Research On Two-stage Short-term Power Load Forecasting Based On SSA-VMD

Electricity provides the foundation for social production and life,which places high demands on the safe and stable operation of the power system.Due to the difficulty and high cost of storing electrical energy,the smooth operation of the power system mainly relies on the production scheduling and planning of the grid and various power plants,which requires accurate short-term power load forecasting as a basis.As an important research direction,many experts and scholars have done related work on short-term power load forecasting: traditional statistical methods,single machine learning methods,neural network methods,and hybrid forecasting methods.However,due to the strong volatility of the electricity load itself,coupled with a wide range of complex influencing factors,it is difficult for a single model to tap the deep time-series features in the power load and obtain highly accurate forecasting values.In order to improve the accuracy of short-term load forecasting,this thesis conducts a research on short-term load forecasting based on mode decomposition and hybrid models.The main works of this thesis are as follows.(1)This thesis analyses the impact trend of power load from the time dimension.The impact trend of power load can be divided into: long-term trend,medium and long-term trend,medium and short-term trend,and short-term trend.Various trends are hidden in different components: the longer the time reflected in the trend,the smoother the load component is,which is regarded as a low frequency component;the shorter the time reflected in the trend,the more volatile the load component is,which is regarded as a high frequency component.(2)This thesis proposes a hybrid SSA-VMD-LSTM-MLR prediction model(SVLM).The original load data are decomposed using variational mode decomposition(VMD).To improve the decomposition quality of VMD,a decomposition evaluation criterion Loss is proposed in this thesis,and it is used as the fitness function of the sparrow search algorithm(SSA)to perform key parameter group search for VMD.The highfrequency component of the decomposition is input to a long short-term memory network(LSTM)for forecasting,and the low-frequency component is input to a multiple linear regression model(MLR)for forecasting,and the final forecasting value is obtained by superimposing and reconstructing the forecasting values of each component.The SVLM model has more accurate forecasting results when compared with several models(3)This thesis introduces feature selection engineering.Pearson correlation coefficient(PCC)and maximum information coefficient(MIC)are used to filter out factors with high degree of influence on load,and autocorrelation function sieve(ACF)is used to select node load values with high influence on the load values of nodes to be predicted.By feature selection,the dimensionality of the feature vector is reduced and the size of the neural network is reduced.(4)This thesis proposes a two-stage hybrid forecasting model based on SSA-VMD and feature selection(SVLM-FE).On the basis of SVLM,instead of directly superimposing the forecasting values of each component,they are input as feature values in the second stage for error correction.The forecasting values of each component and the factors selected by the feature selection are fused to construct a feature vector,which is input to the fully connected layer for the second stage of forecasting to obtain the final forecasting value.Compared to the SVLM model,the accuracy of the SVLM-FE model is further improved.