A Novel Hybrid Model Based on EMD-Improved TCN-Improved TST for Short-Term Railway Traction Load Forecasting
Da Tan, Zhaohui Tang, Fangyuan Zhou, Yongfang Xie
Abstract
Accurately forecasting the short-term traction load of railways is of great significance for achieving rapid response of railway power quality control equipment and optimizing the operation and scheduling of railway departments. Railway load signals have strong volatility and randomness, which makes it difficult for conventional short-term load forecasting (STLF) models to achieve high-precision forecasting. To this end, this article proposes a novel hybrid deep learning model for improving the accuracy of railway load forecasting. The basic structure of the model includes empirical mode decomposition (EMD), temporal convolutional network (TCN), and time-series Transformer (TST). Specifically, the adaptive median filter (AMF) and EMD algorithm are first used to remove pulse noise from the sequence and decompose the sequence to reduce volatility. Then, use the improved TCN (ITCN) model to extract local features from the sequence, and use the improved TST (ITST) model to extract temporal features. Simulation experiments based on measured railway load data from substations in Hunan, China, show that the signal decomposition algorithm proposed in this article can effectively process railway load signals and achieve better signal decomposition performance compared with commonly used signal decomposition algorithms. The mean absolute error (MAE) and mean-squared error (mse) of the proposed hybrid model are 0.0079 and 0.0348, respectively, which significantly improves the prediction accuracy compared with conventional single forecasting models.