Hybrid computational intelligence framework for accurate wind power forecasting and grid integration applications.

Accurate wind power forecasting is essential for the reliable operation and large-scale integration of renewable energy into modern power grids. This study develops and systematically evaluates a hybrid computational intelligence framework that integrates advanced machine learning models with nature-inspired optimization algorithms for wind power prediction. CatBoost (CAT), Long Short-Term Memory (LSTM), and Adaptive Neuro-Fuzzy Inference System (ANFIS) models were optimized using Cuckoo Search Optimization (CSO) and the Stochastic Paint Optimizer (SPO) to determine the most effective model-optimizer configuration under variable wind conditions. A comparative analysis demonstrates that the CAT-SPO hybrid model achieved the best predictive performance, yielding a test RMSE of 0.0338 and an R² of 0.984, outperforming alternative configurations. Feature relevance analysis and multicollinearity assessment using the Variance Inflation Factor (VIF) identified hub-height wind speed (100 m) as the dominant predictor (32.5% relative importance; VIF ≈ 3.96), while lower-height wind speed (10 m) was excluded due to high collinearity. Wind gust measurements at 10 m retained substantial explanatory contribution (≈ 19.2% importance; VIF ≈ 4.34), highlighting the role of short-term atmospheric variability in power modeling. The proposed framework enhances forecasting reliability and supports improved grid stability, reserve allocation, renewable energy integration, and data-driven operational planning. These findings advance intelligent energy management systems and sustainable power grid engineering.