Machine learning-driven hybrid cooling system for enhanced energy efficiency in multi-unit residential buildings

Abstract

The rising demand for energy-efficient cooling systems in Multi-Unit Residential Buildings (MURBs) presents a challenge, as traditional centralized systems often lead to excessive energy consumption, especially during peak demand periods. Addressing this issue requires innovative solutions that can reduce both the size of the central system and overall energy use, while still maintaining thermal comfort for occupants. This study proposes a novel hybrid cooling system that combines a central cooling system with localized thermoelectric coolers. A key innovation in this research is the use of a Machine Learning (ML) model to predict real-time cooling loads based on factors such as temperature, humidity, solar radiation, and occupancy. The system was evaluated through simulations conducted for a 40-unit MURB in Toronto, Canada, over the summer months. System components included solar evacuated tube collectors, absorption chillers, phase change material storage, and thermoelectric coolers. Cooling load analysis revealed that the building operates near peak capacity for less than 10 % of the time, underscoring the potential for hybrid system optimization. A Machine Learning model was developed to control the operation of the thermoelectric coolers, achieving a high R-squared value (R² = 0.9937) and a SMAPE of 15.87 %, ensuring accurate cooling load predictions. Results showed that both the central and hybrid systems provided acceptable thermal comfort, with PMV and PPD values within acceptable ranges. However, the hybrid system demonstrated higher energy efficiency, achieving a COP of 1.36 compared to 1.28 for the central system. These findings establish the hybrid cooling system, integrated with ML-based control, as a viable and sustainable solution for reducing energy consumption and enhancing cooling performance in residential buildings.