基于CNN-VMD-PCA特征融合的光伏发电功率预测研究

Study on photovoltaic power generation forecasting based on CNN-VMD-PCA feature fusion

为了较为准确地预测光伏发电功率,提出一种特征融合的功率预测模型。模型首先使用一维卷积神经网络(1-dimensional convolutional neural network,1D-CNN)提取光伏光电数据深度特征,然后用变分模态分解方法(Variational mode decomposition,VMD)分解数据原特征,再把分解后的特征和深度特征融合,用主成分分析法(Principal component analysis,PCA)提取融合后特征的主成分,最后用Xgboost(Extreme gradient boost,Xgboost)模型进行功率预测。根据对所提模型的实测评估,并与其他预测模型对比,得出基于CNN-VMD-PCA特征融合的预测模型具有较高预测精度,其拟合优度达0.932,能够得到更可靠的功率预测结果。

In order to accurately predict the power of photovoltaic power generation,a power prediction model based on feature fusion was proposed,in which one-dimensional convolutional neural network(1D-CNN)model was used to extract the features of photovoltaic data,and the original characteristics of data were decomposed by variational mode decomposition(VMD),and then the decomposition features were integrated.The main component of integrated characteristic was extracted by principal component analysis(PCA),and power prediction was conducted finally by extreme gradient boost(Xgboost).According to actual assessment of model mentioned,through comparison with other prediction models,it is concluded that the prediction model based on CNN-VMD-PCA feature fusion has higher prediction accuracy with regression fittness 0.932,which means more reliable power prediction.