Neuro-fuzzy approaches to short-term electrical load forecasting

We investigate the application of the Takagi-Sugeno fuzzy models to short-term electrical load forecasting problem. Several learning algorithms for these type fuzzy systems are discussed. For identification of the models with linear antecedents the combination of the cluster estimation and ordinary least squares method are applied. For nonlinear antecedent modelling purposes the fuzzy switched ensemble of feedforward neural networks was used. The performance of the models is compared for two-day ahead peak load prediction in the distribution network.