周期係数モデルによる翌日総需要予測

抄録

This paper reports a method of forecasting the load curve for the following day. Coefficients in the multiple regression analysis used for forecasting the load are estimated by processing data of the previous year. When only a simple multiple regression analysis method is used, an hourly set of coefficients is provided exclusively by data of the corresponding time. To get a better approximation of the hourly coefficients with insufficient data, the proposed method estimates a daily set of the coefficients assuming a daily cyclic change in them. We call our proposed method a cyclic coefficients model.
This assumption does not mean the hourly acute change of the coefficients. The influence of daylight is neutralized in the load data before the estimation of the daily set of coefficients, because of the sharp change in the coefficients of the amount of light at sunset.
Practical data of the Hokkaido area are simulated by both the basic application of the multiple regression analysis and the proposed model. The accuracy of the proposed method is shown by the comparison of results.