Table of Content

20 May 2020, Volume 0 Issue 05

Previous Issue    Next Issue

A Wavelet Transform-based Feature Extraction Method for Workload Prediction

WANG Ke1,2, LI Hui1,2, CHEN Mei1,2, DAI Zhen-yu1,2, ZHU Ming3

2020, 0(05):  1.  doi:10.3969/j.issn.1006-2475.2020.05.001

Asbtract ( 216 )   PDF (1284KB) ( 230 )  

References | Related Articles | Metrics

In resource constraints condition, it is very important to make accurate predictions of the task execution time based on time-series resource and task status generated in real-time during task execution. In order to use time-series data effectively to realize accurate prediction, a load shedding strategy is proposed to determine the time points of prediction and data processing scheme. This strategy uses dynamic time warping (DTW) distance to measure the variation of similarity between subsequences and entire sequences and determine the data used for prediction. Then we use wavelet transform to calculate the wavelet coefficients of the time-series and extract the energy value of wavelet coefficients as the features of prediction. After that, we conduct the prediction for task execution time. Experiments show that the features extracted by this method contain most information than the entire sequence and result in high accuracy in predicting the task execution time.

Short-term Power Load Forecasting Method Based on Dual Graph #br# Regularized Non-negative Low-rank Matrix Decomposition

LIANG Shou-yu1, FANG Wen-chong1, WANG Jin2, ZHOU Zhi-feng1, ZHU Wen3, ZHANG Ji2

2020, 0(05):  7.  doi:10.3969/j.issn.1006-2475.2020.05.002

Asbtract ( 163 )   PDF (2163KB) ( 172 )  

References | Related Articles | Metrics

In the context of smart grids, accurate estimation and prediction of power load has become an important prerequisite for grid power planning, and is of great significance to grid operating safely and economically. Aiming at the cyclical fluctuation and non-periodic influence of power load data, a short-term power load forecasting method based on dual graph regularized non-negative low-rank decomposition is proposed. The method constructs the power load space-time matrix using historical data, and performs robust non-negative low-rank matrix decomposition on the matrix to simultaneously acquire the periodic mode and non-periodic influence of the power load. On this basis, the spatial and temporal correlation of the power load is integrated to further improve the matrix decomposition results, and finally the short-term prediction of the power load is obtained through matrix recovery. This method takes into account the missing complement and trend analysis of power load and derives an effective learning algorithm. The experimental analysis shows that compared with the related methods, the proposed method achieves better accuracy and robustness under multiple evaluation criteria of short-term forecasting of power load.

A Novel Evaluation Model for Urban Smart Growth Based on Principal #br# Component Regression and Radial Basis Function Neural Network

ZHU Hong-zhang1, LI Lian-yan1, REN Xiao-bin2, SUI Xiao-liang1

2020, 0(05):  15.  doi:10.3969/j.issn.1006-2475.2020.05.003

Asbtract ( 165 )   PDF (1714KB) ( 152 )  

References | Related Articles | Metrics

Urban smart growth has become an environmental protection method widely used by urban planners and decision makers to build cities, which has practical significance to measure the level of urban smart growth. In this paper, rational degree (RD) is defined to describe the level of urban smart growth, then RD model is established through principal component regression (PCR) and radial basis function (RBF) neural network. Taking Yumen and Otago as examples, their RD values are 0.04482 and 0.04591 respectively, which indicates that both cities’ smart growth development mode has achieved certain success. The urban development level of Otago is better than that of Yumen. At the same time, this study finds that Yumen should give priority to urban greening and environmental protection, while Otago should give priority to economic development. The proposed model provides a powerful reference for the cities to pursue scientific and smart growth.

Prediction of Power Customer Demands Based on Deep Neural Network

PENG Lu1, ZHU Jun2, ZOU Yun-feng2

2020, 0(05):  22.  doi:10.3969/j.issn.1006-2475.2020.05.004

Asbtract ( 208 )   PDF (1222KB) ( 190 )  

References | Related Articles | Metrics

Customer service of electric power enterprises is related to the vital interests of customers and the business benefits of enterprises. Improving the analyzing and understanding ability of the customer service system for group customers’ electricity consumption problems is one of the important ways to improve the quality of customer service for power industry. In order to solve the concentrated demand of power customers efficiently and pertinently, and achieve “before the customers think”, based on the deep neural network technology, this paper improves the traditional Chinese text segmentation technology and feature extraction method in the field of power, gives the method and flow of power customer demand pre-judgment, and verifies it through experiments. The proposed method can quickly and accurately classify the texts of power customer service order and excavate hidden customer power problems, which changes the service from passive to active and solves the potential demands of power customers at the first time.

An Optimal FCM Clustering Algorithm Based on Improved Bat Algorithm

CHANG Xue, SHI Hong-yan

2020, 0(05):  29.  doi:10.3969/j.issn.1006-2475.2020.05.005

Asbtract ( 188 )   PDF (1087KB) ( 156 )  

References | Related Articles | Metrics

Aiming at the traditional fuzzy C-means (FCM) clustering algorithm implicitly assumes that each sample and each dimension attribute have the same effect on the clustering results, which leads to the degradation of the clustering performance, and is sensitive to the initial center point and easy to fall into a local optimization, an optimal FCM clustering algorithm based on improved bat algorithm is proposed. Firstly, this algorithm improves the bat algorithm by using Logistic map and velocity weight. Secondly, the improved bat algorithm is used to determine the initial clustering center of FCM algorithm. Finally, according to the different effects of each sample and each dimension attribute on the clustering results, the objective function of FCM algorithm is redesigned by using the sample and attribute weighted method. Contrast experimental results show that the improved algorithm has better clustering effect.

An Imbalanced Data Classification of Hybrid Sampling Based on Clustering

SHI Ming-hua, WU Guang-chao

2020, 0(05):  34.  doi:10.3969/j.issn.1006-2475.2020.05.006

Asbtract ( 214 )   PDF (1050KB) ( 317 )  

References | Related Articles | Metrics

The imbalanced classification problem is widely used in real life. For most resampling algorithms, it focuses on the balance between classes and pays less attention to the problem of data distribution imbalance within classes, a hybrid sampling algorithm based on clustering is proposed. Firstly, the original data set is clustered, then the imbalance ratio is calculated for each cluster sample, and the cluster sample is processed according to the imbalance ratio. Finally, the balanced data set is put into the GBDT classifier for training. Experiments show that the algorithm has higher F1-value, AUC and better classification results than several traditional algorithms.


SVM-based Verification Method for New  Energy Bus Operation Mileage

ZHANG Wen-hua, ZHANG Zhi-jun

2020, 0(05):  39.  doi:10.3969/j.issn.1006-2475.2020.05.007

Asbtract ( 166 )   PDF (1242KB) ( 189 )  

References | Related Articles | Metrics

The verification of new energy bus operation mileage is an important part of the new energy bus subsidy declaration work. The verification of declaration data has problems, such as large workload and low efficiency. In this paper, descriptive analysis and one-way ANOVA of the declared data show that the length of the vehicle is related to the annual operating mileage of the vehicle. Based on this, the linear SVM algorithm is used to supervise the operational mileage declaration data. By using the linear kernel function, training the SVM and selecting the appropriate penalty parameter C, the optimal SVM is constructed to identify suspicious values. Experimental results show that the linear SVM algorithm can detect vehicles with suspicious annual operating mileage more effectively, which can provide reference for the verification of new energy bus operation mileage.

Driver Upgrade Strategy Based on Named Pipes and Heterogeneous Communication Mechanisms in Multiple Application Scenarios

YANG Gui-fu1, HU You-rong1, LIU Shu-xia1, LIU Zhen-bang2,3, BAO Yu2

2020, 0(05):  44.  doi:10.3969/j.issn.1006-2475.2020.05.008

Asbtract ( 151 )   PDF (1178KB) ( 109 )  

References | Related Articles | Metrics

In view of the problem that the electrochemistry legacy system drivers are not compatible with new operating systems, this study proposes a driver upgrade strategy based on named pipes and heterogeneous communication mechanisms. The named pipe and WinUSB are used as the middleware to complete the communication between the upper computer and the lower computer, so as to reduce the dependence of the device on the driver and improve the flexibility of communication. The results show that the relative error of experimental data between the proposed method and the original driver is 0.000～0.001. Comparing the upgraded system results with the original driver experimental data, it is consistent within the allowable error range. The method described in the article only needs to modify a small amount of legacy code to solve the compatibility problem between the driver and the new operating system. It not only facilitates the users who are not suitable for the new operating system, but also facilitates the maintenance work of the engineering developers on the legacy system, which realizes the low coupling and code reuse of the system, and has been applied in an electrochemical software.

Real-time Task Fault-tolerant Scheduling Algorithm for Data Center Inspection Robot Information Platform

HU Quan-gui, ZHAO En-lai, JIA Wei-zhao, KAI Bei-qiang

2020, 0(05):  50.  doi:10.3969/j.issn.1006-2475.2020.05.009

Asbtract ( 185 )   PDF (1276KB) ( 241 )  

References | Related Articles | Metrics

In order to realize real-time task fault tolerance of data center patrol robot information platform, a real-time task fault tolerant scheduling model of data center patrol robot information platform based on adaptive feedback equalization and symbol modulation technology is proposed. Firstly, the transmission channel model of patrol robot information platform in data center under route conflict is constructed, and the information transmission protocol of patrol robot in data center is optimized. Then, the information fusion of patrol robot is carried out by using the method of fuzzy C-means clustering, the channel equalization design of patrol robot information transmission is carried out by combining the adaptive feedback equalization method, and the real-time tasks fault-tolerant scheduling of information platform is carried out by symbol modulation method. Finally, the simulation results show that the method has better fault tolerance and better channel equalization for real-time task scheduling of the data center patrol robot information platform, which improves the real-time task scheduling ability of the data center patrol robot information platform.

Crack Recognition of Outcrop Area Based on Deep Learning

LUO Wei, LIANG Shi-hao, JIANG Xin, AN Ni, DU Rui

2020, 0(05):  56.  doi:10.3969/j.issn.1006-2475.2020.05.010

Asbtract ( 343 )   PDF (2235KB) ( 280 )  

References | Related Articles | Metrics

Aiming at the fact that the rock fractures and the surrounding environment in the field of outcrops in the current geological survey are more complicated and the data relies on manual depiction and traditional image processing algorithms, the recognition efficiency and accuracy are low, which makes the research of geological survey difficult, a deep learning rock fracture identification algorithm in outcrop areas is presented, thereby improving the accuracy and efficiency of rock fracture identification. This method is based on the TensorFlow architecture. First, the preprocessed training dataset pictures are manually selected and preprocessed into two types of pictures: cracks and backgrounds. Then the classified pictures are passed to the designed convolutional neural network model for training and saving the parameter data of the model, the trained model data is used to identify the preprocessed rock fracture pictures and record the fracture location information, and the fracture location information is used to locate and display the fractures of the unprocessed primary color rock fracture pictures. The experimental results show that the method can identify fractures with higher accuracy, and provide a more accurate and convenient fracture identification method for geological surveys.

A Survey of Research on Target Detection Algorithms Based on Deep Learning

CAO Yan, LI Huan, WANG Tian-bao

2020, 0(05):  63.  doi:10.3969/j.issn.1006-2475.2020.05.011

Asbtract ( 369 )   PDF (1000KB) ( 512 )  

References | Related Articles | Metrics

Traditional target detection algorithms rely mainly on manually selecting features to detect objects. The artificially extracted feature pairs are mainly for certain specific objects, such as some features suitable for edge detection, and some suitable for texture detection, which is not universal. In recent years, deep learning has flourished, and significant research progress has been made in the field of computer vision such as image classification, target detection, and image semantic segmentation. As a feature learning method, deep learning can automatically learn the useful features of the target, avoiding the problem of manual extraction of features, and at the same time ensuring good detection results. Firstly, the research progress of target detection algorithm based on deep learning is introduced. Secondly, the common problems and solutions in target detection algorithm are summarized. Finally, the possible development direction of target detection algorithm is prospected.

Reduced-reference Crop Image Quality Assessment Based on Random Gabor Feature

WU Shi-hai, BAO Yi-dong, CHEN Guo, CHEN Qiu-shi

2020, 0(05):  70.  doi:10.3969/j.issn.1006-2475.2020.05.012

Asbtract ( 164 )   PDF (1676KB) ( 254 )  

References | Related Articles | Metrics

With the deepening reform of agricultural modernization, informatization and automation, information technology plays an important role in its development process, and gradually serves all aspects of agricultural production. Among them, the new intelligent agricultural system based on computer vision and digital image processing technology has become the research focus and hotspot in the agricultural informatization development. For this reason, this paper takes crop image as the research object, and proposes a reduced-reference crop image quality assessment model based on random Gabor feature. The kernel function of Gabor filter can better describe the field characteristics of simple visual neurons. Its multi-frequency and direction characteristics are similar to the way of humans visual system which perceives images. Based on this fact, Gabor filter is used to analyze the texture and edge distribution characteristics of crop images, and a reduced-reference quality assessment model is established. The experimental results show that the crop image quality assessment model proposed in this paper can recognize and perceive the degradation of crop image quality very well, and plays an important role in the new intelligent agricultural system.

Virtual Reality Training System Based on Main Transformer Fire Emergency Drill

GUO Jian-long1, XIONG Shan1, LI Xiao-ying2, QI Yan-wei2, WU Cheng-kai3

2020, 0(05):  75.  doi:10.3969/j.issn.1006-2475.2020.05.013

Asbtract ( 162 )   PDF (2323KB) ( 125 )  

References | Related Articles | Metrics

In order to improve the training ability of fire emergency drill, a virtual reality training method based on main transformer fire emergency drill is proposed. The single-scale characteristic parameters of the fire environment are calculated under the 3D virtual reality scene model, and the image output model is constructed according to the parameter results. Combined with the fuzzy control method of the main transformer fire emergency drill, the 3D reconstruction of the virtual reality training is carried out, the control of the virtual reality training of the main transformer fire emergency drill is realized, and the scene reconstruction of the main transformer fire emergency drill is carried out in the virtual reality system. 3D virtual reality simulation of main transformer fire emergency drill is carried out in viewpoint position, and virtual reality training design based on main transformer fire emergency drill is realized. The simulation results show that the artificial intelligence and control ability of the virtual reality training system designed by this method are good.

Review of Research on Indoor Positioning

XUE Wei-lian, ZHAO Di, ZHANG Ying-chao

2020, 0(05):  80.  doi:10.3969/j.issn.1006-2475.2020.05.014

Asbtract ( 305 )   PDF (2293KB) ( 198 )  

References | Related Articles | Metrics

At present, as the precision of location based service is more and more demanding, the precision of indoor positioning is far from meeting our demand. Therefore, indoor positioning has become a research hotspot now. By combing and comprehensively reviewing the current research achievements of indoor positioning, we can further clarify the current situation of the research in this field and determine the direction of research in the future. The paper takes the Web of science (WOS) and CNKI as the data source, using Citespace to research analysis of literature of indoor positioning from domestic and foreign. We summarized the research directions of indoor positioning and research hotspots of indoor positioning in current. Finally, we present the future of prospects of research on indoor positioning.

A Heterogeneous Information Network Represention Learning Method Based on GAN

ZHOU Li1,2, SHEN Guo-wei1,2, ZHAO Wen-bo1,2, ZHOU Xue-mei1,2

2020, 0(05):  89.  doi:10.3969/j.issn.1006-2475.2020.05.015

Asbtract ( 194 )   PDF (1782KB) ( 147 )  

References | Related Articles | Metrics

Heterogeneous information networks contain rich structural and semantic information. It is a hot topic of current research to retain the structural and semantic information of heterogeneous information networks through network representation learning. Traditional heterogeneous information network representation learning methods are limited to preserving semantic information in heterogeneous information networks in the form of meta-paths, which is lack of considering the distribution of all nodes in the network, and the retained information is insufficient. Therefore, this paper proposes a heterogeneous information network representation learning method (HINGAN) based on a generative adversarial networks (GAN), which can better retain the information in the network and improve the robustness of representation learning. First of all, the GAN generation model generates fake data that retains network information, then samples the real data from the Gaussian distribution, and finally sends the real and fake data into the GAN discriminant model at the same time. Through the game of the generated model and the discriminant model, the node representation with more information is obtained. The experimental results based on two real data sets show that the proposed model is better than the traditional heterogeneous information network method in node classification and link prediction tasks.

Single Intersection Traffic Signal Coordination Control Based on Q-learning

HU Yu, LIU Mei-ling, ZHOU Zi-ang, ZHANG Min

2020, 0(05):  96.  doi:10.3969/j.issn.1006-2475.2020.05.016

Asbtract ( 161 )   PDF (1077KB) ( 102 )  

References | Related Articles | Metrics

Q-learning uses the interaction with the external environment to carry out the traffic signal adaptive control of a single intersection. In the background of the increasingly congested urban traffic, in order to alleviate the traffic congestion, a Q-learning algorithm combined with the green signal ratio optimization method of SCOOT system is proposed. In this paper, the method of green signal ratio optimization in SCOOT system is combined with Q-learning, that is, a new mathematical model is established as the cost function of the algorithm by combining the time factors such as average vehicle delay rate, parking times and economic factors, and a continuous reward and punishment function is established. On this basis, the operation process of Q-learning algorithm on a single intersection is introduced in detail, and through the horizontal comparison with Webster delay rate and Q-learning based on the minimum average vehicle delay rate, it is verified that this algorithm is superior to the timing control and Q-learning algorithm based on average vehicle delay. Compared with these two algorithms, the algorithm proposed in this paper is more suitable for the single intersection green signal ratio optimization.

A System of Sports Monitoring and Fall Warning for Elderly

LIU Bo1, WANG Ming-wei2, CHANG Li-bo3

2020, 0(05):  101.  doi:10.3969/j.issn.1006-2475.2020.05.017

Asbtract ( 244 )   PDF (1145KB) ( 167 )  

References | Related Articles | Metrics

Physical exercise is one of the effective means to promote the old people’s health and longevity. In order to monitor the motion state of the elderly in real time, master the motion state parameters, and alarm the elderly for falls caused by an accident or a sudden disease, a portable human fall monitoring system is designed, which can monitor falls of the elderly in real time and send the location and alarm information to remote receiver. The system collects the motion attitude data of human body in real time by the three-axis acceleration sensor placed at the waist. The embedded processor and wireless network are used to realize data processing, wireless transmission and remote alarm. The three-level threshold human fall detection algorithm is used to extract the acceleration characteristics of human fall posture change, grade the human motion state, and predict the serious fall behavior. The experimental results show that the system has the characteristics of stable performance, high accuracy, light and convenient, which is very suitable for the elderly to wear and use, and can ensure the elderly movement safety, and has a broad application prospect.

A Rate Watt-hour Data Processing System for Gate Internet of Things Terminal

HUANG Guo-bing1, WANG Jia-hao1, HUANG Kai2, WANG Qian1

2020, 0(05):  106.  doi:10.3969/j.issn.1006-2475.2020.05.018

Asbtract ( 142 )   PDF (1183KB) ( 240 )  

References | Related Articles | Metrics

Aiming at the requirement of rates watt-hour data uploading for gate energy remote terminal, it is necessary to extend the design of terminal software that the reading, local saving and uploading of rate electric energy need be realized. Because of the large capacity of minute power curve and the low access efficiency of embedded terminal electronic disk, information compression and the vertical and horizontal two-level structure are adopted to store files by day in order to locate and access information quickly and improve the performance of information processing. Considering the strictness of terminal processing information, multiple means are adopted to realize the integrity and traceability. At last, IEC 60870-5-102 protocol is extended for rates watt-hour data transmission, and these data are uploaded to master station of power billing system. This software is tested and verified in laboratory and substation field, all functions and performance indicators meet the requirements, and the software extension has achieved expected goal.

An IDS Alerts Preprocessing Method Based on Information Entropy

ZHANG Yu1,2, GUO Chun1,2, SHEN Guo-wei1,2, PING Yuan3

2020, 0(05):  111.  doi:10.3969/j.issn.1006-2475.2020.05.019

Asbtract ( 329 )   PDF (1911KB) ( 285 )  

References | Related Articles | Metrics

Focus on the issue for the large number of repeated alerts, high false alert rate, and low alert quality of the current intrusion detection system, an IDS alerts preprocessing method based on information entropy is proposed to reduce false alerts, aggregate similar alerts, and generate super alerts that represent the intent of a single step attack. First, the feature extraction of the IDS alerts are performed. The information entropy fusion result of the four characteristics of the alert density, the alert period value, the source IP address corresponding destination IP address numbers, and the attack source threat degree indicates the amount of feature information that an alert has. By calculating the Rainey information entropy with the feature vector of the false alert, the false alert is recognized and removed. Then, the alerts that the false alerts have been removed are classified into two types according to the IP correspondence: one type of source IP address corresponding to one destination IP address and one source IP address corresponding to multiple destination IP addresses. The two types of alert features are counted separately, and the 5-dimensional feature information entropy vectors are constructed. The DBSCAN algorithm is used to cluster the alerts with the same or similar information. Finally, the dynamic time windows are divided for each category of alerts, and the super alerts that represent the intention of single-step attacks are constructed. The experimental results show that the alerts preprocessing method based on information entropy has a false alert reduction rate of 87.43% and an alert aggregation rate of 98.63%, which has a good false alert reduction effect and a high alert aggregation rate.

A Low-rate DDoS Attack Detection Method Based on BiLSTM

JIANG Wan-ming1,2, GUO Chun1,2, JIANG Chao-hui1,2

2020, 0(05):  120.  doi:10.3969/j.issn.1006-2475.2020.05.020

Asbtract ( 243 )   PDF (1270KB) ( 244 )  

References | Related Articles | Metrics

 Low-rate distributed denial of service (LDDoS) attack is a new type of DDoS attack. Because of its characteristics of low-rate, periodicity and concealment, it avoids the traditional detection technology of DDoS attack and is more difficult to be detected and defended. This paper proposes a LDDoS attack detection method based on feature selection and bidirectional long short term memory (BiLSTM) neural network. In this method, recursive feature elimination CV (REFCV) feature selection algorithm of layered cross validation is used to mine the optimal 11 feature sets in two-way flow as input to the neural network, and a LDDoS attack detection classifier based on BiLSTM neural network model is established for classification, which achieves the purpose of LDDoS attack detection. Experimental results show that this method has higher detection rate than Kalman filter and NCAS algorithm, and lower false positive rate and false negative rate.