A novel method based on PCA and LS-SVM for power load forecasting

Load forecasting plays a key role in power system operation and planning. However, the influencing factors of electric power load are very complex and variable. To achieve higher precision, as many of factors as possible are input in the forecast model at the cost of complex computing. Principal components analysis (PCA) is one of multivariate statistic analysis, which achieves parsimony and reduces dimensionality to simplify computation by extracting the smallest number of irrelevant components with little loss of information. In this paper, a new method for load forecasting based on PCA and least squares support vector machines (LS-SVM) is presented. Firstly, principal components are extracted from various factors of load by PCA to be inputs of LS-SVM. Then LS-SVM is applied to train and predict. The model is characterized by all-sided influencing factors and simple computing. Analysis of the experimental results proved that the method proposed achieved greater accuracy and efficiency than conventional LS-SVM.