Publications / 2018 Proceedings of the 35th ISARC, Berlin, Germany

A Data-driven Framework to Estimate Saving Potential of Buildings in Demand Response Events

Xinran Yu and Semiha Ergan
Pages 1044-1051 (2018 Proceedings of the 35th ISARC, Berlin, Germany, ISBN 978-3-00-060855-1, ISSN 2413-5844)
Abstract:

In the U.S., the increasing electricity demand gives pressure on the power grids because of its limited capacity to serve demand. Instead of building new power plants to meet the increasing demand, Demand Response (DR) programs incentivize end-consumers to reduce certain electricity demand during certain periods (e.g., peak demand and emergency times). In the current practice, saving potential of buildings, i.e., the amount of electricity that end-consumers can save during an event, is usually determined using the technical specifications of equipment installed, which is unrealistic and leads to over or underestimation of the expected saving potential. In this study, the authors developed a data-driven framework to quantify the electricity saving potential in buildings. The framework was applied to nineteen campus buildings. Several prediction algorithms were used to fit models to the integrated datasets of these buildings, and models were evaluated using four criteria to avoid over-fitting and under-fitting. The best performance of the models resulting in 0.86 of R2 , which represents high capability to quantify the electricity saving potential. The contribution of this study is the proposed data-driven framework, which provides facility operators with reliable tools to accurately quantify saving potential of buildings. The conducted case study using the framework on 19 test buildings showed that facility operators could avoid unnecessary penalties by eliminating them to sign up for unrealistic targets, and help them to gain the most value out of the DR programs by knowing the true potential of their buildings.

Keywords: Facility Management, Demand Response, Electricity Saving Potential, Data-driven, Decision Trees, Ada Boost, Random Forest, Energy Efficiency