A decomposable multi-period mixing algorithm for long-term load forecasting

With the growing population and the rapid development of technologies such as smart grids, electric vehicles, and renewable energy, accurate long-term load forecasting (LTLF) has become increasingly important. Achieving accurate LTLF is not only crucial for grid planning and operation but also provides a scientific foundation for power dispatch, ensuring efficient energy use and maintaining supply–demand balance. However, the nonlinearity, periodicity, and temporality of load pose multiple challenges to LTLF. In this paper, a long-term load forecasting model based on a decomposable multi-period mixing algorithm is proposed. Firstly, the features that may affect load are analyzed and selected. Secondly, by combining load decomposition methods with the Multilayer Perceptron (MLP) structure, a long-term point forecasting approach based on a decomposable multi-period mixing algorithm is proposed. Thirdly, a probabilistic forecasting method is proposed on the basis of this model. This method utilizes a residual simulation technique to quantify the uncertainty of LTLF, effectively complementing the point forecasting results. Finally, the model’s effectiveness is validated across datasets from a Chinese dataset, Australian electricity markets, and GEFCom2014, consistently demonstrating superior accuracy and computational efficiency.