基于CNN-LSTM-lightGBM组合的超短期风电功率预测方法

近年来，风电装机规模逐年增加，风电数据采集量呈现规模化增长，面对海量多维、强波动的风电数据，风电功率预测精度仍面临一定的挑战。为提高风电功率预测精度，提出了基于卷积神经网络(convolutional neural networks, CNN)-长短期记忆网络(long short-term memory, LSTM)和梯度提升学习(light gradient boosting machine, lightGBM)组合的超短期风电功率预测方法。首先，分别建立CNN-LSTM和lightGBM的风电功率超短期预测模型。其中，CNN-LSTM模型采用CNN对风电数据集进行特征处理，并将其作为LSTM模型的数据输入，从而建立CNN-LSTM融合的预测模型；然后，采用误差倒数法对CNN-LSTM和lightGBM的预测数据进行加权组合，建立CNN-LSTM-lightGBM组合的预测模型；最后，采用张北曹碾沟风电场的风电数据集，以未来4 h风电功率为预测目标，验证了组合模型的有效性。预测结果表明：相较于其他3种单一模型，组合模型具有更高的预测精度。

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

In recent years, the installed scale of wind power has increased year by year, and the amount of wind power data collection has shown a large-scale growth. Facing the massive multidimensional and strongly fluctuating wind power data, the accuracy of wind power prediction still faces certain challenges. To improve the accuracy of wind power prediction, an ultra-short-term wind power prediction method based on the combination of CNN-LSTM and lightGBM is proposed. First, the ultra-short-term wind power prediction models of CNN-LSTM and lightGBM are established respectively. Among them, the CNN-LSTM model uses CNN to feature the wind power dataset and uses it as the data input of the LSTM model, so as to establish the prediction model of CNN-LSTM fusion; then, the error inverse method is used to combine the prediction data of CNN-LSTM and lightGBM in a weighted way to establish the combined CNN-LSTM-lightGBM prediction model; finally, the wind power dataset of a wind farm in Zhangbei was used to verify the effectiveness of the combined model with the future 4-hour wind power as the prediction target. The prediction results show that the combined model has higher prediction accuracy compared with other three single models.