A hybrid fuzzy, neural network bus load modeling and predication

A hybrid approach utilizing a fuzzy system and artificial neural network fo r bus load forecasting is proposed in this paper. This approach models the behavior of load on those areas where it is primarily a function of tempera ture. Load sequences were broken down into a non-weather sensitive, normal load sequence and a pure weather sensitive load sequence. It has been shown that normal load has a stationary characteristic and can be modeled by back propagation neural networks. The weather sensitive load has been modeled by a set of three fuzzy logic systems trained by least squ are estimation of an optimal fuzzy basis function coefficient. The model was tested with 1994 historical data from the town of Hinton,West Virgina (part of the Appalachian Power Company). The results show an avera ge MAPE (mean absolute percentage error) of 2%! which is comparable with sy stem load forecasting methods reported in the literature.