Skip to Main Content
Short-term load forecasting (STLF) is critical for risk management of utility companies in a competitive electricity market, especially for those that are weather-sensitive. Traditional STLF approaches such as time series models and causal models have the advantage of the models' physical interpretations, and the theoretical foundation is well-established, but their application is limited because they are essentially linear models, while electricity load exhibits a nonlinear relationship with the model variables, especially for weather-sensitive load. The knowledge-based auto-regression with exogenous variables (ARX) model proposed in this paper tackles this problem by applying the concept of weather segmentations. This model is tested with an application in the central Texas area. It is shown to produce satisfactory forecast results with a significant improvement over the traditional approach by achieving over 50% error reduction.