Short-Term Power Load Forecasting Based on STLSSA-GRU Modeling

Aiming at the problems of high randomness and low prediction accuracy of the time series data of short-term power load, this paper adopts the short-term power load prediction model with the combination of Seasonal-Trend Decomposition Procedure Based on Loess (STL), sparrow optimization algorithm (SSA), and gated recurrent unit (GRU). The model first performs STL time-series decomposition on the raw data, which decomposes the raw data into trend terms, seasonal terms and residual terms. The parameters of the GRU model are optimized by SSA to obtain the optimal number of neurons, and then the three data after STL decomposition are trained and predicted respectively, the learning rate and weights of the GRU model are corrected by using Adam during the training process, and the three predicted data were reconstructed to produce the final prediction. Experimental results confirmed that the method contributes to improving the stability and accuracy of system prediction.