Computer and Modernization

Intelligent Test System of Terminal Equipment of Relay Protection Information System

WU Cong-yun, LIU Bin, LI Hai-yong, XU Xiao-feng

2021, 0(07): 1-5.

Asbtract ( 172 )

PDF (926KB) ( 129 )

References | Related Articles | Metrics

In order to achieve fast and accurate test results of relay protection information system terminal equipment, the intelligent test system of relay protection information system terminal equipment is designed. The terminal equipment data collected by the data acquisition module is transmitted to the data processing module through wireless mode; the interference data in the original data is removed by using signal processing algorithm to make it meet the requirements of the monitoring and early warning module; the real-time monitoring of the change of the operation state of the terminal equipment is realized, and the warning level information is sent out; the intelligent test module forms the terminal equipment data higher than monitoring and early warning indicators into recording data, uses the protection device and recording data to obtain the results of protection operation, completes the intelligent test of terminal equipment by evaluating the results of operation behavior. The experimental results show that the system can realize the intelligent test of relay protection information system terminal equipment, improve the working efficiency of test terminal equipment and the economic benefits of power enterprises.

Passenger-vehicle Matching and Route Optimization of Network Carpooling

CHEN Ling-juan, KOU Si-jia, LIU Zu-peng

2021, 0(07): 6-11.

Asbtract ( 425 )

PDF (1353KB) ( 135 )

References | Related Articles | Metrics

Urban road congestion and the prevalence of shared concepts have brought the rise of carpooling. Passengers with similar travel routes share the same car, which can increase the vehicle’s seat resources, save costs and relieve traffic pressure. Taking the problem of multi-vehicle static carpooling without transfer with time window constraints as the research background, the objective function of passenger-vehicle matching and path optimization is established from three aspects: vehicle usage fee, travel cost on the way and penalty cost of arrival time window, constructing model constraint conditions based on vehicle capacity, passenger departure and arrival time windows, no detours, no overlap between passenger and vehicle matching, etc. The evolution strategy algorithm is used to solve the problem, and the coding and decoding rules are designed according to the model characteristics. The decoding results can obtain the matching relationship between the vehicle and the passengers and the traveling path at the same time, and the cross-mutation operation is used to update the iterative individual population to obtain the optimal solution. Using MATLAB to solve the calculation example to verify the feasibility of the model and the effectiveness of the algorithm, the results show that the algorithm can quickly respond to the static carpooling problem, and can provide the matching relationship between passengers and vehicles and the path of the vehicle in a short time, the carpooling scheme can save more costs than traveling alone.

Link Prediction Algorithm Based on Resource Allocation and Graph Embedding Weighting

WAN Yang-ye, GUO Jin-li

2021, 0(07): 12-17.

Asbtract ( 138 )

PDF (1198KB) ( 90 )

References | Related Articles | Metrics

Most of the weighted link prediction algorithms with weight information have higher accuracy. However, most of the existing weighting algorithms are based on the external weight information, and there are few studies based on the weight of network topology. To solve this problem, a weighted link prediction algorithm is proposed, which uses the structural features of the unweighted network to generate the structure weight. Firstly, the resource allocation index is calculated to obtain the local structure similarity of the network. Then, the DeepWalk algorithm is used to learn the network structure features to generate the node vector to obtain the cosine similarity. The two similarities are combined to define the network structure weight. Finally, experiments are carried out on four network datasets, and three different types of similarity indices W-CN, W-LP and W-RWR are compared with those without weight information. The results show that the algorithm with structural weight information has higher prediction accuracy.

Named Entity Recognition Algorithm Based on Active Learning

ZHANG Cen-fang

2021, 0(07): 18-22.

Asbtract ( 225 )

PDF (715KB) ( 90 )

References | Related Articles | Metrics

The purpose of named entity recognition is to identify the boundaries and categories of entities in the text. In the process of training named entity recognition models, a large number of labeled samples are usually required. By implementing effective selection algorithms, this paper reduces the labeling of samples from a large number of samples suitable for model updates. By using five sets of comparison experiments, it is verified that a better set of samples can be obtained by effective selection algorithm, and a targeted sample of annotations is realized. Through experiments designed on microblog network data sets, it is verified that the current-based active learning algorithm can select more appropriate sample sets for a large amount of Internet text data, which can effectively reduce the cost of manual labeling. This paper uses two models to realize the boundary extraction and classification of entities. The sequence labeling model extracts the position of the entity in the sequence, the entity classification model realizes the classification of the labeling results, and uses the active learning method to realize the training on the unlabeled data set. Experiment on two data sets is done by using the training method in this article. Experiments on the Weibo dataset show that the algorithm can learn text features from the unlabeled dataset. The experimental results on the MSRA data set show that when the proportion of the pre-training data set reaches more than 40%, the F1 score of the model on the test data set is stable at about 90%, which is close to the result of using all the data sets, indicating that the model in unlabeled data sets has certain feature extraction capabilities.

Recommendation Algorithm Based on User Information Vector Clustering and Improved SAMME#br#

WANG Shan-wen, OU Ou, MA Wan-min, CHEN Jian-lin

2021, 0(07): 23-28.

Asbtract ( 181 )

PDF (1294KB) ( 88 )

References | Related Articles | Metrics

Aiming at the problem of imperfect user information acquisition and long recommendation time in the current mainstream recommendation algorithms, this paper proposes a recommendation algorithm based on user information vector clustering and improved SAMME. The algorithm analyzes basic user information (region, time, interest, tags, etc) to find user information keywords; weights different user information keywords based on the TF-IDF method to construct user information vectors; then uses the K-means algorithm to perform user clustering analysis, and uses the user clustering results as improved SAMME training sample set; finally, the prediction results are recommended to the user by the improved SAMME algorithm, and the model is saved during the training process, which greatly reduces the recommendation time. Finally, the algorithm of this paper is tested on the real Weibo user data set and compared with other mainstream algorithms. The results show that the algorithm of this paper achieves good results in accuracy, recall and F-value.

A Survey of Text Classification Based on Deep Learning

JIA Peng-tao, SUN Wei

2021, 0(07): 29-37.

Asbtract ( 585 )

PDF (1071KB) ( 264 )

References | Related Articles | Metrics

With the continuous development of the Internet, there is an increasing number of text data on the Internet. If these data can be effectively classified, it is more conducive to mining valuable information. Therefore, the management and integration of text data is very important. Text classification is a basic task in natural language processing tasks. It is mainly used in the fields of public opinion detection and news text classification. The purpose is to sort and classify text resources. The text classification based on deep learning shows a good classification effect in the processing of text data. The article elaborates on the deep learning algorithms used for text classification, classifies according to different deep learning algorithms, and analyzes the characteristics of various algorithms, and finally summarizes the future research directions of deep learning algorithms in the field of text classification.

Low-resource Neural Machine Translation Based on ELMO

WANG Hao-chang, SUN Meng-ran, ZHAO Tie-jun

2021, 0(07): 38-42.

Asbtract ( 198 )

PDF (601KB) ( 70 )

References | Related Articles | Metrics

The difficulty in low-resource neural machine translation is lack of numerous parallel corpus to train the model. With the development of the pre-training model, it has made great improvements in various natural language processing tasks. In this paper, a neural machine translation model combining ELMO is proposed to solve the problem of low-resource neural machine translation. There are more than 0.7 BLEU improvements in the Turkish-English low-resource translation task compared to the back translation, and more than 0.8 BLEU improvements in the Romanian-English translation task. In addition, compared with the traditional neural machine translation model, the simulated low-resource translation tasks of Chinese-English, French-English, German-English and Spanish-English increase by 2.3, 3.2, 2.6 and 3.2 BLEU respectively. The experimental results show that the ELMO model is effective for low-resource neural machine translation.

A Dynamic Search Algorithm of Picture Context Information Clustering in Mobile Crowdsensing

CAI Li-ping, ZHANG Chen-chen, LI Shi-bao, LIU Jian-hang

2021, 0(07): 43-48.

Asbtract ( 140 )

PDF (1012KB) ( 69 )

References | Related Articles | Metrics

Mobile device camera sensing is one of the main forms of mobile crowdsensing. The context information clustering of photos in advance can reduce the similarity calculation of image features and improve the efficiency of redundant judgment of photos. In order to improve the accuracy of context information clustering, this paper proposes a clustering dynamic search algorithm, which solves the problem of dynamic clustering near edge similarity. Firstly, according to whether the PTree clustering algorithm clusters to the existing interval, it is divided into real branches and virtual branches. The data of real branches and leaves are uploaded directly, and the virtual branches and leaves are further dynamically searched for the best similarity matching interval; then, based on the idea of using local expansion and then dynamic reduction, the average distance between dynamic clustering data points is reduced; finally, the redundant photos are removed by clustering to the same interval image set. By collecting photo data with context information through the designed APP, the results show that, compared with the existing schemes, the number of photos to be uploaded can be reduced and the effect of redundancy can be improved.

Local Road Network Traffic Flow Prediction Based on Improved eRCNN

YAO Si-jia, GUI Zhi-ming, GUO Li-min

2021, 0(07): 49-53.

Asbtract ( 157 )

PDF (1047KB) ( 77 )

References | Related Articles | Metrics

Error feedback recurrent convolutional neural network can only handle time series error sequences when it is applied to short-time traffic flow forecasting. The spatial topological characteristics implied in traffic flow error data are ignored. Furthermore, at model initialization time, it uses the general convolutional neural network initialization method to reduce the training efficiency of the model. Aiming at these problems, this paper presents an optimal error feedback recurrent convolutional neural network model. The error feedback layer structure in the model is strengthened according to the spatial and temporal characteristics of prediction error data.The model can deal with the error sequence with simple spatial relation. Through separating the history prediction error and training error, the training strategy is also improved to accelerate the convergence speed of the model. Experiments on traffic data of Beijing fourth ring road show that the optimized error feedback recurrent convolution neural network forecasting model can effectively improve the prediction accuracy, construction efficiency and robustness.

A Differential Evolution K-mediods Clustering Algorithm Based on Dynamic Gemini Population

DENG Bin-tao, XU Sheng-chao

2021, 0(07): 54-59.

Asbtract ( 149 )

PDF (848KB) ( 67 )

References | Related Articles | Metrics

With the appearance of massive big data, some new parallel computing models have been proposed for clustering algorithm. A dynamic gemini population based differential evolution K-mediods clustering algorithm called DGP-DE-K-mediods in cloud environments is proposed in this paper. In DGP-DE-K-mediods, gemini population scheme is adopted to improve the problem of being easily trapped into a local optimum while maintaining population diversity. The differential evolution algorithm is also used to make DGP-DE-K-mediods have strong global optimization capabilities. The DGP-DE-K-mediods clustering algorithm is designed and implemented in parallel under the Hadoop MapReduce framework and thus the time of the big data process has been significantly reduced. The programming model of MapReduce has been also described in detail for parallel cluster algorithm. A serial of simulation experiments are also done using UIC datasets and the intrusion detection processing of big data. Experimental results show that the overall detection effect of DGP-DE-K-mediods is significantly better than the existing intrusion detection algorithms.

Intelligent Talent Recommendation Method Based on Big Data Technology

WEI Yun-dong

2021, 0(07): 60-64.

Asbtract ( 241 )

PDF (1586KB) ( 135 )

References | Related Articles | Metrics

Complex and diverse job information makes it difficult for many job seekers to find suitable job information. In order to improve the quality of human resource recommendation, this paper designs a targeted talent market recommendation model based on gradient lifting tree and hybrid convolution neural network. The flow distributed method is used to collect the information of job seekers and convert it into a unique hot code which can be used for algorithm analysis. The gradient lifting tree is used to extract the features of job seekers. After training, the hybrid convolution neural network can achieve targeted talent recommendation. Compared with the hybrid convolution neural network without gradient lifting tree and the convolution neural network with gradient lifting tree, the recall rate and F1 score of this model are improved by 9.78% and 10.1% respectively. This shows that the hybrid convolution neural network algorithm combined with gradient lifting tree can effectively improve the quality of human resource recommendation.

CGAN-based Adversarial Example Defense Method

LI Shi-bao, CAO Da-peng, LIU Jian-hang

2021, 0(07): 65-70.

Asbtract ( 176 )

PDF (831KB) ( 83 )

References | Related Articles | Metrics

Artificial Intelligence has been well applied in many fields at present. However, the classification of neural network model output errors can be achieved by adversarial example. It is of great significance to study how to improve the robustness of the neural network model while taking into account the efficiency of the algorithm operation. To solve the above problems, this paper proposes a defense method Defense-CGAN based on conditional countermeasure generation network. Firstly, the generator of CGAN is used to generate the reconstructed image according to the input noise and label information, and then the MSE is used to extract the image features before and after the reconstruction. The reconstructed image is selected and fed to the classifier for classification, so as to remove the antagonistic perturbation and realize defense of the adversarial example. Finally, a large number of experiments are carried out on the MNIST data set. The experimental results show that the proposed defense method is more versatile, can defend against various kinds of adversarial attacks, and the time consumption is low at the same time. Therefore, this method can be applied to the real scene with extremely strict time requirement.

Intention Recognition and Classification Based on BERT-FNN

ZHENG Xin-yue, REN Jun-chao

2021, 0(07): 71-76.

Asbtract ( 759 )

PDF (989KB) ( 157 )

References | Related Articles | Metrics

Intention recognition classification is an important question in the field of natural language processing. How to understand the user’s intention based on context is a key and difficult problem in intelligent robots and intelligent customer service. Traditional intention recognition classification is mainly based on regularization methods or machine learning methods. However, there are problems of high computational cost and poor generalization ability. In response to the above problems, the design of this paper is based on Google’s BERT pre-training language model to perform context modeling and sentence-level semantic representation of the text, uses the vector corresponding to the ［cls］ token to represent the context of the text, then, extracts the feature of sentences through fully-connected neural network (FNN). In order to make full use of the data, this paper uses the idea of disassembly method to convert the multi-classification problem into multiple binary classification problems. Each time, one category is used as a positive example, and the remaining categories are used as negative examples, which generates multiple two-classification tasks so as to achieve intention classification. Experimental results show that the performance of this method is better than the traditional model, and the accuracy of this method is 94%.

An Improved Algorithm for Small Target Detection Based on SSD

CHENG Kai-qiang, ZHANG Xu, KOU Xu-peng

2021, 0(07): 77-82.

Asbtract ( 241 )

PDF (1567KB) ( 84 )

References | Related Articles | Metrics

Target detection algorithms cannot effectively use the edge texture and semantic information of small targets in the feature map due to low data resolution and noise interference, resulting in poor detection results. To solve this problem, this paper proposes an improved algorithm for small target detection based on SSD. Firstly, common convolution and deep separable convolution are used for synchronous feature learning and fusion, and the information-rich shallow features are obtained. Then the channel and space adaptive weight distribution network is added after the inherent 5 scale feature layer, so that the model pays more attention to the important feature information of the channel and space. Finally, the candidate target frame is subjected to non-maximum suppression screening to obtain the detection result. By comparing the improved method with Faster RCNN, SSD and other methods on the VOC2007 data set, the method reduces the false detection rate of small targets and improves the accuracy of the overall target. The proposed model mAP reaches 78.94%. It is 3.13% higher than the SSD model.

Application of Medical Image Fusion Technology in Tumor Radiotherapy

LI Wen-yu, ZHOU Ling-hong

2021, 0(07): 83-88.

Asbtract ( 143 )

PDF (2285KB) ( 153 )

References | Related Articles | Metrics

Due to the small amount of effective information of conventional medical images, the reference to conventional medical images will affect the judgment of the tumor condition, resulting in low clinical effect of tumor radiotherapy. Therefore, medical image fusion technology is proposed and applied to tumor radiotherapy. This paper uses medical equipment to collect the initial medical image of the patient, pre-processes the medical image through grayscale, negative image and histogram. After the registration space transformation of the medical image, the reference image and the floating image are imported into the registration space to ensure that the mapping point coincides with the origin of the pixel to realize the registration processing of multi-modal medical images. The fusion operator is used to fuse each level and coefficient of the medical image to obtain the reconstructed medical fusion image. The medical image fusion result is output and applied in tumor radiotherapy to determine the location of the target area and adjust the radiation dose so as to achieve radiotherapy for patients. The experimental results show that the application of medical image fusion technology to the actual tumor radiotherapy work can effectively improve the clinical treatment effect of tumors and has high application value.

Vehicle Taillight Detection Method Based on Improved YOLOv3

LI Long, ZHANG Chong-yang

2021, 0(07): 89-94.

Asbtract ( 320 )

PDF (1586KB) ( 124 )

References | Related Articles | Metrics

In the automatic driving scene, the detection of the front taillights is an extensive and significant problem. Darknet53 is the feature extraction network of YOLOv3. It uses five residual units to extract features from the original image, and uses three scale feature map for fusion prediction. The smaller the size is, the stronger the feature expression ability of large target is. Because taillight detection belongs to small target detection, this paper omits the last residual unit of Darknet53, and increases the repetition times of small-scale feature extraction residual unit. Aiming at the problems of K-means clustering algorithm which is difficult to determine K value and sensitive to the initial clustering center, this paper uses K-means+〖KG-*3〗+ clustering algorithm to obtain anchor value and combines IOU distance measurement index. The experimental results show that the accuracy and speed of taillight detection on the improved YOLOv3 network are higher than those before. The mAP is increased from 79.63% to 89.32%, and the detection time of single image is shorten from 0.014 s to 0.01 s. Compared with other mainstream target detection frameworks, the improved YOLOv3 model has superior detection performance.

An Efficient Attribute Revocation Scheme of Supporting Rights Management

LIU Xue-zhen, CUI Yan, DENG Xiao-fei, PENG Jie

2021, 0(07): 95-101.

Asbtract ( 124 )

PDF (1166KB) ( 74 )

References | Related Articles | Metrics

Aiming at the problem of permission determination after attributes revocation existing in the attribute based access control model, the paper proposes an efficient attribute revocation scheme supporting rights management. The scheme implements ciphertext access control by introducing attribute encryption mechanism CP-ABE based on ciphertext policy. On the basis of that, the scheme uses the main disjunctive normal form to express the access tree. Every subset in the main disjunctive normal form is called the minimum attribute set of the restrictive condition that the access subject needs to satisfy to access resource. Once occurring attribute revocation, the scheme considers the relationship between the minimum attribute set and the revoked attributes to determine whether the subject’s access permission is changed. The performance analysis shows that the scheme has high security, which not only can determine the authority after the attribute is revoked, but also can resist collusion attacks.

Cross Platform Access Control of Technology Achievement Data Based on Portal Authentication Technology#br#

ZHAO Wei, QIN Li-han, YUN Chen-chao

2021, 0(07): 102-106.

Asbtract ( 124 )

PDF (754KB) ( 71 )

References | Related Articles | Metrics

In order to avoid the leakage of scientific and technological achievements data, a cross platform access control method based on Portal authentication technology is designed. Portal authentication technology is used to construct the access control model of request access platform and scientific and technological achievements data service platform. When these two platforms pass the visitor identity authentication, the trust degree evaluation and access request authorization are carried out. The service provider platform uses the policy enforcement point (PEP) to complete the collection of user attribute information of the access request and transmit it to the PEP, and uses the recommendation operator to calculate the existing access ask the user’s trust degree, and obtain the user’s trust in the scientific and technological achievements data service platform through the integration operation. The obtained trust degree is transmitted to the policy decision point (PDP). The trust degree is analyzed by the PDP to determine whether the access request is authorized or not, so as to realize the cross platform access control of scientific and technological achievements data. The experimental results show that the method has high efficiency and accuracy, short platform response time and good practicability.

Network Data Stream Classification Based on Concept Drift Detection

ZHANG Heng, JU Shi-guang

2021, 0(07): 107-114.

Asbtract ( 234 )

PDF (2251KB) ( 83 )

References | Related Articles | Metrics

With the rapid development of the Internet environment, the concept drift may exist in the network data stream. The classification of the data stream has changed from the traditional static classification to the dynamic classification. The key of dynamic classification is how to detect the concept drift. In this paper, an adaptive classification algorithm for network data streams based on concept drift detection is proposed. The algorithm detects concept drift by comparing the differences of distribution difference between the data in the sliding window and historical data, and then the window data is oversampled to reduce the imbalance between the samples, finally, the processed data sets are input into OS-ELM classifier for online learning, it updates the classifier to cope with the concept drift in the data stream. In this paper, the proposed algorithm is tested on the MOA experimental platform by using synthetic data sets and real data sets. The results show that the classification accuracy and stability of the algorithm are improved compared with the traditional ensemble learning algorithm, and with the increase of data flow, the advantage of time performance begins to show, which is suitable for complex and changeable network environment.

Crowdsourcing Privacy Protection Method to Local Differential Privacy

ZHAO Long, LONG Shi-gong,

2021, 0(07): 115-119.

Asbtract ( 207 )

PDF (953KB) ( 92 )

References | Related Articles | Metrics

With the development of mobile Internet, significant progress has been made in mobile Internet based on location service (LBS) technology. Aiming at the problem of the risk of leakage of data and information privacy when individual users perform precise positioning, this paper proposes a Geo-indistinguishable disturbance method based on localized differential privacy. Before the user’s real location data information flows out of the client, a Geo-indistinguishable location disturbance method is used to act on the real location to obtain approximate location data. After the server receives it, the secondary area grid map is made, and then differential privacy worker count of the graph is disturbed, and finally the experiment is verified under the spatial range query, and compared with the satisfaction-localized differential privacy disturbance algorithm, the accuracy is increased by 2.7%, and the experiment is compared with the average division of the privacy budget allocation method. Area counting accuracy is improved by 4.57%。

A Method to Generate Features of Mimicry Honeypot Based on Generative Adversarial Networks

LIU Yi-hao

2021, 0(07): 120-126.

Asbtract ( 357 )

PDF (1240KB) ( 110 )

References | Related Articles | Metrics

Mimicry honeypot is a kind of dynamic honeypot technology, which refers to the idea of biological mimicry game, Comprehensively uses the protective color mechanism of "honeypot simulate service features" and the warning color mechanism of "service simulate honeypot features" to carry on the decoy game. Its core strategy is feature generation and evolution. Generative Adversarial Networks(GAN) is a feature generation method, which can make the data generated by generator reach the effect of "mix the spurious with the genuine" through the antagonistic game between generator and discriminator. The idea of antagonistic game is very similar to the idea of mimicry honeypot. In this paper, based on generative adversarial networks, a method for feature generation of mimicry honeypot(MMHP-GAN) is proposed. By optimizing the structure and parameters of MMHP-GAN, new features of honeypot or service can be generated, which are difficult to distinguish between true and false. The experiment shows that through the evolution of feature data generated by this method, the service can effectively resist attacks, and by comparison, the scheme proposed in this paper is better than the existing scheme for feature generation.

Table of Content