Abstract
This paper describes a load profile prediction algorithm that has recently been developed for use in an optimal operation of heat storage systems. It is highly desirable to devise an on-line prediction technique such that the difference between predicted load and actual load is as small as possible. To accomplish this, an ARIMA model is adopted. The modelling is first done for the past load data. Next, this model is used to predict the load profiles for the next day. The load profiles are updated every hour on the basis of the newly obtained load data.
When this on-line prediction algorithm is executed on a computer in field operation, some of the more important questions are left open:
1) What kind of evaluations should be considered for the model validity?
2) After air-conditioning system starts, how many data should be required to predict the load profile?
3) Half days and off days have different load profiles from ordinary days. How should the load profile on these days be predicted?
4) Does the forecasted highest ambient temperature for the next day become available to update the predicted load profile?
In this paper, the solutions to these questions are considered. Performance of the load profile prediction algorithm is assessed by recording cooling load and ambient temperature history (1987) at The Tokyo site. Results of this test show generally good agreement between actual and predicted loads. Errors are typically less than 10% on ordinary days; however, somewhat higher errors occur on half days and off days (errors being on the order of 20%).
