EHP-AHU 시스템의 급기 온도 최적화 알고리즘에 대한 시뮬레이션 기반 에너지 절감 효과 평가

Abstract

This study presents the development and simulation-based validation of a real-time optimization algorithm for the supply air temperature setpoint in an Electric Heat Pump – Air Handling Unit (EHP-AHU) system. The objective is to minimize overall energy consumption while ensuring compliance with indoor thermal comfort standards. Traditional HVAC systems often rely on fixed setpoints for supply air temperature, which can lead to suboptimal energy use, particularly under varying internal and external load conditions. To address this limitation, the proposed control algorithm dynamically adjusts the supply air temperature in response to real-time load fluctuations, using machine learning–based predictive models. Two separate power consumption prediction models were developed as the core components of the optimization algorithm. A deep neural network (DNN) model, achieving R² 0.87 and CV(RMSE) 16.3%, was used to predict the outdoor unit (ODU) power consumption, which exhibits complex and nonlinear behavior. In contrast, a polynomial regression model, with R² 0.83 and CV(RMSE) 18.7%, was designed to estimate the fan power consumption based on load ratio and supply air temperature. This differentiation in modeling approach allowed for a balance between computational efficiency and predictive accuracy, facilitating real-time application of the algorithm. A co-simulation framework was constructed by integrating EnergyPlus with Python to emulate realistic operational conditions. The optimization routine was executed every 20 minutes using data from the preceding 20-minute interval. The algorithm selected the supply air temperature setting that minimized the sum of the predicted ODU and fan power consumption. Simulation results demonstrated that, compared to the baseline case using a constant 12°C supply air temperature, the proposed algorithm effectively reduced total energy consumption. Notably, energy savings of 19.3% on the lowest floor, 23.2% on the middle floor, and 13.0% on the top floor were achieved. Additionally, indoor air temperature and relative humidity were maintained within ±1°C and ±5%, respectively, indicating that occupant comfort was not compromised by the control strategy. These results confirm that predictive, load-responsive control strategies can significantly improve the operational efficiency of HVAC systems. The proposed algorithm not only provides a practical method for energy cost reduction but also offers a scalable framework for deployment in real-world commercial building applications. Future work will focus on expanding the control parameters, incorporating internal load prediction, and conducting field validation to assess real-world performance and robustness.