A Wind Power Prediction Model Based on Recurrent Highway Network and Multi-Layer Semantic Fusion Attention Mechanism

Accurate prediction of wind turbine power generation is of significant importance for the utilization of wind resources and the operational services of wind farms. However, the complex intermittent, and uncertain nature of wind poses challenges to accurate predictions. In this study, an improved algorithm based on recurrent neural networks is proposed, presenting a wind power prediction model (RHN-MSA) utilizing a recurrent highway network and a multi-layer semantic fusion attention mechanism. The model fully exploits hidden state information at multiple semantic layers. Finally, through numerical calculations and analysis of actual wind field data, the results indicate a significant improvement in the prediction accuracy of the RHN-MSA wind power prediction model compared to models such as GRU, LSTM, and RHN. The research provides a more flexible and comprehensive decision basis for the stable operation of the power system.