Short Term Load Forecasting Based on WLS-SVR and TGARCH Error Correction Model in Smart Grid

摘要

Smart grid is the main development goal of future power grid while the short-term load forecasting is the significant premise of making management, power supply and trading plan in market circumstance. The forecasting accuracy directly determined the safety and economy of electric system. Support Vector Machines (SVM), as the new machine learning method, has applied successfully to short-termed load forecasting. However, research finds out that the singular points of the initial data have impact on forecasting accuracy. So in this paper, firstly, based on the analysis of SVM, we render Weighted Least Square and Support Vector Regression (WLS-SVR) applying to short-termed load forecasting, which overcomes the disadvantage of singular points. Secondly, we offer Threshold Generalized Autoregressive Conditional Heteroskedasticity (TGARCH) model to construct error prediction model to modify the initial predicted value. Finally, according to the PJM historical data, we get the results showing that the accuracy is greatly improving by implementing our methods which makes our methods founded.%W artykule przedstawiono model przewidywania krótkookresowego obci??enia sieci elektroenergetycznej. W proponowanym rozwi?zaniu wykorzystano metod? SVM (ang. Support Vector Machin?). W celu eliminacji istniej?cego wp?ywu warto?ci syngularnych na dok?adno?? wyniku, zastosowano regresj? ze ?redni? wa?on?. Dodatkowo wykorzystano model TGARCH w okre?laniu b??dów predykcji. Przedstawiono wyniki badań weryfikacyjnych, przeprowadzonych na rzeczywistych danych.