Multi-factor offshore short-term wind power prediction based on XGBoost

Fast and accurate wind power prediction is important for grid-connected operation and grid dispatch of offshore wind power. A gated recurrent neural network (GRU) multi-factor prediction model based on extreme gradient boosting (XGBoost) algorithm and particle swarm optimization is proposed to address the problems of random fluctuation of offshore wind power, susceptibility to climate factors, and poor prediction accuracy due to large data volume. The relevance and influence of historical offshore wind power, wind speed, wind direction and temperature are analyzed, ranked in order of importance and downscaled, and then predicted using an optimized gated recurrent neural network. The results are validated with real data from an offshore wind farm and show a decrease in MAE, RMSE and SMAPE. The results show that the MAE, RMSE and SMAPE are all reduced, and that the MAE is still highly accurate when the offshore wind farm is in a different season or in a different location, thus demonstrating the adaptability of the model.