Neural-Accelerated Dynamic modeling of heat pumps

Abstract

In response to escalating climate challenges and the transition to renewable energy, integrating heat pumps with other energy systems has gained significant attention. Fast and accurate dynamic models for heat pump are essential in predicting its transient thermal behavior and simulating its interactions with other energy systems, and effectively controlling and optimizing its performance in real time. Among the heat pump’s components, heat exchangers involve complex heat transfer phenomenon and contributes to majority of the dynamics of the heat pump. Among the common numerical approaches used in the dynamic modeling of heat exchangers, the finite-volume method is well-recognized for its robustness and high accuracy. However, it is often outpaced in terms of computational efficiency compared to other methods. This paper introduces novel Neural-Accelerated Dynamic (NAD) models for heat pump’s heat exchangers, integrating the strengths of numerical modeling and machine learning. The NAD model’s architecture comprises of simplified heat exchanger model and a deep neural network, which are dynamically strongly coupled, achieving a computationally efficient model while maintaining high precision. Furthermore, in validation, the NAD models demonstrated excellent computational performance, achieving an average error below 0.4% and operating on average 271 times faster than the state-of-the-art finite-volume model. This highlights the capability of the NAD model to rapidly and precisely approximate complex heat transfer interactions in heat exchangers of the heat pump, making it particularly suitable for model control applications, optimization tasks, and long-duration simulations.