Correlation-Based Short-Term Electric Demand Forecasting Using ANFIS Model

Abstract

Prediction of electrical demand in advance plays a key role in power system planning. More specifically for the utilities, market operators, and system aggregators, as it helps them to make critical decisions related to uncertainties in the generation side as well as the demand side. The accuracy of the demand forecasting model is influenced by multiple issues such as uncertainty in demand, weather factors, intermittency of renewable energy sources, etc. A small improvement in system-level and meter level forecasting can contribute to the proper utilization of renewable energy sources and can reduce the system cost. To achieve the better accuracy of demand forecasting, this paper shows an adaptive neuro-fuzzy inference system (ANFIS) for short-term demand forecasting. The ANFIS model can provide good results by considering correlated weather factors, hour of the day, and day type for demand forecasting. The mean absolute percentage error (MAPE), mean absolute error (MAE) and root mean square error (RMSE) of the ANFIS model are obtained for different day types. The results are compared with the ANN model and achieved better accuracy. Also, the impact of anomalous days on forecasting accuracy is analyzed.