A special period peak load forecasting method based on order relations

In order to forecast demand precisely through a year, we should consider various types of days. This paper presents a peak load forecasting method for special periods. In Japan, there are three special periods whose peak loads are usually less than normal days' loads. These peak loads are influenced by annual variation, date and day of the week. These facts request a new method that is different from the forecasting methods for normal days. A proposed method detects order relations of labelled days in the special period based on the days' peak loads. A labelled day is a date labelled day of the week, such as January 1st (Sun) or a date labelled abstracted day of the week, such as January 1st (weekday). The proposed method refines order relations by changing labels from detailed levels to abstracted levels. Final order relations show a structure among labelled days in the special period on peak loads. By using results of final order relations, a peak load of each day in the special period is predicted. Performance of the method which is verified with simulations on actual load data of Tokyo Electric Power Company (TEPCO) is also described.