摘要
为解决现有用电数据异常检测算法准确率低的问题,首先,文章分析了用户用电数据具有时间关联特性、高维度特性,且容易受外部因素影响等特性。其次,基于数据内在特性和LSTM理论,提出了基于数据内在特性和LSTM的用户用电数据异常检测算法。该算法采用有放回的构造数据集策略,构造K个数据集合,采用4层LSTM网络,实现高维数据特征提取,利用两层全连接的隐含层组成的神经网络,实现用户特征数据匹配,采用大概率事件将K个数据集的结果中出现最多的分类作为该节点的分类。通过实验,验证了文章算法比传统算法好,提高了准确率,降低了误报率。
In order to solve the problem of low accuracy of the existing power data anomaly detection algorithm,the characteristics of user power consumption data with time correlation and high dimensionality are analyzed,and user power consumption data are easily affected by external factors.Secondly,based on the intrinsic characteristics of data and LSTM theory,a power data anomaly detection algorithm based on data intrinsic property and LSTM is proposed.The algorithm constructs K data sets by using the constructed dataset strategy with back-conversion.The four-layer LSTM network is used to realize high-dimensional data feature extraction.The neural network composed of two layers of fully connected hidden layers is used to realize user feature data matching.The probability event takes the classification that appears the most in the results of the K data sets as the classification of the node.Experiments show that the proposed algorithm improves the accuracy and reduces the false positive rate compared with the traditional algorithm.
作者
吴刚
Wu Gang(China Telecom Fufu Information Technology Co.,Ltd.,Fuzhou 350000,China)
出处
《无线互联科技》
2019年第10期96-99,112,共5页
Wireless Internet Technology
