Abstract: Wind power forecasting is important in power system planning and decision-making. However, accurately predicting wind power has always been a challenging problem due to the randomness of meteorological events. To solve this problem, this paper proposes a prediction model based on long short term memory(LSTM)-light gradient boosting machine(LGBM), which combines the advantages of LSTM and LGBM to improve the short-term prediction ability of future wind power. In this paper, the LSTM model is used to capture the timing patterns and trends of wind power and generate a hidden state containing sequence information, and the LGBM model is used as a supplement to the LSTM model to further predict future wind power by receiving the hidden state extracted by the LSTM as input. Experimental results show that the proposed LSTM-LGBM model is superior to other models in global training, which proves the temporal feature extraction ability of LSTM and the predictive performance of LGBM. The application of this model helps to improve the accuracy of wind power generation forecasts and provide effective support for the operation and resource allocation of power systems.

Key words: LSTM, LGBM, wind power forecasting, feature extraction