超临界RP-3航空煤油热物性替代模型研究  

Study on Surrogate Models for Thermophysical Properties of Supercritical Aviation Kerosene RP-3

作　　者：沈扬 刘源斌 曹炳阳[1] SHEN Yang;LIU Yuan-Bin;CAO Bing-Yang(Key Laboratory for Thermal Science and Power Engineering of Ministry of Education,School of Aerospace Engineering,Tsinghua University,Beijing 100084,China)

机构地区：[1]清华大学航天航空学院,热科学与动力工程教育部重点实验室,北京100084

出　　处：《工程热物理学报》2022年第4期1048-1054,共7页Journal of Engineering Thermophysics

基　　金：国家科技重大专项(No.2017-III-0005-0030)。

摘　　要：为准确预测RP-3航空煤油在超临界压力下的热物性,本文在四组分模型的基础上,分别基于遗传算法(GA)和人工神经网络(ANN),提出了两种构建航空煤油替代模型的方法,并对比了两种方法在预测不同热物性上的性能。关注的热物性包括密度、黏度、比定压热容以及热导率。结果表明对于密度和黏度,两种方法均能得到高精度的替代模型;对于比定压热容,GA构建的模型精度更高,而ANN构建的模型表现反而变差;对于热导率,但由于缺乏跨临界区的实验数据,GA可以通过约束适应度函数使模型仍能准确预测伪临界温度,而ANN的灵活性则较差。To accurately predict thermophysical properties of aviation kerosene RP-3 at supercritical pressures, we propose two surrogate formulation methodologies for a unitary property based on genetic algorithm(GA) and artificial neural network(ANN), respectively. We compare the performance of the two approaches on different thermophysical properties including density, viscosity,constant-pressure heat capacity, and thermal conductivity. The results indicate that for density and viscosity, both the two methods can formulate accurate surrogate models;For constant-pressure heat capacity, GA can generate better surrogate models whereas the performance of ANN is poor;For the thermal conductivity, due to the lack of experimental data across the trans-critical region, GA can still generate the model which can accurately predict the pseudo-critical temperature by constraining the fitness function, while ANN is less flexible.

关 键 词：热物性 航空煤油 超临界状态 遗传算法 神经网络 

分 类 号：V312[航空宇航科学与技术—航空宇航推进理论与工程]