Chaotic Time Series Forecasting Base on Fuzzy Adaptive PSO for Feedforward Neural Network Training

Short-term electricity demand forecasting for the next hour to several days out is one of the most important tools by which an electric utility plans and dispatches the loading of generating units in order to meet system demand. But there exists chaos in electricity systems to a great extent. Complicated electricity systems are nonlinear systems and the forecasting is very complex in nature and quite hard to solve by conventional algorithm. The inaccuracy or large error in the forecast simply means that load matching is not optimized and consequently the generation and transmission systems are not being operated in an efficient manner. In this paper, feedforward neural network trained by fuzzy adaptive PSO algorithm is proposed for chaotic load time series global prediction. The results which are compared with feedforward neural network trained by Levenberg-Marquardt back-propagation (LMBP) algorithm show much more satisfactory performance, converges quickly towards the optimal position, convergent accuracy and can avoid overfitting in some extent.