ISSN 0253-2778

CN 34-1054/N

2019 Vol. 49, No. 2

Display Method:
Gas outburst prediction based on rough set and particle swarm optimization support vector machine
LIU Haibo, QIAN Wei, WANG Fuzhong
2019, 49(2): 87-92. doi: 10.3969/j.issn.0253-2778.2019.02.001
In view of coal and gas outburst intensity forecast problems in coal mines, on the basis of comprehensive influence factors of gas outburst, a decision table of gas outburst intensity was established by employing the rough set theory and support vector machine, and selecting coal thickness variations, geological structures, coefficient of the solid coal, roadway pressure, gas change, gas desorption value of drilling chip, and ten main influence. Using the attribute reduction algorithm in rough set theory to eliminate redundant information, and particle swarm optimization to optimize parameters of Support Vector Machine, the main control factors of gas outburst were mapped to high-dimensional space through kernel function, and the nonlinear relationship between main control factors and intensity of gas outburst was fitted. A gas outburst prediction model based on rough set theory and particle swarm optimization support vector machine was established. A typical example of gas outburst was selected as a study sample, and a prominent example of a mine in Henan was used as a test sample for prediction. The experimental results show that the model can meet the requirements of gas outburst prediction, with the prediction results being consistent with the actual results.
Top-k query of relational database based on CP-net
LUAN Yanhong, SUN Xuejiao
2019, 49(2): 93-99. doi: 10.3969/j.issn.0253-2778.2019.02.002
CP-net is a simple and intuitive graphical preference representation tool that can be used to describe the qualitative preference relationships implied in relatively tight, intuitive, structured conditional preference information. Full-featured qualitative decision-making with multiple dependencies between attributes in complete cases. Top-k queries are designed to retrieve the top k results that meet user requirements, thereby improving search efficiency. Aiming at implement Top-k queries with relational databases that have CP-net preferences. First, CP-net is induced into multiple tables for representation and storage, Then, the traditional Pareto composition is extended into the model so as to maintain a strict partial order relationship between preferences. Finally, based on “Lattice” theory, the Top-k query of relational database based on CP-net preference has been implemented.
Equipment identification from power load profile
2019, 49(2): 100-104. doi: 10.3969/j.issn.0253-2778.2019.02.003
The power load profile of the equipment varies with time, and it is essentially time series data. A new ensemble learning method for identifying electrical equipment through load power profile is proposed, which uses convolution neural network (CNN) as base learner to train the multi-granular load profile to improve the accuracy of classification. First, the raw data with different granularities are divided and some different new data sets are obtained. Then, these new data sets were used to train different base learners and get the weight of different base learners according to the accuracy of validation sets. In the testing process, testing data are divided based on different granularities in the same way as the training data are fed into base learners and the final results are obtained by weighting the output of each base learner. The proposed model are compared with a single CNN model on the electrical equipment load data. The experimental results show that the proposed method has higher accuracy in the identification of electrical equipment.
Facial expression recognition based on fusion of deep learning and dense SIFT
PENG Yuqing, WANG Weihua, LIU Xuan, ZHAO Xiaosong, WEI Ming
2019, 49(2): 105-111. doi: 10.3969/j.issn.0253-2778.2019.02.004
With the wide application of facial expression recognition in the field of human-computer interaction, accurate and efficient expression recognition methods are of particular important. A hybrid model that combines the convolutional neural network with Dense SIFT features is proposed. The network structure used in the hybrid model is improved in the idea of depth-separable convolutional neural network MobileNet. Based on the separation of channel convolution ( depth convolution)and space convolution (point convolution), the multi-scale convolution kernel is used in the point convolution part of the MobileNet structure, which ensures the diversity and subtleness of the extracted features and is more suitable for facial expression feature extraction, and the introduction of DenseNet network structure ideas improve the performance of the network structure. Using Dense SIFT's 128-dimension descriptors to provide greater advantages for feature descriptions, the improved MobileNet network is integrated with its fully connected layer, and the Eltwise layer is used to compare the elements of the fully connected layer, taking the maximum value to ensure the diversity of features, as well as greater representation. Using this hybrid model on FER2013 and JAFFE face expression data sets, the recognition rate can reach 73.2% and 96.5%.
Research on product reviews hot spot discovery algorithm based on MapReduce
SU Hao, LIU Qicheng, MU Chunxiao
2019, 49(2): 112-118. doi: 10.3969/j.issn.0253-2778.2019.02.005
A parallel algorithm based on MapReduce framework for finding hot spots from commodity reviews (PR-HD algorithm) is proposed. The PR-HD algorithm uses crawler technology to extract an electricity supplier. A review data set is generated from the review data of a popular mobile phone under the platform, and the weight of the feature words is calculated by the TF-IDF algorithm. The final weights of the feature words are obtained by adding position weights of the feature words, and a vector space model (VSM) calculation is established. The similarity of different comment sentences is combined using Canopy algorithm and K-means algorithm to realize hot spot discovery from commodity reviews. This allows product developers to obtain more direct and effective suggestions and feedback.
Comparative study of short-term electrical load forecast models
YAN Huifeng, HUANG Dingjiang, XIE Yao, CHENG Xiao, XIE Jiyang, ZHU Xiaomeng, MA Zhanyu
2019, 49(2): 119-124. doi: 10.3969/j.issn.0253-2778.2019.02.006
In order to solve the problems of electrical load prediction performance improvement, more efforts are being made to apply artificial intelligence methods in electrical load prediction. Using the electricity load data of Hunan Province from 2014 to 2017, the autoregressive (AR) model, BP neural network (BPNN), and exponential smoothing (ES) model were compared in terms of their performance of predicting both daily and monthly electrical load, respectively, and analyze the differences among the aforementioned three models. According to the experimental results, it was that the autoregressive model performs better in daily predictions than the other two models, while the exponential smoothness model gives better monthly predictions.
Optimizing design of PID controller with time varying undetectable changes based on multiple reference points
LI Erchao, ZHAO Yumeng
2019, 49(2): 125-131. doi: 10.3969/j.issn.0253-2778.2019.02.007
A kind of time-varying parameter non measurable dynamic multi-objective optimization genetic algorithm based on reference points is proposed for the design of PID controller with variable parameters and multiple objective functions. The algorithm is a dynamic multi-objective optimization genetic algorithm which joins the reference point and local search and population updating mechanism to optimize the parameters of the PID controller under the conditions of different environment and undetectable changes. In order to verify the effectiveness of the algorithm, a typical test function is used to compare this algorithm with the DNSGA2-A algorithm. The thought of PID controller design is as follows. Firstly, a dynamic multi-objective model of a PID controller is established, and the designing PID controller tuning problem is formulated as a dynamic multi-objective optimization problem. Secondly, reference points are established and then a dominant Pareto dominance relationship based on reference points is defined. In addition, the population is processed through a local search and archive update. And the dynamic multi-objective optimization algorithm is used to optimize the PID parameters. Finally, the method is applied to the optimization problem of the diesel engine. To shorten the error and variance, as the optimization goal, the three parameters of the PID controller are optimized. The dynamic multi-objective optimization evolutionary algorithm for PID controller parameter optimization is validated effectively.
A mobility evaluation routing protocol for delay tolerant networks
ZHANG Fuquang, WU Yin, YANG Xubing
2019, 49(2): 132-137. doi: 10.3969/j.issn.0253-2778.2019.02.008
DTN (delay tolerant networks) has the characteristic of communication-connection intermittence. Currently, more and more researchers focus on mobility and show that mobility is important to network connection and coverage in mobile wireless networks. A routing protocol that uses the mobility of nodes for evaluating the possibility of delivery is proposed. The evaluation is based only on the node speed without the requirement of any other information. Simulation results show that the proposed scheme performs better than those in related works.
Trend information for time series classification
LIN Qianhong, WANG Zhihai, YUAN Jidong, ZHANG Wei
2019, 49(2): 138-148. doi: 10.3969/j.issn.0253-2778.2019.02.009
One of most important parts of time series data analysis is to choose the appropriate similarity measurement. Among all similarity measurements, the longest common subsequence is a commonly used and effective method. However, the original method only measures the numerical differences of point-to-point sequences, which neglects the trend of the changing sequence. Therefore, a time series discretization method based on the trend information is proposed and the longest common subsequence is employed to carry out similarity measurements. This method can measure time series trend information well. In addition, it is linearly combined with the point-to-point comparison function. In contrast to well-known measures from the literature, the proposed method can take both the trend information of time series and point-to-point comparison function into consideration. The new similarity measurement is used in classification with the nearest neighbor rule. In order to provide a comprehensive comparison, a set of experiments have been conducted, testing its effectiveness on 42 real time series. The experimental results show that our method can effectively improve the accuracy rate of time series classification.
A modeling socialization point process sequence prediction algorithm
JIANG Haiyang, WANG Li
2019, 49(2): 149-158. doi: 10.3969/j.issn.0253-2778.2019.02.010
Predicting the type and time of the next event according to the sequence data is a subject worth studying.At present, the point process intensity function only considers the background knowledge and historical influence from the time dimension, and has no influence on the social relations from the spatial dimension.Aiming at this problem, a sequence prediction algorithm (SPSP algorithm) is proposed based on the spatio-temporal deep network.In this model, firstly the background knowledge and historical influence of the intensity function are modeled with the dual LSTM (long short-term memory).Then the output of two LSTMs are combined by the union layer to generate the vector representation of event type and time.Finally, the influence of social networks on the spatial dimension is added to optimize the intensity function.Through multiple training of the Spatio-temporal deep neural network, the optimal network model is obtained.Sina weibo data sets are used to verify the validity of the algorithm, and it has been proven by experiments that the proposed algorithm can predict the event type and time efficiently accurately.
Mixed linear matrix completion model based on auxiliary information
2019, 49(2): 159-165. doi: 10.3969/j.issn.0253-2778.2019.02.011
The matrix completion technology has been applied in many fields in recent years. A matrix completion model that mixes bilinear and unilateral linear relationship is proposed, considering the correlation between row information and column information and their respective characteristics, so that the mixed linear model can approximate the original matrix entries. The convergence of using the ADMM algorithm to solve the convex optimization problem is proved, and makes two sets of experiments with synthetic datasets and real datasets, which proves that the proposed method is more effective compared with the existing model using auxiliary information, whose error under RMSE evaluation standard has been reduced by more than 25%.
XSS attack detection based on Bayesian network
WANG Peichao, ZHOU Yun, ZHU Cheng, ZHANG Weiming
2019, 49(2): 166-172. doi: 10.3969/j.issn.0253-2778.2019.02.012
Cross-site scripting (XSS) attack is one of the most serious cyber-attacks. Traditional XSS detection methods mainly focus on the vulnerability itself, relying on static analysis and dynamic analysis, which appear weak in defending the flood of various kinds of payloads. An XSS attack detection method is proposed based on the Bayesian network, in which the nodes are acquired with domain knowledge. The ontology constructed with domain knowledge provides a good basis for feature selection, and 17 features have been abstracted from it; besides, malicious IPs and malicious domain names collected from open source channels make effective complement rules for the detection of new attacks. To validate the proposed method, experiments were conducted on a collected real-world dataset about XSS attacks. The results show that the proposed method could maintain a detection accuracy of above 90%.