基于人工神经网络模型的传热系数预测模型

R1234ze(E) and R600a are considered as promising environmental-friendly alternative refrigerants and accurate heat transfer data is essential when R1234ze(E) and R600a are applied in engineering practice. In this study, heat transfer coefficient prediction models are developed by using R1234ze(E) and R600a experimental data through a back-propagation (BP) neural network. The prediction result of new models is better than the classical heat transfer model in the literature. To expand the prediction ability of the model, a more general heat transfer prediction model based on R1234ze(E) and R600a is also developed. For R1234ze(E), the average relative deviation (ARD) of prediction results is 4.08%, average absolute relative deviation (AARD) is 8.46%, λ_10% is 70.2%; for R600a, the ARD of prediction results is −3.59%. AARD is 6.98%, λ_10% is 76.4%. The ARD range of the prediction results for six works of literature is −17.9% to 26.8%, AARD is no more than 27.6%, and λ_30% is not less than 60.0%, which shows that the model in this study has certain prediction accuracy and universality.