基于LSTM-XGboost组合的超短期风电功率预测方法

受数值天气预报信息影响，风电功率变化具有较强的随机波动性，传统单一预测模型精度较低，难以满足现实预测需求。为此，提出基于LSTM-XGboost组合的超短期风电功率预测方法。首先，基于风电场的气象数据，采用皮尔逊相关系数法筛选与风电功率强相关的气象数据，建立风电功率预测模型数据集；然后，将归一化处理的数据集作为LSTM(long short-term memory)和XGboost (extreme gradient boosting)的模型输入，分别构建LSTM和XGboost的超短期风电预测模型，在此基础上，采用误差倒数法对LSTM和XGboost的预测数据进行加权构建组合预测模型；最后，以张家口某示范工程风电场实际运行数据验证组合模型的有效性。结果表明，相较于其他4种单一预测模型，组合模型具有更高的预测精度。

An ultra-short-term wind power prediction method based on LSTM-XGboost combination

Affected by numerical weather prediction, wind power changes have strong stochastic volatility, and the accuracy of traditional single prediction model is low, which is difficult to meet the realistic forecasting demand. To this end, a ultra-short-term wind power prediction method based on the LSTM-XGboost combination is proposed. First, based on the meteorological data collected from wind farms, the Pearson correlation coefficient method is used to screen the meteorological data with strong correlation with wind power and establish the wind power prediction model dataset; then, the normalized dataset is used as the model input of LSTM and XGboost, and the ultra-short-term wind power prediction models of LSTM and XGboost are constructed respectively. Finally, the effectiveness of the combined model is verified by the actual operation data of a demonstration wind farm in Zhangjiakou. The results show that the combined model has higher prediction accuracy compared with the other four single prediction models.