fault detection and the platform of MATLAB GUIDE, the design scheme and the implementation process of the visual detection toolbox are proposed. According to the features of the monitored data, the relevant detection method can be chosen in the designed toolbox to improve the fault detection ratio effectually. The fault detection has been simulated by the toolbox in the Tennessee-Eastman process simulation system. The results show that the univariate method can realize the fault detection and even the fault isolation. But the method may produce a lot of occupying space of the detection charts. Besides, compared with the univariate method, the multivariate method can get the detection results more directly with higher sensitivity.

Analysis of Spatio-temporal Interaction Characteristics of Urban Area Based on Taxi GPS Trajectory

YANG Wen-liang, FENG Hui-fang

2021, 0(01):  87-93. 

Asbtract ( 175 )   PDF (4553KB) ( 141 )  

References | Related Articles | Metrics

Urban big data provides data support for exploring the behavior characteristics of urban residents’ travel. Combining data mining with visualization technology, the resident travel law and urban space interaction characteristics of Lanzhou are studied based on the taxi GPS trajectory. Firstly, the travel characteristics and the inter-district spatial interaction characteristics of the four districts are analyzed. Then, the traffic trips between the urban grid and urban traffic hotspots of weekday and weekend are studied by the CLARA clustering algorithm. Finally, a directed-weighted complex network model is established to analyze the space interaction characteristics between urban traffic hotspots. The results show that there are significant differences in the spatial and temporal characteristics of urban travel behaviors and urban space interaction characteristics in the weekday and weekend. Compared with weekend, urban travel in the weekday are more compact and purposeful. The cluster structure of travel topology presents a "dumbbell" distribution shape matching the valley topology of Lanzhou. The spatial interaction between adjacent clustering areas close to the city center is strong. The results of this study can provide decision-making services for urban traffic management and residents’ travel.

Semi-supervised Clustering Ensemble with Pairwise Constraints Based on PCA Dimension Reduction 

HUANG Xin-chen, GAO Jun, HUANG Hao-jie

2021, 0(01):  94-99. 

Asbtract ( 156 )   PDF (852KB) ( 72 )  

References | Related Articles | Metrics

Aiming at the problem that the existing clustering integration algorithms are unsupervised and cannot deal with high-dimensional data well, this paper proposes a semi-supervised clustering ensemble with pairwise constraints based on PCA dimension reduction (SSCEDR), the SSCEDR method uses PCA principal component analysis to reduce the dimension of the original data. Combined with semi-supervised clustering integration technology, the prior knowledge such as pairwise constraints is substituted into the clustering integration process in the reduced dimension space. The effectiveness of the algorithm is verified by experiments on multiple data sets.

Analysis of Airline Customer Churn by Random Forest Algorithm Based on RFM Model

YANG Lin, BAI Zhao, KOU Yong-gang

2021, 0(01):  100-104. 

Asbtract ( 387 )   PDF (823KB) ( 78 )  

References | Related Articles | Metrics

In recent years, with the rapid development of the aviation market, it is urgent for airlines to control the loss of customers while increasing their market share. Based on the random forest algorithm, according to the data of aviation customers, a loss prediction model is established to predict whether customers have lost. The traditional RFM customer value model is improved and the random forest algorithm is used to predict customer churn. The experimental results show that the customer churn model based on RFM stochastic forest algorithm has a more persuasive index selection, an AUC value reaches 0.92 and the accuracy is higer. The model can be used to predict the loss of airline customers accurately, classify the lost customers and provide marketing strategies for civil aviation enterprises.

Requirements Document Named Entity Recognition Based on Deep Learning and Grammatical Regulations 

XU Meng-di, WANG Jin-hua

2021, 0(01):  105-110. 

Asbtract ( 102 )   PDF (1377KB) ( 79 )  

References | Related Articles | Metrics

Named entity recognition is particularly critical in natural language processing. There are overlong entities in the requirements document: virtual function, which makes it hard for pervasive traditional named entity recognition method to recognize entire entity. This paper conducts an in-depth research on the entity recognition model of requirements documents, introduces CNER method, which is based on Deep Residual Network (ResNet), to combine with the method based on grammatical regulations to perform word segmentation of Chinese requirements documents. This paper’s NER model is an encoder-decoder model, applies Bidirectional Long Short-Term Memory network (BiLSTM with attention) to encode, which obtains the context features and sentence pattern features of the text after word segmentation, employs conditional random field (CRF) method to decode, then identifies the requirements document entities with the intervention of grammatical regulations as a combination. Experiments show that the proposed method has better recognition effect than the pervasive traditional methods.

Semantic Similarity Calculation of Cilin Based on Logistic Function

YANG Quan

2021, 0(01):  111-119. 

Asbtract ( 126 )   PDF (1024KB) ( 71 )  

References | Related Articles | Metrics

At present, there is a certain gap between the calculation results of semantic similarity of words and the results of artificial discrimination, mainly because the semantic similarity calculation based on knowledge ontology generally uses the semantic classification dictionary directly from the perspective of mathematical calculation, but does not make full use of the linguistic knowledge in the dictionary from the perspective of lexicology. Therefore, we use the theory of semantic field to analyze the organizational relationship between words in Cilin, and expound the decisive role of depth in semantic similarity and the auxiliary role of branch information. On the basis of the combination of depth with branch information of Cilin, the Logistic function calculation model is proposed. The Pearson correlation coefficient between the calculated result of semantic similarity of MC30 and the manually labeled value is 0.9540; the average root error is 0.0191; the Pearson correlation coefficient between the calculated result of semantic similarity of RG65 and the manually labeled value is 0.9434; the average root error is 0.0193.

Algorithm of Answer Sentence Selection Based on Q and A Interaction

HOU Jia-zheng, ZHANG Shao-yang, YU Kun

2021, 0(01):  120-126. 

Asbtract ( 96 )   PDF (1294KB) ( 69 )  

References | Related Articles | Metrics

The accuracy of answer selection task has an important influence on the application of question answering system, text processing and so on. In order to solve the problem that the semantic information and the shallow features of question and the candidate answer are not fully utilized in the answer selection model, an answer selection model based on the interaction of question and answer is proposed. Given question Q and candidate answer A, the model first uses BiLSTM encoder to encode them, concatenates the two directions encode results of each time step of Q and A, and then uses Feed-Forward attention to encode question sentence Q; all the time steps of question Q are matched with all the time steps of A. According to the matching results, the weight of each word of the answer sentence is calculated, and then the sentence code of the answer sentence is calculated according to the weight of each word of the answer sentence. Finally, the sentence coding of Q and A sentences is input into the full connection layer after the aggregation operation, and the final judgment result is output by fusion with the co-occurrence feature of words. The experimental results on the DBQA data set show that this model can effectively improve the effect of answer selection task compared with the mainstream Siamese neural network.