Research On Short-term Power Load Forecasting Model Based On Singular Spectrum Analysis

In the rapid development of electric power industry,power load forecasting has become an indispensable part of power system load scheduling,power equipment maintenance and power planning.With the current proposal of the "Dual Carbon Goal",in order to build a new power system,a large number of new energy sources such as wind power and photovoltaics are continuously connected to the power grid,which has an impact on the power grid,and the stability and volatility of the power load become more complicated.Accurate predictive models are extremely necessary.In this regard,this paper proposes a short-term power load forecasting model based on singular spectrum analysis,and applies singular spectrum analysis(SSA)and improved rat swarm optimization(IRSO)to optimize the gated recurrent unit(GRU)in the short-term power load forecasting model.The main research contents are as follows:(1)The acquired power load data is filled with missing values and normalized to the maximum and minimum values.The Pearson correlation coefficient method and the Spearman correlation coefficient method are used to analyze the correlation between the weather,electricity price and other characteristics of the data set and the power load.The factors affecting the power load are analyzed in detail.(2)The feature sequences and load sequences used for modeling have the characteristics of poor stationarity and strong volatility.In this regard,SSA is used to decompose these sequences.The subsequences obtained by the decomposition represent the main trends,periodic components,and regularities of the original sequence.Compared with the original sequence,the subsequence is less volatile and more stable.Modeling prediction is performed on each group of subsequences separately,and the prediction results of each subsequence are superimposed to obtain the final prediction result,which greatly reduces the difficulty of sequence prediction.(3)Aiming at the fact that the shallow machine learning model cannot well capture the time series between features and power loads,this paper uses GRU as the basic prediction model.As a deep learning model,GRU can effectively learn the original features and power loads.The temporal dependencies of the sequences themselves and each other are crucial to the improvement of prediction accuracy.(4)In order to prevent the threshold and bias of the GRU model from falling into local optimal values,and to further enhance the prediction accuracy of the prediction model,this paper proposes to use IRSO to optimize the threshold and bias of GRU full connection layer.IRSO is an improved algorithm of rat swarm optimization(RSO).It introduces the crossover operator of genetic algorithm,which has stronger optimization ability and is more obvious to promote the improvement of prediction accuracy of GRU.This paper uses real public data in Singapore for simulation modeling,and verifies the validity and performance of each part of the model.After experimental demonstration,the model in this paper can more effectively predict the short-term power load.