| Ultra short-term load forecasting is an important research topic In the field of power system.It is of great significance to ensure the reliability of automatic generation control system and the safety monitoring.Compared with the traditional statistical learning and the machine learning methods,the recurrent neural network can effectively improve the accuracy of load forecasting.However,the forecasting performance still needs to be further improved.In order to improve the reliability and accuracy of ultra short term load forecasting,this thesis establishes an improved LSTM and the gating cycle unit(GRU)forecasting model combined with VMD decomposition,and realizes the ultra short term load forecasting in Shaanxi Province.Firstly,aiming at the slow convergence of Adam algorithm in the LSTM neural network,the optimization algorithm with momentum parameters is realized,and the ultra short term load forecasting model of improved LSTM network is constructed,which improves the convergence speed of the traditional LSTM forecasting model.Then,the ultra short term load forecasting model of VMD-LSTM network is constructed.Aiming at the problem of the mode number selection in VMD dccomposition,the dynamic time warping method is used to measure the Gauss of mode distributiotion,and the mode number selcction is realized.The VMD-LSTM model is constructed to realize the load reconstruction and prediction.Finally,by analyzing the structure characteristics of GRU network,the VMD-GRU forecasting model is constructed to realize the ultra short term load forecasting.The experimental results of different dates show that the accuracy of the improved LSTM model is better than that of the traditional LSTM forecasting model.The VMD-LSTM has higher accuracy than the LSTM model by sacrificing the learning time.Duc to the simpler structure of.GRU network,the VMD-GRU model has less learning time,takes into account the accuracy and real-time,and has highcr application value and practical significance. |
PDF Full Text Request