MTS-LSTDM: Multi-Time-Scale Long Short-Term Double Memory for power load forecasting

Abstract

As reducing greenhouse emissions and carbon footprint have drawn more and more attention, predicting the power load of distribution transformers in an intelligent way is critical for electric safety and equipment protection. It is essential to perform accurate regression and generate stationary residual series in critical fields such as power load forecasting. In this paper, we introduce a fusion operator to combine critical historical information of different kinds and propose an ensemble Long Short-Term Double Memory based on Multiple Time Scale (MTS-LSTDM) for power load prediction in smart grid. Compared with some state-of-the-art and existing regression methods, both excellent accuracy and residual series stationarity are shown with MTS-LSTDM, which is conducive to the stability and robustness of power load forecast and is easier to popularize in real-world application scenarios. In addition, the comprehensible reason for generating stationary residual series is given, and generalized applications are discussed in detail. This work not only conducts a complete model research and analysis on the smart grid, but also has great reference significance for the regression of multi-time-scale time series data.