Abstract: Based on the experimental data of variable structure heat exchangers, the feasibility of neural network model in the performance for the prediction of water-air finned tube heat exchangers is studied. The back propagation (BP) neural network models of 2 rows and 3 rows of finned tubes under refrigeration and heating conditions are established which optimize and determine the optimal network structure under the condition of single hidden layer and double hidden layer. The prediction error of the models is about 1%. The specified structural heat exchanger data is set as a test set, and the performance of the single hidden layer and double hidden layer network model is compared. The research results show that for the refrigeration condition, the double hidden layer model cannot improve the accuracy of the model, but will even reduce the prediction accuracy of some parameters due to over-fitting. For the heating condition, the double hidden layer model has a better accuracy in prediction.

Key words: finned tube heat exchanger, back propagation (BP) neural network, variable structure, network architecture