A novel neuro-fuzzy based self-correcting online electric load forecasting model

Abstract

The paper presents a neuro-fuzzy short-term load forecasting (STLF) model. The proposed ANN function approximator models the relationships between the system hourly peak load and system variables affecting it, namely, weather and temperature variations, type and time of day, the inherent parameters of historical load patterns such as trend, cyclic oscillations, regular seasonal and irregular ‘special’ events. The load predictor forecasting input vector was extended to account for most of the input dominant variables affecting the short-term forecast load. The model utilizes a preprocessor for input vector generation and priority classifications using historical load and system data. A postprocessor fuzzy logic block provides error correction and data filtering and online tuning and adjustment of electric load forecast data.