Accurate prediction models for the viscosity and thermal conductivity of refrigerants are of great importance and have drawn wide attention from scholars. Considering the great advantage of artificial neural network (ANN) models in solving non-linear problems, two fully connected feed-forward ANN models were proposed to predict the viscosity and thermal conductivity of the HFC/HFO refrigerants in this paper. The reduced pressure (p_r), reduced temperature (T_r), mole mass (M) and acentric factor (ω) of the refrigerants were selected as the input variables for both ANN models. Regarding the ANN model for viscosity, the neural number of the hidden layer was optimized to be 9 by trial-and-error method. The prediction results coincided with the experimental data very well. The correlation coefficient and the average absolute deviation (AAD) of regression were 0.9998 and 1.21%, respectively. The prediction of thermal conductivity also showed a good agreement with the experimental data, and the AAD of the model was 1.00%. The paper is expected to provide valuable methods to predict the viscosity and thermal conductivity of HFC/HFO refrigerants.

制冷剂粘度和导热系数的准确预测模型非常重要，已引起学者的广泛关注。考虑到人工神经网络（ANN）模型在解决非线性问题方面的巨大优势，本文提出了两种完全连接的前馈神经网络模型来预测HFC / HFO制冷剂的粘度和导热系数。减压（p _r），减压（T _r），摩尔质量（M）和偏心因数（ω）两种制冷剂均被选作两个ANN模型的输入变量。对于ANN模型的黏度，通过试错法将隐层的神经数优化为9。预测结果与实验数据非常吻合。回归的相关系数和平均绝对偏差（AAD）分别为0.9998和1.21％。导热系数的预测也与实验数据吻合良好，模型的AAD为1.00％。预期该论文将提供有价值的方法来预测HFC / HFO制冷剂的粘度和热导率。