Utilizing nonlinear models in power systems for short-term load forecasting

This paper investigates the application of nonlinear forecasting models in Short-term Load Forecasting (STLF). Given the limitations of traditional linear models in dealing with complex load data, nonlinear forecasting models are invoked to enhance forecasting accuracy. Multilayer Perceptron (MLP) and Extreme Learning Machine (ELM) are constructed and validated, and the initial use of ensemble learning idea to integrate single nonlinear models is proposed. The empirical analyses showed that the nonlinear models outperform the linear ones in terms of prediction accuracy and stability. The results demonstrated that MLP-SDLW models with 10 variables performed best with Mean Average Percentage Error (MAPE) 5.047%. Meanwhile the ensemble models reflected the advantage of higher accuracy in observation.