Based IGARCH error correction of the PLS-SVR short-term load forecasting

Due to the complexity in the influencing factors of the prediction accuracy, using single forecasting method to improve the prediction accuracy is just impossible in practice. In this chapter, the partial least square (PLS)method was used to diminish the sample input data, which can improve the traditional Support Vector Regression (SVR) for short-time electricity load. Then, there is error sequence between the predictive value and the actual value, and the error sequence was considered as the forecasting data, which has the characteristics of obvious peak and fat tail. Next, Integrated Generalized Autoregressive Conditional Heteroskedasticity (IGARCH) model was used to build the electricity load error predicted model, and modify the original predictive value. Lastly, the forecasting method of this chapter based on PJM historical data was verified. The result shows that the mean absolute percentage error (MAPE) and mean square prediction error (MSPE) are 3.56 % and 1.75 %, respectively. Compared to other traditional predictive value, the model presented in this chapter has higher accuracy, which can be applied to predict the short-term electricity load.