Investigation on forecast of offshore wind power generation hybrid attention mechanism and bi-directional long short-term memory based on deep learning

Offshore wind energy plays a pivotal role in the global transition to renewable energy, offering vast potential for sustainable and large-scale power generation. Accurate forecasting of wind power is critical for efficient grid integration, resource optimization, and operational planning. This study introduces a novel hybrid model, CNN-BiLSTM-Attention, specifically designed to improve the precision of offshore wind power forecasting. The model synergistically combines convolutional neural networks (CNN) for capturing spatial and local features, bidirectional long short-term memory (BiLSTM) networks for modeling bidirectional temporal dependencies, and an enhanced Attention mechanism with a dynamic offset term (δ_t), which dynamically adjusts attention weights to enable real-time and task-specific feature prioritization. The model is trained and tested using 5920 sets of high-resolution observational data collected from an offshore wind farm located in Fujian Province in the East China Sea, with a rated installed capacity of 280 MW. Comparative analyses against baseline models—including LSTM, BiLSTM, CNN-LSTM, and CNN-BiLSTM—demonstrate that the proposed model reduces RMSE by 13.53 %, increases R² by 8.00 %, lowers MAE by 23.97 %, and decreases MAPE by 28.66 % compared to the CNN-BiLSTM best-performing baseline. The model captures both short-term fluctuations and long-term trends under complex offshore conditions, including wind fluctuations, temperature gradients, atmospheric pressure variability, enhancing forecasting adaptability across multiple time scales. These findings underscore the robustness and accuracy of combining advanced deep learning techniques with enhanced Attention mechanisms for offshore wind power forecasting, providing a powerful tool for facilitating renewable energy integration and sustainable grid management.