Modelling and forecasting monthly electric energy consumption in eastern Saudi Arabia using abductive networks

Abductive network machine learning is proposed as an alternative to the conventional multiple regression analysis method for modelling and forecasting monthly electric energy consumption. The AIM (abductory induction mechanism) is used to model the domestic consumption in the eastern province of Saudi Arabia in terms of key weather parameters and demographic and economy indicators. Models are synthesized by training on data for 5 years and forecasting new data for the sixth year. Compared to regression models previously developed for the same data, AIM models require fewer input parameters, are more accurate and are easier and faster to develop. An AIM model that uses only the mean relative humidity and air temperature gives an average forecasting error of about 5.6% over the year. Our study demonstrates the advantage of using actual values for monthly average weather data rather than means of such averages over a few years.