Short-Term Load Forecasting by Integration of Phase Space Reconstruction, Support Vector Regression and Parameter Tuning System

Forecasting electricity consumption is an important index for system planning, operation and decision making. In order to improve the accuracy of the forecasting, we apply an integrated architecture to optimize the prediction. Based on an integration of two machine learning techniques: artificial fish swarm algorithm search approach based on test-sample error estimate criterion (AFSAS-TEE) and support vector regression (SVR), we proposed a novel forecasting model for future electricity load forecasting. On the one hand the theory of phase space reconstruction (PSR) technique was used for nonlinear dynamic system analysis with the chaotic load series and on the other hand a AFSAS-TEE tuning system is proposed to choose the parameters of SVR automatically in time series prediction. The effectiveness of the proposed model is demonstrated with actual data taken from the Australia Power Grid, and the actual data are compared with the presented and neural networks (NN) methods.