A Robust Short-term Wind Power Forecasting Algorithm Based on LSTM-XGBoost Model

The generalization ability of a single wind power prediction model is insufficient. When using this model to predict complex weather conditions, the prediction accuracy in a small range is low, which affects the overall prediction accuracy. To solve these problems and improve the overall robust performance, this paper presents a dynamic weight combination model based on LSTM-XGBoost. First, this paper analyzes the correlation between numerical weather forecast data and historical power data. Then, the pre-processed data sets are used as LSTM model and XGBoost model inputs. Next, the prediction results of single models and extracted features are used as the input of the XGBoost-based dynamic weight combination model to get the expected wind power prediction value. The results show that compared to a single model, the proposed model not only performs better in the case of large power variations but also has lower overall error.