The power load forecasting by kernel PCA

Abstract

We use one year's subset to train the Support Vector Machines (SVM) and the next year's data was used for testing with Kernel Principal Components Analysis (KPCA). This is clearly not optimal for a non-stationary time series such as we have here nevertheless the MAPE of peak load data set with back-propagation neural network [Chuang et al., 1998] is 3.0 and Support Vector Machine is 2.6. © 2010 Springer-Verlag Berlin Heidelberg.