中圖分類號(hào)： TP301.6,；TK01+8 文獻(xiàn)標(biāo)識(shí)碼： A DOI： 10.19358/j.issn.2096-5133.2021.05.008 引用格式： 黃家續(xù),，曾獻(xiàn)輝，施陳俊. 基于實(shí)時(shí)數(shù)據(jù)流特征提取的設(shè)備能耗異常識(shí)別算法研究[J].信息技術(shù)與網(wǎng)絡(luò)安全,，2021,，40(5)：45-50.
Research on equipment energy consumption anomaly identification algorithm based on real-time data stream feature extraction
Huang Jiaxu1，Zeng Xianhui1,，2,，Shi Chenjun1
(1.College of Information Science and Technology，Donghua University,，Shanghai 201620,，China； 2.Engineering Research Center of Digitized Textile & Apparel Technology,，Ministry of Education,，Shanghai 201620，China)
Abstract： Energy saving of energy consumption equipment is a very important part of the enterprise energy saving and emission reduction, timely discovery of abnormal energy consumption equipment operation, to reduce unnecessary energy consumption of enterprises is of great significance.In this paper, based on the collected real-time data stream of equipment energy consumption, a new method of equipment energy consumption anomaly recognition and classification based on multi-feature extraction was proposed.Firstly, the characteristic parameters such as low energy consumption time ratio, high energy consumption time quantity and DTW distance were extracted from the sample data. Then, the isolated forest algorithm and K-means clustering algorithm were used to carry out type calibration for each sample data. Finally, the abnormal classification algorithm of equipment energy consumption combining attention mechanism and LSTM was constructed.The experimental results show that the classification accuracy of the algorithm reaches 97.76%, which can effectively identify the abnormal energy consumption of different types of equipment, so as to provide a decision-making basis for enterprises to make timely treatment and reduce the loss of energy consumption.
Key words : feature extraction,；attention mechanism,；LSTM；classification algorithm
目前國內(nèi)外許多研究者對(duì)用電數(shù)據(jù)的異常檢測(cè)進(jìn)行了大量的研究,。黃悅?cè)A等提出了一種基于用電特征分析的無監(jiān)督方式異常檢測(cè)方法[2],，具有較高的準(zhǔn)確性；張春輝等提出了基于小波檢測(cè)電力負(fù)荷異常的方法,，利用ARFIMA統(tǒng)計(jì)方法結(jié)合小波,，能夠快速準(zhǔn)確全面地發(fā)現(xiàn)電力負(fù)荷異常數(shù)據(jù)[3],；趙嫚等利用模糊聚類和孤立森林算法相結(jié)合進(jìn)行異常檢測(cè)[4]；徐瑤等采用卷積神經(jīng)網(wǎng)絡(luò)挖掘用戶時(shí)間序列中的用電規(guī)律,，并通過反向傳播來實(shí)現(xiàn)網(wǎng)絡(luò)參數(shù)的更新,，利用支持向量機(jī)檢測(cè)出異常用電行為[5]；ANGELOS E W S等人使用模糊分類矩陣來改進(jìn)C均值聚類,，歸一化度量距離最大的即為異常用電行為[6],；ARISOY I等基于電力公司長期運(yùn)營的專家知識(shí)，對(duì)用電數(shù)據(jù)的時(shí)間關(guān)聯(lián)關(guān)系進(jìn)行了數(shù)學(xué)建模,，實(shí)現(xiàn)了用戶異常用電量的檢測(cè)[7],。
