Fault Detection Method for Power Distribution Network Based on Ensemble Learning

Abstract

The State Grid Shanghai Institute of Electrical Science and Technology fault statistics system requires implementation of data analysis and index control capability improvement. This requires the purchase of corresponding hardware and ongoing algorithm and rule optimization in accordance with the application’s actual needs. By enhancing the loss function of the unsupervised LSTM prediction model based on the enhanced experience of the semi-supervised time series anomaly detection algorithm based on LSTM autoencoder and attention mechanism, a semi-supervised LSTM prediction model is proposed in this paper to address the issue that the prediction model is affected by the abnormal data in the training set. This paper further proposes a semi-supervised anomaly detection algorithm based on ensemble learning after verifying the enhanced semi-supervised LSTM prediction model. The algorithm consists of two parts: an anomaly detection model based on LSTM autoencoder and attention mechanism, and a semi-supervised LSTM prediction model. In order to create a semi-supervised model with superior performance, we combined the semi-supervised LSTM prediction model with the semi-supervised LSTM autoencoder model after applying the improved experience of the LSTM autoencoder to the LSTM prediction model. Both algorithms can successfully address the issues of high marking costs and significant time series data reliance in order to produce improved time series anomaly detection results.