Anomaly Prediction Method for Time Series Data Based on ARIMA and Multilayer LSTM Networks

Time series analysis has consistently captivated the scholarly community. Within the realm of power trading security assurance systems, valuable information can be obtained by analyzing the time series data of power trading. Consequently, this study advocates for a novel time-series anomaly detection approach predicated on the amalgamation of ARIMA and multi-layer LSTM networks. The time series is partitioned into linear and nonlinear components, with the linear segment undergoing processing via the ARIMA module. Subsequently, the processed outcomes are juxtaposed with the original data for residual and white noise detection to derive more suitable nonlinear mode insights for processing within the multi-layer LSTM model. This methodological framework effectively addresses the challenge of abnormal data detection in network systems. Comparative analysis against various historical detection models reveals an average performance enhancement exceeding 5% in key metrics.