Short-term Load Forecasting Of Power Load Characteristic Analysis And Error Analysis

The construction of smart grid improves the efficiency of power grid operation,and plays an important role in energy saving,environmental protection,optimizing the utilization of power s ystem assets and improving the service qualit y of power suppl y.At the same time,it also put s forward higher requirements for the work of all departments of power s ystem.Load forecasting is an important part of the power management s ystem,to provide decision support for multiple departments of power system.It is also the key to the planning,c onstruction,operation,organization,maintenance,and even sales of the entire power grid s ystem.Load forecasting is the key of research of the domestic and foreign in power field.The cause of load is rather complex.It is the result of man y factors.Th e continuous load sampling at different time periods constitutes a time sequence combined with regularit y and randomness.It is the main object of research on load forecasting.Based on the domestic and foreign existing short-term load forecasting technology,for an in-depth anal ysis of the load sequence inherent regularit y and stabilit y,improving the prediction accuracy,this paper constructs the framework of short-term load forecasting b y using frequency domain decomposition,Select the appropriate predi ction model respectivel y,and finall y reconstruct the prediction value of each model as the final prediction result.The specific research contents are as follows:(1)Firstly,according to the anal ysis of load characteristics,the eigenvalues of load sequence are selected as the sample data of short-term load forecasting,and the forecasting date.The wavelet multiresolution anal ysis is used to anal yze the spectrum of sample data,and the low frequency components and high frequency components are obtained.(2)Anal yze the characteristics o f different component sequences,using time series anal ysis to establish ARIMA model to forecast the low frequency component.For high frequency components,the m RMR feature selection algorithm is used to filter out the mai n factors that affect the load changes,considering the factors multiple non-linear relationship,using LS-SVM and the same t ype of day-weighted average idea to build high-frequency component prediction model.(3)The prediction values of the above models are reconstructed as the final prediction results,and at the same time,models of the same sample data are established b y other methods.Experiments show that,compared with other prediction models,the method used in this paper has achieved better predic tion results.Through the anal ysis research and experimental verification of this paper,the "decomposition-reconstruction-prediction" technical solution and quantitative screening of influencing factors provide a effective prediction method for short-term load forecasting.