Hybrid FMD-TCN-BiLSTM-ECA network for enhanced offshore wind power prediction and grid stability
Anping Wan, Shuai Peng, Khalil AL-Bukhaiti, Xiaomin Cheng, Xiaosheng Ji, Yunsong Ji, Shidong Ma
Abstract
Accurate offshore wind power prediction is vital for grid stability and energy optimization, yet its nonlinearity and volatility challenge forecasting accuracy. This study introduces a novel hybrid network, FMD-TCN-BiLSTM-ECA, to enhance offshore wind power prediction. Feature Mode Decomposition (FMD) extracts modal components from wind power data, enriching input features. Temporal Convolutional Network (TCN) captures local dependencies via stacked convolutional layers, while Bidirectional Long Short-Term Memory (BiLSTM) processes global dependencies bidirectionally. Efficient Channel Attention (ECA) dynamically weights key features, boosting model performance. Using SCADA data from four wind turbines in Yangjiang City, China, collected from March 22, 2022, to March 22, 2023, the proposed model achieved superior results compared to actual data, with R²=0.991, MSE=5.44×10⁻⁴, and reduced error rates (e.g., RMSE=0.022, MAE=0.0155). The Kruskal-Wallis test confirmed the model's novelty against TCN and BiLSTM (P<0.05). These results demonstrate that FMD-TCN-BiLSTM-ECA significantly improves prediction accuracy, supporting stable grid operation and efficient wind energy utilization. This approach offers a robust renewable energy forecasting framework adaptable to broader applications.