One-day ahead electric load forecasting with hybrid fuzzy-neural networks

Short-term electrical load forecasting is essential to maintain economic operation of electric power systems. Although several techniques have surfaced in the field of load forecasting, efforts are still being made to develop a model that can achieve a reliable forecast with accurate results. This paper describes the development and implementation of a one-day ahead load forecaster based on a hybrid fuzzy-neural approach. Kohonen's self-organizing feature map with unsupervised learning is used for the classification of daily load patterns. Supervised back-propagation neural networks are then used for learning the temperature-related corrections of the load curves. A post-processing fuzzy controller is employed for fuzzy corrections for unusual load conditions, making the fuzzy-neural model robust in generating accurate predictions on all days of the week.