Volume 62

Data-Physics Hybrid-Driven Plate Heat Exchanger Design Methodology Rui Zhang , Ce Zhang , Minxia Li

Abstract

With increasing demands for energy efficiency, plate heat exchangers have gained widespread application in
heating and cooling pump systems due to their highly efficient and compact heat transfer performance. Addressing the limitations of traditional selection methods in terms of accuracy and adaptability, this paper proposes a hybrid prediction model integrating data-driven approaches with physical constraints. By incorporating the narrow point theory to define the physical boundaries of the heat transfer area and leveraging the non-linear fitting capabilities of neural networks, the method achieves high-precision prediction
of the plate heat exchanger’s heat transfer area under various operating conditions. Further global optimisation of neural network weights and biases through genetic algorithms enhances the model’s generalisation capability and predictive performance. Results demonstrate that the maximum prediction error of the hybrid model, prior to optimisation, did not exceed 10.36%, reducing to below 7.97% post-optimisation.
Overall accuracy improved by 11.9% compared to conventional methods. This approach provides effective
support for the precise selection of plate heat exchangers and the optimised design of heating and cooling pump systems.

Keywords Heat pump, Refrigeration system, Plate heat exchanger, Heat transfer area, Hybrid model