Abstract :
In this paper, by presenting a new combination model of artificial neural networks and fuzzy expert system, the initial daily load forecasting of Iran is done more accurately using the knowledge of an expert and the fuzzy expert system for special days (including the new year vacation, official holidays, days before and after official holidays). The effect of temperature is also included in the initial forecasted model of the Kohonen Self-Organizing Map (SOM) neural network for all days of the year, normal, and special. The Mean Absolute Percentage Error (MAPE) values of the years 1380, 1381 and 1382 (in the Iranian solar calendar) are consist of normal and special days are 1.89%, 1.90% and 1.68% respectively. By studying the performance of the designed software based on the present model it is possible to understand the valuable acts of this model. Some of them are: high accuracy in forecasting normal days, high improvement in the forecasting of special days using the fuzzy expert system, high speed of training and forecasting stages, and suitable sensitivity toward temperature, the possibility of accounting other environmental factors such as humidity, cloud overlay, and wind speed.
Keywords :
environmental factors; fuzzy reasoning; load forecasting; Iran; Kohonen self-organizing map neural network; artificial neural networks; cloud overlay; environmental factors; fuzzy expert system; fuzzy inference self-organizing-map; humidity; initial forecasted model; mean absolute percentage error; short term load forecasting; wind speed; Calendars; Forecasting; Load forecasting; Load modeling; Neural networks; Predictive models; Temperature; artificial neural network; fuzzy expert system; kohonen neural network; load forecast;
