Solar Power Generation Forecasting Based on Improved Elastic Net Regression

A solar energy power prediction method based on improved elastic net regression is proposed to further enhance the accuracy of ultra-short-term photovoltaic power prediction. The method integrates three key components: Spearman correlation-based feature selection, Elastic Net regression, and hyperparameter optimization via the Tree-structured Parzen Estimator (TPE). Spearman filtering reduces irrelevant features; Elastic Net handles multicollinearity by combining [Formula: see text] and [Formula: see text] regularization; and TPE improves the efficiency and effectiveness of hyperparameter tuning compared to traditional grid search. Using real-world data from a PV station in Heilongjiang Province, China — including Numerical Weather Prediction (NWP) and grid-connected power data — the proposed method demonstrates superior performance. Case studies show a 31.96% reduction in mean absolute error compared to standard Elastic Net, and a 31.84–45.97% improvement over other machine learning models. These results confirm the method’s effectiveness and practicality for PV power forecasting.