Forecasting short term load demand using double seasol arima model

• Main Authors: Mohamed, Norizan, Ahmad, Maizah Hura, Suhartono, Suhartono
• Format: Article
• Published: International Digital Organization for Scientific Information (I D O S I) 2011
• Subjects: CB History of civilization
• Online Access: http://eprints.utm.my/id/eprint/44931/; http://www.idosi.org/wasj/wasj13(1)2011.htm
• Institution: Universiti Teknologi Malaysia

Load demand is a time series data and it is one of the major input factors in economic development especially in a developing COlUltry such as Malaysia. Forecasting load demand with high accuracy is hoped to help the cOlmtry, especially the Malaysian electricity utility company to generate an appropriate load of required power supply which can avoid energy wasting and prevent system failure. A half hourly load demand of Malaysia for one year, from September 01, 2005 to August 31, 2006 measured in Megawatt (MW) is used for this study with the mean absolute percentage error (1.1APE) as a forecasting accuracy. Statistical Analysis System, SAS package was used to analyze the data. The best model was selected based on the mean absolute percentage error (1.1APE) and the theoretical autocorrelation fWlction (ACF) was presented to prove that the best model satisfies the load data. The ARIMA(O,I,1 XO, 1,1 )48(0,1,1 )336 with in-sample MAPE of 0.9906% was selected as the best model for this study. Comparing the forecasting performances by using k-step ahead outsample forecasts and one-step ahead forecasts, we fOWld that the 1.1APE for the one-step ahead out-sample forecasts from any horizon were all less than 1%. In other words it can be concluded that the one-step ahead out-sample forecasts was more accurate. There was a reduction in 1.1APE percentages for all lead time horizons considered, ranging between 89% to 96%. Furthermore a time series plot of out-samples of actual load data, kstep ahead and one-step ahead out-sample forecasts showed that one-step ahead out-sample forecasts followed the actual load data more closely than k-step ahead out-sample forecasts. Therefore we propose that the theoretical ACF must be considered in proving the best model satisfies load demand and that the one-step ahead out-sample forecasts must also be considered in forecasting load, especially in Malaysia load data.