Two-Stage Forecasting of Tcn-Gru Short-Term Load Considering Error Compensation and Real-Time Decomposition

Abstract

With the continuous development of the power system and the growth of load demand, efficient and accurate short-term load forecasting (STLF) provides a reliable guide for power system operation and scheduling. Therefore, this research suggests a two-stage STLF scheme. In the first stage, Improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) is used to process the original load sequence. The subsequence features are extracted by TCN and the initial load sequence prediction is realized based on GRU. Moreover, to eliminate the problem of insufficient adaptability of the model built from the original subsequence in the newly decomposed subsequence, the prediction target is decomposed in real time to boost the generalization ability of the model. In the second stage, the error sequence is constructed with the absolute error of the original load sequence and initial predicted sequence. The adaptive variational mode decomposition (AVMD) is introduced to decompose the error sequence, which is compensated back to the initial predicted sequence by predicting the error sequence through the TCN-GRU hybrid model. The experimental result displays that the scheme is able to better capture the nonlinear and nonstationary properties in the load sequence and has high accuracy in STLF.

Keywords: Short-term load, Error compensation, Real-time decomposition, Two-stage forecasting, TCN-GRU