Modeling and forecasting hourly electric load by multiple linear regression with interactions

Abstract

Short-term electric load modeling and forecasting has been intensively studied during the past 50 years. With the emerging development of smart grid technologies, demand side management (DSM) starts to attract the attention of electric utilities again. To perform a decent DSM, beyond when and how much the demand will be, the utilities are facing another question: why is the electricity being consumed? In other words, what are the factors driving the fluctuation of the electric load at a particular time period? Understanding this issue can also be beneficial for the electric load forecasting with the purpose of energy purchase. This paper proposes a modern treatment of a classic technique, multiple linear regression, to model the hourly demand and investigate the causality of the consumption of electric energy. Various interactions are discovered, discussed, tested, and interpreted in this paper. The proposed approach has been used to generate the 3-year hourly energy demand forecast for a US utility.