A Hybrid Deep Learning Framework for Wind Speed Prediction with Snake Optimizer and Feature Explainability

Abstract

Renewable energy, especially wind power, is required to reduce greenhouse gas emissions and fossil fuel use. Variable wind patterns and weather make wind energy integration into modern grids difficult. Energy trading, resource planning, and grid stability demand accurate forecasting. This study proposes a hybrid deep learning framework that improves forecasting accuracy and interpretability by combining advanced deep learning (DL) architectures, explainable artificial intelligence (XAI), and metaheuristic optimization. The intricate temporal relationships in wind speed data were captured by training Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), LSTM-GRU hybrid, and Bidirectional LSTM-GRU following data preprocessing and normalization. To enhance transparency, Local Interpretable Model-Agnostic Explanations (LIMEs) were applied, revealing key time-step contributions across three urban datasets (Los Angeles, San Francisco, and San Diego). The framework further incorporates the Snake Optimizer Algorithm (SOA) to optimize hyperparameters such as LSTM units, dropout rate, learning rate, and batch size, ensuring improved training efficiency and reduced forecast error. The model predicted 2020–2040 wind speeds using rolling forecasting; the SOA-optimized LSTM model outperformed baseline and hybrid models, achieving low MSE, RMSE, and MAE and high R2 scores. This proves its accuracy, stability, and adaptability across climates, supporting wind energy prediction and sustainable energy planning.