To increase short-term power load forecasting accuracy,this paper proposes a weighted combination prediction strategy.Firstly,in order to reduce the instability of load data,the variational mode decomposition(VMD)is used to decompose the load data into three feature mode components:high-frequency,low-frequency,and residual.Secondly,considering the temporal characteristics of the load data,an adaptive error importance quantification function is designed based on the principle of exponential weighting,and the objective function and constraint conditions of the improved optimal weighting method are designed based on the mean-square prediction error of the historical load data within the time window,in order to achieve accurate weight variation of the submodels.Finally,XGBoost and CNN-LSTM models are selected for the high-frequency components with strong fluctuations,and the improved optimal weighting method is used for combination prediction.The MLR model is used to predict the low-frequency components,and the LSTM model is used to predict the residual components.By superimposing the prediction results of each mode component,accurate prediction of short-term load data is achieved.The experimental results show that the average absolute percentage error of the combined model using this strategy is 4.18%.Compared to the combined model using existing combination strategies,the average absolute percentage prediction error is reduced by 0.87%.
short-term load forecastingVMDimproved optimal weighting methodcombined model
NETL
NSTL
万方数据
