Abstract:

The supercritical carbon dioxide (SCO₂) Brayton cycle system is a promising power generation energy conversion system in the future. The CO₂ physical property characterization model has a significant impact on the prediction accuracy of the power equipment shaft seal and bearing performance in the Brayton cycle system. On the basis of summarizing the experimental data of CO₂ physical properties under different temperatures and pressures in authoritative literature, the prediction accuracy of CO₂ density, viscosity and thermal conductivity prediction model in the classical physical property query software REFPROP software is compared and analyzed, and the physical prediction model with the highest prediction accuracy is obtained. The artificial neural network algorithm is used to obtain the prediction model of CO₂ physical property with higher precision in near critical region. The results show that the FEK model, VS1 model and TC1 model in REFPROP software have the highest prediction accuracy for CO₂ density, viscosity and thermal conductivity, respectively, but the maximum and average error predictions in the near critical region are still more than 40% and 8%, the CO₂ physical property prediction model obtained by the neural network algorithm can reduce the maximum and average error prediction of the near critical point area to 30% and 4%, respectively.

Key words: carbon dioxide, REFPROP software, physical property characterization model, neural network, near critical region

摘要：

超临界二氧化碳（SCO₂）布雷顿循环系统是未来极具潜力的发电能量转换系统，CO₂物性表征模型对布雷顿循环系统中动力设备转轴密封和轴承性能的预测精度影响显著。在总结权威文献中不同温度和压力下CO₂物性实验测试数据的基础上，对比分析了经典物性查询软件REFPROP软件中CO₂密度、黏度和热导率预测模型的预测精度，获得了预测精度最高的物性预测模型及对应临界点附近误差较大的区域，采用人工神经网络算法获得了近临界区预测精度更高的CO₂物性预测模型。结果表明：REFPROP软件中的FEK模型、VS1模型和TC1模型分别对CO₂的密度、黏度和热导率具有最高的预测精度，不过其在近临界区的物性预测最大和平均误差仍分别达到40%和8%以上，利用神经网络算法所获得的CO₂物性预测模型可使近临界点区的物性预测最大和平均误差分别降至30%和4%以下。

关键词: 二氧化碳, REFPROP软件, 物性表征模型, 神经网络, 近临界区