Fuzzy short-term electric load forecasting using Kalman filter

Abstract

A linear time-varying fuzzy load model for solving the short-term electric load forecasting problem is presented. The model utilises a moving window of current values of weather data as well as the recent past history of load and weather data. The parameters of this model are assumed to be fuzzy numbers with a triangular membership function yielding a fuzzy load that has both central and spread values. Both the load and load error are predicted for the following 24 hours on an hourly basis. The forecasting method is based on state space and the Kalman filtering prediction approach in conjunction with fuzzy rule-based logic. The technique is used recursively to estimate the optimal load forecast fuzzy parameters for each hour of the day. The central values of the fuzzy parameters represent the crisp forecast values while the spread values represent the amount of variation of the forecast. The predicted load spread value provides an approximate envelope of the extremes the load possibly takes. The effectiveness of the approach is demonstrated on real load and weather data which show the load forecast with a mean absolute percent error of less than 0.7% and absolute percent error standard deviation of 0.9%.