PCA-LSTM Anomaly Detection and Prediction Method Based on Time Series Power Data

Smart grid puts forward accuracy and reliability requirements for power core data. The abnormal situation of power data still relies on human observation, and traditional neural networks still have large errors in power data prediction. In light of the aforementioned instance, this study suggests an anomaly detection and prediction method for time series power data based on PCA-LSTM. First, in the data preprocessing part, we use statistical methods such as box plots and the 3σ criterion to eliminate abnormal data. Secondly, the multidimensional data is reduced in dimension, and the primary influencing elements are extracted using the principal component analysis (PCA) approach. Thirdly, we predict short-term power data with a network with long short-term memory (LSTM). Finally, we conduct comparative experiments on LSTM, GRU (Gated Recurrent), BP (Back Propagation) and PCA-combined versions of the above three network models. The experimental findings demonstrate the superior prediction accuracy and applicability of the PCA-LSTM model.