Analysis of the impact of supply air temperature on outdoor unit power consumption in EHP-AHU systems and development of a DNN-Based predictive model

Recent studies on optimizing heating, ventilation, and air conditioning (HVAC) system control using data-driven predictive models are making significant strides toward achieving the decarbonization goals in the building sector. However, previous research often faced challenges in acquiring measured data that reflected control setpoint variations, leading to the use of simulation-derived data as training data for predictive models. Additionally, performance data provided by manufacturers for the simulation modeling of direct expansion air handling unit (AHU) systems like electric heat pumps (EHP)-AHU was based on experimental data using variable refrigerant flow systems connected to cassette-type and duct-type indoor units. This approach inadequately captured the heating and cooling performance and energy consumption characteristics of EHP-AHU systems. This study was conducted as a foundational investigation aimed at addressing the limitations of previous research and supporting the development of an optimal control algorithm for the EHP-AHU system. To this end, a multi-calorimeter experiment was carried out to quantitatively evaluate the cooling performance under varying supply air temperature conditions and to assess the feasibility of algorithm development through regression analysis. In addition, to develop a prediction-based optimal control algorithm, this study aimed to construct a deep neural network (DNN)-based prediction model capable of accurately capturing the dynamic variations in operating conditions, including outdoor air temperature, thermal load, and humidity, under real-world environments. Experimental results demonstrate that as the supply air temperature increases, the low pressure of the EHP-AHU system rises, leading to a reduction in the power consumption of the outdoor unit. Depending on outdoor conditions, power consumption reduction rates vary. For example, changing the supply air temperature from 12 °C to 14 °C and 16 °C resulted in maximum reduction rates of 20.7 % and 32.3 %, respectively. Thus, supply air temperature setpoints were identified as critical control variables for effectively reducing the power consumption of outdoor units. Using multi-calorimeter experimental data reflecting EHP-AHU system characteristics, a DNN-based power consumption prediction model was developed. Additionally, hyperparameter optimization achieved a highly accurate predictive model with an R² score of 0.87 and a coefficient of variation of the root mean square error of 16.3 %.