A hybrid machine learning framework integrating feature selection and improved sparrow search optimization for electricity consumption forecasting

This is the authors' abstract. We don't add key points for this paper.

Accurate electricity consumption forecasting is essential for power-system planning, grid investment, and renewable electricity management. However, annual forecasting remains challenging because of limited sample size, nonlinear relationships, redundant input information, and the sensitivity of machine learning models to hyperparameter settings. To address these challenges, this study proposes a hybrid mRMR-ISSA-SVR framework for forecasting China’s annual electricity consumption. In this framework, the maximum relevance minimum redundancy (mRMR) algorithm is used to select informative predictors, while an improved sparrow search algorithm (ISSA) incorporating logistic chaotic mapping and Cauchy-Gaussian mutation is developed to optimize the hyperparameters of support vector regression (SVR). Using China’s annual electricity consumption data from 1986 to 2023, the proposed model is compared with a series of benchmark models. The results show that mRMR-ISSA-SVR achieves superior out-of-sample forecasting accuracy, with robustness further verified through rolling-origin and multi-horizon evaluations. Scenario forecasts for 2024–2035 reveal varying electricity consumption growth trajectories across different development pathways.