An Attribute Associated Isolation Forest Algorithm for Detecting Anomalous Electro-data

With the popularization of the power information collection system, the vast amount of electricity data can be collected from distribution network. It is inevitable to have abnormal data which have an effect on data analysis. This paper proposes an attribute associated isolation forest algorithm for detecting anomalous electro-data. When constructing an isolation tree, attributes are regrouped on the basis of attribute association and the method of generating partitions is changed. Using data samples to traversal the isolation forest, the abnormal data can be identified by its anomaly score and corrected by the modified wavelet neural network method. The method detects the anomalous data quickly and has high accuracy for detecting and correcting anomalous data. A dataset of electricity consumption is used to verify this method. The experimental results indicate that the algorithm has high efficiency and accuracy for the anomaly identification of power data.