一种针对电力大数据融合与异常检测的改进方法

An improved method for power big data fusion and anomaly detection

针对传统电力大数据异常值检测算法所存在的漏报率高、误报率高、互信息量小、准确率低的问题,提出了一种针对电力大数据融合与异常检测的改进方法,该方法通过深度受限玻尔兹曼机将异构数据统一嵌入到向量空间,而后再结合向量空间内数据来构建正常行为画像,如果新数据与正常行为画像的偏离值超过了一定阈值,那么就可判定为异常状态。将本文算法与其他7种常用算法的异常检测效果进行对比,结果表明:本文算法采用Dee-plearning4j来剖析嵌入式向量数据,进而获取数据画像,其他7种常用算法通过Spark自带算法来获取数据画像,本文算法的互信息量最大,准确率最高,漏报率、误报率在8种算法中处于最低值,由此可见,本文算法具有较高的可行性,可较好地实现电力大数据融合与异常检测,值得推广应用。

Aiming at the problems of high false alarm rate,high false alarm rate,small mutual information and low accuracy in the traditional outlier detection algorithm for large power data,an improved method for power data fusion and outlier detection is proposed,the method uses Restricted Boltzmann machine to embed heterogeneous data into vector space,and then combines the data in vector space to construct normal behavior portraits,if the deviation of the new data from the normal behavior profile exceeds a certain threshold,the abnormal state can be determined.The results show that the proposed algorithm uses Dee-plearning4j to analyze the embedded vector data and obtain the data sketch,the other seven commonly used algorithms are used to acquire data images through Spark's built-in algorithm.The mutual information of this algorithm is the largest,the accuracy rate is the highest,the false alarm rate and the false alarm rate are the lowest among the eight algorithms,the algorithm in this paper has high feasibility,can realize power large data fusion and anomaly detection well,and is worth popularizing and applying.