Application of neural network and support vector machines to power system short-term load forecasting

Abstract

Power system load was effected by many factors such as weather conditions, holidays, day types, so that the build of short-term load forecasting model is very important. The author analyzed the theory of support vector machine, studied the learning discipline of minimize the structural risk, solved the problem of insufficient training samples better. At the base of support vector machine, The author studied different kernel function and parameter, established the optimal kernel function and parameter, took network training with support vector machines algorithm, established network structure, built a support vector machine short-term load forecasting model; and applied this model to power system's short-term load forecasting. The forecasted results are compared with BP artificial neural network (ANN) methods. The result shows support vector machine short-term load forecasting model is more superiority. © 2010 IEEE.