Research On Dynamic Modeling Method Of Short-term Load Forecasting Based On Fisher Information And Online Support Vector Regression

With the further deepening of power market reforms and the booming development of smart grids(Smart Grids,SG),the effective management and rational allocation of smart grid resources are becoming increasingly important.Short-term load forecasting is an important part of smart grid resources and energy management systems,and accurate forecasting models are the key to achieving high-precision load forecasting.Therefore,the accurate modeling technology of short-term load forecasting has always been the focus and difficulty of research in this field.This paper seeks an accurate modeling technology and method suitable for short-term load forecasting of smart grids,and gives a generally applicable solution to the meteorological factors that affect short-term load forecasting.Based on the study of online Support Vector Regression(OSVR)and the analysis of the impact of meteorological factors on Short-term Load Forecasting(STLF),this paper proposes a dynamic modeling method of online Support Vector Regression,processing of Fisher Information(FI)meteorological factors and Features Selection(FS).The main work done is as follows:First of all,in view of the widely used Support Vector Regression(SVR)model and STLF method of BP neural network,both use offline learning algorithms,and offline samples cannot fully reflect the current characteristics of the system and thus lead to prediction model performance.Degradation,this paper derives the online SVR algorithm based on the KKT(Karush Kuhn Tucker)condition of the SVR model,so that the SVR model can be updated online without repeated offline training,which significantly improves the accuracy and efficiency of model prediction.Secondly,considering the influence of meteorological factors on short-term load forecasting,this paper gives meteorological factors processing method based on Fisher information theory.By calculating the Fisher value of historical meteorological factors to weight the current meteorological factors and meteorological comprehensive index,the real-time effect and cumulative effect of meteorological factors on load forecasting are solved well.In addition,the input feature quantity selection of the prediction model is another issue that needs to be considered in the accurate modeling of Short-term Load Forecasting(STLF),which has not been well solved for a long time.In this paper,Fisher information theory is used to select the input amount of the prediction model,and the expected effect is achieved,which reduces the input amount of the model,improves the prediction accuracy,and improves the prediction speed.Finally,the feature input processed by Fisher information is input to support vector regression(SVR),least squares support vector regression(Least Squares Support Vector Regression,LSSVR)and the online support vector regression(OSVR)proposed in this paper.Among the three prediction models,numerical experimental results show that the online support vector regression model based on Fisher information has higher prediction accuracy and is more suitable for short-term load forecasting,especially ultra-short-term load forecasting.