Swift and Efficient Solar Energy Prediction Based on an Improved Coati Optimization Algorithm

Accurate short-term solar power forecasting is essential for the stable operation of power systems and the efficient integration of renewable energy, yet remains challenging due to the strong intermittency, volatility, and nonlinearity of solar resources. Although machine learning and deep learning models have been widely adopted, their forecasting performance is highly sensitive to hyperparameter selection, while existing optimization methods often suffer from high computational cost and limited search efficiency. To address these issues, an improved Coati Optimization Algorithm, termed SFCOA, is proposed for efficient hyperparameter optimization. The proposed SFCOA is evaluated across multiple forecasting models, among which Random Forest (RF) and Neural Basis Expansion Analysis for Interpretable Time Series Forecasting (Nbeats) exhibit superior performance and are therefore highlighted as representative cases. Comparative experiments demonstrate that SFCOA significantly accelerates the optimization process, reducing the optimization time of the Nbeats model by 32.81%. In the 15-hour-ahead solar power forecasting task, the optimized RF and Nbeats models achieve mean squared error (MSE) values of 0.0094 and 0.0098, respectively. Furthermore, five repeated experiments and additional tests on different forecasting zones are conducted to further evaluate the stability and robustness of the proposed framework. The Diebold–Mariano test confirms the statistical significance of the forecasting improvements achieved by the proposed SFCOA-based framework.