Thermal response time prediction-based control strategy for radiant floor heating system based on Gaussian process regression

Abstract

One of the main obstacles to wider applications of radiant floor heating systems with great thermal comfort and energy efficiency is the long response time caused by large thermal inertia and little research has been conducted to effectively address this issue. In this study, an optimal control strategy based on a thermal response time prediction model for radiant floor heating systems was proposed by applying the gaussian process regression (GPR) algorithm. First, a comprehensive database of the thermal response time for various scenarios based on different building configurations, radiant floor, and weather characteristics was obtained by parametric simulations after base-case building validation. Then a sensitivity analysis of the response performance of the seven explanatory variables by Pearson correlation coefficients was conducted and a theoretical explanation was provided. Besides, the response time prediction model based on the GPR algorithm with a larger R2 of around 0.96 was obtained by an in-depth comparison with commonly used machine learning algorithms such as multiple linear regression, random forest, and support vector machines. Its accuracy was verified by cross-validation and a 25% database. Finally, an optimal control strategy based on the response time prediction model was proposed which can reduce the response time from 96 ∼ 188 mins to 44 ∼ 75 mins, achieving a reduction of 41% ∼ 64% while keeping the comparative power consumption. Therefore, the proposed optimal control strategy can effectively reduce the thermal response time of radiant floor heating systems and improve the indoor thermal comfort during the thermal response phase without sacrificing power consumption.