SHORT-TERM FORECASTING OF MULTIVARIATE LOAD BASED ON DIGITAL TWIN AND MULTI-MODEL FUSION
Aiming at the problems of poor stability and low accuracy of the forecasting model caused by the volatility and nonlinear of multivariate load,a short-term forecasting method of multivariate load based on digital twin and multi-model fusion is proposed.Firstly,according to the meteorological and load information in the digital twin,the maximum information coefficient(MIC)is used to analyze the coupling characteristics between multi-source data information.Information features are constructed and filtered based on data temporality and periodicity.Secondly,adaptive local iterative filtering(ALIF)is used to decompose the historical multivariate load data to obtain intrinsic mode function(IMF)at different frequencies.Then,kernel extreme learning machine(KELM)and bi-directional long short-term memory network(BiLSTM)are used to forecast high-frequency and low-frequency load components.The short-term forecasting results of the initial load are obtained by fusing and reconstructing high-frequency and low-frequency components.Finally,the final load forecasting results are obtained by compensating the initial forecasting results with the digital twin.Compared with the single forecasting model and the non-digital twin forecasting model,the proposed method can effectively deal with the fluctuation and nonlinearity of multivariate load,and has better stability and accuracy.
digital twinload forecastingadaptive filteringnew power systemkernel extreme learning machinebidirectional long short-term memory network
万方数据
维普
