Electricity consumption prediction - A case study of a university library

Abstract

Estimating and controlling electricity consumption (EC) for public facilities is important for the efficient management of public use facilities. Of particular interest is EC and understanding the patterns that influence it. Therefore, the focus of research is on an effective predictive model, to objectively consider the usage characteristics of electrical structures, to accurately estimate the relationship between changes in EC and future trends in electricity demands. EC in public buildings is typically cyclic with variation in usage throughout the year, presenting a unique modeling challenge. This study takes the library of a public University (YunTech) as the subject, using the total EC and monthly EC data of the library over five years to build a predictive model using Fourier's theory. Firstly, the study estimated the EC of the library by using simple regression to verify whether the historical data has a significant trend of increasing EC. The results showed that with time series regression analysis, the increasing trend in EC had no statistically significant relationship. Whereas, the Fourier series model, with three-years of monthly data provided the best EC prediction ability with performance comparable with more complex machine learning based methods. A robust model of EC will help to develop and test strategies for reducing the environmental impact of public use buildings with unique cyclic usage patterns.