Air conditioning facilities tend to be designed excessively because of the difficulties in estimating peak heat loads. Therefore, their energy efficiencies decrease owing to an increase in low load operations. The demand for using accurate heat load predictions for optimal designs is increasing. In this study, a hybrid annual heat load prediction model combining dynamic heat load calculation and machine learning was developed. The model was then evaluated considering 16 floors in 5 office buildings in which multi-split air-conditioning systems were installed. The hybrid model learns the relations between explanatory variables and errors of dynamic heat load calculation via machine learning during learning periods. Then, future heat loads are estimated by correcting the calculated values of dynamic heat load using errors estimated via machine learning. We confirmed that the hybrid model was more accurate than dynamic heat load calculation and machine learning in both cases of cooling and heating load prediction. Additionally, for simplifying building models and diminishing learning periods, the hybrid model was confirmed to be advantageous. Heat loads consist of robust and sensitive components. The robust components are decided based on whether changes and heat transmissions in buildings. They are easy to explain by physical laws. On the contrary, the sensitive components are decided based on human behavior, for example, internal heating and operation of air conditioning facilities. Those components should be considered as probabilistic events. Dynamic heat load calculation is superior in estimating the robust components. Machine learning is superior in estimating the sensitive components. Thus, the hybrid model is naturally more accurate than dynamic heat load calculation and machine learning.

To apply one-dimensional simulation for the airflow and heat transfer analysis of underground stations, it is necessary to preliminarily estimate the loss factors and heat transfer coefficients as the simulation inputs. However, it is difficult to obtain them because the spatial configurations show significant three-dimensionality, and the flow almost becomes turbulent. This study aimed to obtain adequate loss factor and heat transfer coefficients for one-dimensional simulation by using a high-accuracy numerical simulation in an H-shaped channel with counterflows and heat transfer, which is a simplified two-dimensional configuration model of underground stations. First, a numerical simulation of the H-shaped channel flow is carried out to obtain the loss factor in the flow. For the proper calculation of the turbulent flow, three low-Reynolds-number k-epsilon models and LES are adopted. As a result, the drastic change in the flow pattern as a function of the Reynolds number and/or the length of mixing region of an H-shaped channel are clarified. Further, the characteristic behaviors of the loss coefficients of the counterflows in various H-shaped channels are revealed.

A heating ventilating air conditioning (HVAC) system comprising a heat source unit, an air conditioner, and a water / air transport system accounts for approximately 40% of the energy consumption of a building, so highly efficient design and operation is important. There are many energy-saving methods for air-conditioning systems related to automatic control, but few studies have quantitatively evaluated their energy-saving effects. At the design stage of HVAC system, the equipment is selected according to the load, but detailed automatic control logic and control parameters are rarely examined. Therefore, the control state deteriorates during operation, and the expected energy-saving effect cannot be obtained. To properly evaluate the energy-saving effect at the design stage, programs that can handle various controls are required. the major programs in Japan and overseas are built for the purpose of calculating annual energy consumption. Thus, the actual control was not sufficiently reflected. Therefore, in this study, we constructed a unique program that can quantitatively calculate the energy-saving effect of various controls by reproducing the behavior of the device according to the control logic and control parameters in detail. Further, we examined the energy-saving effect of variable air volume (VAV)・variable water volume (VWV) and CO₂ concentration control. In this study, the VAV / VWV control logic introduced in the experimental building was modeled as faithfully as possible for the existing experimental building. In addition, to quantify the influence of the control state of the equipment, the simulation was constructed by incorporating the physical model and the equipment model in as much detail as possible. Then, to validate the constructed simulation, we confirmed the room temperature calculation model and compared it with the measured values regarding the behavior of the room temperature during the operation of the HVAC system. Through the constructed simulation, VAV/VWV control, constant air volume (CAV)/VWV control, and VAV/ constant water volume (CWV) control were defined as comparison targets, and the annual power consumption was calculated. By comparing the results, the energy-saving effect of the VAV / VWV control was quantified. From the analysis of the control state, the superiority of the VAV / VWV control over CAV or CWV control was confirmed. In addition, the detailed calculation results of each month by the VAV / VWV control showed that the controllability deteriorated at other times when the parameters were adjusted according to the time when the load was heavy. Thus, readjusting the parameters may improve controllability. In the future, by using this simulation, the behavior of each device according to the control can be calculated in detail; therefore, it is expected to be used as an energy-saving design tool at the time of design, parameter adjustment at the operation stage, and defect detection / diagnosis. In the next report, we considered the energy-saving effect and the effect on room temperature controllability when the parameters of the control logic for VAV / VWV / CO₂ concentration control were changed.