Development of a hybrid system based on neural networks and expert systems for short-term electricity demand forecasting

Electrical power is one of the most important commodities in terms of high levels of welfare and comfortable living standards in the modern world. The provision of electricity supply security requires accurate electricity demand forecasts. In this study, a hybrid system using neural networks and expert systems has been developed considering Turkey's electricity market and the seasonal conditions in order toobtain short-term electricity demand forecasts with high degree of accuracy. The new forecast system, which is called EPSIM-NN, estimates daily average per hour demand and 24-hour shape function using two different artificial neural networks. The results from these two separate networks are combined to obtain 24-hour daily demand estimates. Forecast errors are further minimized by an expert system module using correction factors derived from recent demand data. By comparing the estimated values with the actual values for typical Turkish demand scenarios, we conclude that degree of accuracy is quite high for EPSIM-NN generated forecasts.