A Two-phase SVR Model with Information Reduction for Forecasting Electric Quantity at Gateway

Abstract

With the rapid developments of electricity market it is very important that prediction of electric quantity helps to have insight into the characteristics of power consumption of different customers so as to make a correct decision of power distribution for the optimal resource scheduling and planning. Based on an accurate forecasting, operation center of power grid can effectively decrease the cost of power generation and simultaneously increase the profit of power services. Although there are many techniques available for short-term electric quantity forecasting, mid-term electric quantity forecasting is relatively little. In this paper, we proposed a new two-phase model combining Support Vector Regression (SVR) with information reduction using L1-norm SVR (L1-SVR) in order to improve the accuracy, efficiency and interpretability of prediction. The former is used to obtain the optimal multiplier vectors and forecast electric quantity at gateway with the reduced data set. At the same time, the latter employs a new information reduction approach for regression to find the important variables and remove the less important variables. The results of our experiment and comparison with the single SVR approach showed that the proposed model improved forecasting accuracy, interpretability of each variable or feature and efficiency of electric quantity at gateway on the independent test set.