基于改进VMD和聚类权值共享的负荷预测

Load forecasting based on improved VMD and clustering weight sharing

针对常见的数据分解加预测算法的组合负荷预测方法具有参数多、训练慢以及模态间共有信息不能有效提取的问题,提出了一种基于改进变分模态分解(variational mode decomposition,VMD)和聚类权值共享的负荷预测模型。模型首先引入互相关函数以寻找VMD的最优分解K值,然后利用K-means将分解后的模态分量进行聚类以突出模态分量的时序特征,最后利用权值共享思想对聚类后的分量进行快速准确的建模预测。仿真结果表明:该模型的平均绝对百分比误差(mean absolute percentage error,MAPE)和均方根误差(root mean square error,RMSE)分别为5.29%和986.50,与传统的单一预测模型相比,所提算法的MAPE和RMSE平均降低了7.50%和982.41;与常见的数据分解加预测算法的组合相比,所提算法的MAPE和RMSE平均降低了3.09%和268.93,训练速度也有一定提升。

A load prediction model based on improved variational mode decomposition(VMD)and clustering weight sharing is proposed for the problems that the common combined load prediction method of data decomposition plus prediction algorithm has many parameters,slow training and ineffective extraction of common information between modes.The model first introduces the cross-correlation function to find the optimal decomposition K value of VMD,then uses K-means to cluster the decomposed modal components to highlight the temporal characteristics of the modal components,and finally uses the idea of weight sharing to model the clustered components for fast and accurate prediction.The simulation results show that the mean absolute error and root mean square error of the model are 5.29%and 986.50,respectively,which are 7.50%and 982.41 lower on average compared with the traditional single prediction model.Compared with the common combination of data decomposition plus prediction algorithm,the mean absolute error and root mean square error of the proposed algorithm are 3.09%and 268.93 lower on average.There is also a certain improvement in the training speed.