基于线性判别分析和密度峰值聚类的异常用电模式检测

Anomaly Detection for Power Consumption Patterns Based on Linear Discriminant Analysis and Density Peak Clustering

现有的异常用电检测方法存在未考虑电力用户的位置信息、模型参数选取困难的问题。据此,提出了一种基于线性判别分析(LDA)和密度峰值(DPeaks)聚类的双判据无监督异常用电检测模型。该模型遵循“特征构造—维度规约—聚类—异常检测”的流程,借助聚类算法将用电模式类别不同的用户进行分类后再检测。在维度规约模块,使用线性判别分析将用户的台区号输入检测模型,提升了模型的检出率和精确率;在异常检测模块,设置双判据检测标准,减小了模型对参数摄动的敏感程度。采用该模型检测爱尔兰智能电表数据,结果表明用户位置信息的引入可以提高异常检测模型的准确度。

Existing methods for anomaly power consumption detection have the problems of ignoring the location information of power users and difficulty in selecting model parameters. On this basis, a double-criterion unsupervised anomaly power consumption detection model based on linear discriminant analysis(LDA) and density peaks(DPeaks) clustering is proposed. This model follows the process of“feature construction—dimension reduction—clustering—anomaly detection”. Users with different power consumption patterns are classified by clustering algorithm and then detected. In the dimension reduction module, LDA is used to input the station area codes of users into the detection model, which improves the detection rate and accuracy of the model.In the anomaly detection module, a double-criteria detection standard is set to reduce the sensitivity of the model to the parameter perturbation. The proposed model is used to detect Irish smart meter data, and the results show that the introduction of location information of users can improve the accuracy of the anomaly detection model.