改进特征选择的光伏功率预测融合算法

Photovoltaic Power Prediction Fusion Algorithm Based on Improved Feature Selection

为提高电站光伏功率预测准确率,该文提出了改进特征选择的融合预测模型。首先耦合包裹式和过滤式方法筛选特征参数;然后根据气象特征分类构建XGBoost、LightGBM和MLP的单一模型;最后使用双隐藏层多层感知器(MLP)构建融合模型进行预测。实验结果表明,通过改进特征选择以及使用对非线性描述能力更佳的MLP融合算法,融合预测模型相比单一模型具有更高的预测准确率以及更强的泛化能力,可较好地满足短期光伏功率预测的需求。

To improve the accuracy of photovoltaic power prediction,a fusion prediction model based on improved feature selection was proposed.Firstly,the Pearson correlation coefficient and the information gain method were combined to select characteristic parameters.Then,the dataset was classified to construct the single model of XGBoost,LightGBM and multilayer perceptron(MLP).Finally,a MLP with two hidden layers was used to build a fusion model.The results show that the fusion prediction model has higher prediction accuracy and stronger generalization ability than the single model,and can better meet the needs of short-term photovoltaic power prediction.