A localized NARX Neural Network model for Short-term load forecasting based upon Self-Organizing Mapping

As a routine within the planning and operation of the electrical power system, Short-term load forecasting (STLF) is an essential issue in energy fields. Its prediction accuracy and precision specifically have an effect on the basic safety, stability and economic efficiency of the power system. Moreover, the actual forecasting result also has an impact on operations such as startup and shutdown of the power units, power switching, equipment maintenance, etc. Nonlinear Autoregressive models with Exogenous Input (NARX) Neural Network has been utilized for STLF and proved its effectiveness. This paper proposes a localized Bayesian-Regularization NARX Neural Network model combined with Self-Organizing Mapping (SOM). SOM Neural Network is utilized to extract the meteorological distribution and K-means is utilized to cluster the data. Assessment results depending on half-hourly Australian Grid data and Meteorological data demonstrate that the enhanced model can provide a higher accuracy prediction, which could bring additional economic benefits and extensive social advantages.