Learning-Based personal models for joint optimization of thermal comfort and energy consumption in flexible workplaces

Abstract

Due to distinct preferences across individuals, a large proportion of people are not satisfied with the thermal environments of their workplaces. Although recent studies have investigated flexible workplaces to meet the different preferences of individuals, the feasibility of adopting them to real-world environments is limited by the room capacity and potential for extra energy consumption. To address the knowledge gaps, this paper proposes a framework to jointly optimize thermal comfort and energy consumption of buildings through flexible workplaces by considering personal thermal preferences. A joint optimization algorithm that integrates building energy and personal thermal comfort models is developed based on the concept of the Large Neighborhood Search (LNS) algorithm. Analytical energy prediction models are obtained through Nonlinear Polynomial Regression (NPR) and personal thermal comfort models are established using Support Vector Machine Models (SVM). To verify the algorithm, a case study considering two scenarios (with and without existing personal thermal comfort data) is presented. The results indicate the proposed optimization algorithm can significantly improve the thermal comfort of the occupants while saving 22% and 14 % of energy consumption in the two scenarios, respectively. The proposed optimization method is valuable for building managers while adopting the concept of flexible workplaces.