Time-delay reservoir computing based on VCSEL for short-term load forecasting

Abstract

It is essential for the effective functioning of the power system to guarantee an adequate electricity supply while reducing unnecessary generation. This research presents time delay reservoir computing (TD-RC) system that employs a vertical cavity surface emitting laser (VCSEL) for short-term load forecasting(STLF). The model employs a sliding window of variable length for both its inputs and outputs, using VCSEL as nonlinear node in the reservoir to enhance the model’s nonlinear dynamic characteristics. Additionally, Ridge Regression is employed as a post-processing method to predict power load. The study also investigates the impact of parameters on the performance of the TD-RC system, and its performance is validated using load datasets from Panama and Johor, Malaysia. Compared with existing models, the TD-RC model reduces the mean absolute percentage error (MAPE) by 26% to 39% (35% to 50%) and decreases computation time by 45% to 62% (52% to 67%) across both datasets. The results demonstrates that the proposed model offers advantages such as low complexity, fast convergence, and strong nonlinear expression capabilities, enabling accurate load forecasting.