A Short-term Load Forecasting Based on Support Vector Regression

Abstract

Load forecasting has become a significant part in national power system strategy management. In this paper, the Support Vector Regression (SVR) for Short-term Load Forecasting (STLF) is presented to predict the primacy of the industry power in the electricity composition. The Support Vector Machine (SVM) is introduced to learn a regression model from training samples with relaxation factors. Our experimental data come from a real-time data acquisition system, which is running for industrial users in a city of Eastern China. As input to the regression model, the feature vector of training samples combines meteorological factors with power system data collected from meters. In order to study the effect of different kernel functions on the accuracy of prediction, this paper respectively tests the linear, polynomial kernel function and Radial Basis Function (RBF). We evaluate the method with two types of predictions, discrete prediction of random samples and continuous prediction of sequential samples. The results indicate that the linear regression model is suitable to forecast with a high fitting degree. However, in the continuous date power prediction, the polynomial kernel function shows preferable prediction ability from the impact of emergencies. KeywordsLoad Forecasting; Support Vector Machine; Machine Learning; Telecommunication; Regression Model