Input dimension reduction for load forecasting based on support vector machines

The traditional methods for load forecasting can not supply the required accuracy for the engineering application because we only get limited history data sets and the factors that affect the load forecasting are complex. This paper presents a new framework for the power system short-term load forecasting: firstly, this paper establishes the feature selection model and uses floating search method to find the feature subset; then this paper makes use of the support vector machines to forecast the load and takes full advantage of the SVM to solve the problem with small sample and nonlinear. Hence the accuracy of the estimation result is improved and a better generalization ability is guaranteed. The EUNITE network is employed to demonstrate the validity of the proposed approach.