A hierarchical Bayesian regression model for predicting summer residential electricity demand across the U.S.A. (1st December 2017)

• Record Type: Journal Article
• Authors: Wang, Siyan; Sun, Xun; Lall, Upmanu
• Is Part Of: Energy. Volume 140:Part 1(2017)
• Page Start: 601
• Page End: 611
• Publication Date: 2017-12-01
• Subjects: Hierarchical Bayesian model -- Electricity demand -- Per capita electricity consumption -- Prediction -- Cooling-degree-days -- Clustering; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2017.08.076
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: A hierarchical Bayesian model is developed for predicting monthly residential per capita electricity consumption at the state level across the USA. The summer period is selected to target cooling requirements that are generally directly associated with electricity use compared with winter heating requirements that are derived from a mix of energy sources that has changed over time. Historical monthly electricity consumption data from 1990 to 2013 are used to build a predictive model with a set of corresponding climate and non-climate covariates. A clustering analysis was performed first to identify groups of states that had similar temporal patterns for the cooling-degree-days of each state. Then, a partial pooling model is applied to each cluster to assess the sensitivity of monthly per capita residential electricity demand to each predictor (these are cooling-degree-days, gross domestic product (GDP) per capita, per capita electricity demand from the previous month and previous year, and the residential electricity price). Most of the predictors are significant for most of the states across USA. The sensitivity of residential electricity demand to cooling-degree-days has an identifiable geographic distribution with a stronger relationship in northeastern United States. This predictive model considers interactions between causal factors at large spatial scales and quantifies the model uncertainty. This study can be applied as guide in the future energy management. Highlights: State level electricity demand is modeled using climate, economic, and demographic factors. States with similar cooling-degree-days characteristics are clustered. Local and regional information are jointly modeled to reduce prediction uncertainty. The predictive model can be used to guide future energy management strategy.