Performance improvement of a hybrid heat pump system with PVT using DNN and optimization algorithms

• System design specifications were optimized using Particle Swarm Optimization (PSO) to minimize life cycle cost (LCC). • Deep Neural Network (DNN)-based flowrate control was developed to enhance real-time operational performance under dynamic conditions. • The optimized system reduced LCC by 10.6 % and improved Return on Investment (ROI) by 3.03 years. • DNN-based control improved system COP by 5.0% and reduced annual energy consumption by 2.8% compared to rule-based control. To meet growing demands for energy-efficient and carbon–neutral buildings, hybrid energy systems combining photovoltaic-thermal and air-to-water heat pump technologies are gaining attention. These systems can simultaneously supply electricity and thermal energy, but their complex interactions, nonlinear dynamics, and load variability require highly accurate, physics-based models for optimal configuration and operation. Traditional approaches—either focused on theoretical design optimization or artificial intelligence-based control methods—often fall short in maintaining system performance under unexpected operating conditions. Without validated models, system reliability, scalability, and comfort assurance remain limited. This study develops a fully validated simulation framework using empirical data from a demonstration building in Busan, Korea, enabling integrated optimization of system design and operation. Economic optimization over the full life cycle was conducted using a particle swarm optimization algorithm, and a control strategy based on deep neural networks was implemented to manage the flowrates in key subsystems, including the air-to-water heat pump, fan coil units, and photovoltaic-thermal solar collectors. Results show that the optimally designed system reduced life cycle cost by 10.6% and improved thermally comfortable hours by 13.1% compared to a conventional setup. Furthermore, the intelligent control method enhanced the system’s coefficient of performance by 5.0%, reduced annual thermal discomfort by 3.2%, and improved economic return. These improvements were primarily attributed to the control algorithm’s ability to adapt flowrates in response to real-time thermal demands, thereby reducing the electricity consumption of circulation pumps while maintaining performance under dynamic conditions. This study demonstrates that integrating simulation-based design with intelligent control can significantly enhance the energy, economic, and comfort performance of hybrid solar-assisted heat pump systems, providing a practical foundation for large-scale deployment in zero-energy-buildings.