Mid-term Load Forecasting Using a Neuro-Fuzzy system

Forecasting usage of electricity load for producing energy, and needed planning for the future, and choosing needed equipment, attract great attention. The electricity distribution companies need an accurate forecast for producing, distributing, and selling of energy to reduce their expenses. Moreover, the safety of the production would increase too. There are lots of factors that affect the amount of electricity load consumption in a region, such as climate conditions, changing of city structures, and increasing population of a region. Since this forecasting is a nonlinear forecasting with large sizes of parameters, and it depends on lots of factors, which are impossible to predict, the usual classic methods cannot exhibit a great accuracy. Thus, it is essential to use intelligent methods. Notice that the horizon of prediction can include various time periods. Among all, forecasting for a short time period, and middle time period have attracted more attention for producing and distributing companies. Short time period ranges from a few hours to a few days. Middle time period varies from a few days to a few months. In this thesis we have forecasted the electricity load consumption for a middle time period using neural networks in a way that, by the use of available information of recent years, we can forecast the future load consumption. Since there is no explicit relation between climate conditions, and the amount of load consumption, by the use of fuzzy methods, uncertainties between effective events and load consumption can be put in a fuzzy logic. Therefore, using a neuro-fuzzy method can put in account the environmental effects on load consumption. By comparison of neural network, and neuro-fuzzy method, it can result that, the environmental factors such as temperature, and moisture can affect the load consumption. Finally, by the use of simulations it has been revealed that neuro-fuzzy method can give more accurate results of load consumption.