Hybrid steady-state modeling of a residential air-conditioner system using neural network component models

This paper presents a new hybrid steady-state modeling approach for air-conditioning systems. At the system level, it follows the first principles to keep the conservation of mass, energy, and momentum, respectively. At the component level, the physics-based component models are entirely or partially replaced with neural networks. Component neural networks can be trained using the data from laboratory or from the well-validated physics-based component models. Numerical comparisons between the system model consisting of component neural networks and that consisting of physics-based component models indicate that the two system modeling approaches give very close predictions in a wide range of operating conditions. With the proposed hybrid modeling approach, robustness and speed of system modeling can be significantly improved.

► We model a residential air-conditioner system using a hybrid modeling approach.
► First principles are applied to the system level modeling and neural networks are used for component modeling.
► The component neural networks provide accuracy, speed, and robustness to the system modeling.
► The hybrid approach is recommended for large-scale HVAC and refrigeration system modeling.