Skip to content
Conference

PatchTST-based short-term load forecasting with sliding-window residual correction for distribution networks

Nov 2026 · International Conference on Optical Signal Processing and Intelligent Algorithms (OSPIA 2026) · 0 citations

Abstract

Short-term load forecasting plays a vital role in day-ahead scheduling, demand response, and energy-storage dispatch in distribution networks. The common recurrent neural networks always present lagged response around the peak-valley transition points and abrupt load change points, and the traditional transformer always incurs huge computational complexity when dealing with long point-wise sequences. In light of these shortcomings, in this paper, we propose a PatchTST-driven multivariate time series forecasting framework with residual sliding-window calibration to forecast the 15-minute load in the regional 10 kV distribution network. Multivariate inputs consist of load, meteorology, calendar, and lagged historical loads as multiple channels. Patch tokenization transforms the continuous point-wise sequences into segment-wise sequences to reduce the attention computation cost while maintaining the local load structure information. We adopt the channel-independent Transformer encoding method to model the heterogeneous variables separately before doing the prediction to avoid the mixing of features in the multivariate sequence learning. On this basis, we further construct a residual sliding-window calibration module to find the systematic errors in peak, flat, valley, and abrupt-change stages. Residual library is established online to correct the initial PatchTST forecast result without any retraining of the backbone network. Experiment results show that our method obtains the following scores: MAE=59.26 kW, RMSE=79.48 kW, MAPE=1.73%, and R²=0.987. Comparing with PatchTST without residual calibration, MAE, RMSE, and MAPE can be decreased by 13.41%, 12.82%, and 16.83%, respectively.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.