基于改进DBN和K-means的变压器数据异常检测

Abnormal Detection of Transformer Data Based on Improved DBN and K-means

随着智能变电站监测数据的多样化和复杂化,数据异常问题日益突出,由于深度学习可以综合考虑各种因素提高数据异常检测水平,文章提出一种基于改进深度置信网络(deep belief network,DBN)和K-means的变压器数据异常检测方法。首先从智能变电站二次系统中获取变电站监测数据,并从中选取一个变量作为标签变量,计算各变量与标签变量的互信息值,确定输入矩阵;其次,利用改进DBN对输入矩阵进行特征提取;最后,采用K-means对特征向量进行聚类,得到异常数据。其中在结构方面,改进DBN引入了高斯–伯努利受限玻尔兹曼机,使其输入数据不局限于二项分布;在性能方面,采用粒子群算法确定隐含层节点个数,提高DBN的特征提取水平。实验分析表明,该方法在对变压器数据异常检测中具有很好的准确性与鲁棒性。

With the diversification and complexity of monitoring data in smart substations,the problem of data anomalies has become increasingly prominent.Since deep learning can comprehensively consider various factors to improve the level of data anomaly detection,this paper proposes an improved abnormal detection method based on deep belief network(DBN)and K-means for transformer data.Firstly,the substation monitoring data is obtained from the secondary system of the smart substation,and a variable is selected from it as the tag variable,the mutual information value of each variable and the tag variable is calculated,and the input matrix is formed.Secondly,the improved DBN is used to extract features of the input matrix.Finally,the K-means algorithm is used to cluster the feature matrix to obtain abnormal data.Among them,the improved DBN introduces Gauss-Bernoulli restricted Boltzmann machine in structure,so that its input data is not limited to binomial distribution.In terms of performance,particle swarm optimization is used to determine the number of hidden layer nodes and improve the feature extraction level of DBN.Experimental analysis shows that this method has good accuracy and robustness in detecting abnormalities in transformer data.