Supercritical thermophysical properties prediction of multi-component hydrocarbon fuels based on artificial neural network models

A good understanding of the thermophysical properties of hydrocarbon fuels at supercritical pressure is important to research on experiment and numerical simulation of fuel supercritical spray. Experimental measurements are difficult to conduct directly because of the extremely high pressure and high temperature. In this study, back propagation (BP) neural network, BP optimized by mind evolution algorithm (MEA-BP) and BP neural network optimized by genetic algorithm (GA-BP) are established to determine the nonlinear temperature-dependent thermophysical properties of density, viscosity, and isobaric specific heat (C-p) of hydrocarbon fuels at supercritical pressure Meanwhile, approximate formulas for these properties prediction are primarily proposed using polynomial fitting In this paper, models that can predict three types of physical properties of three kinds of hydrocarbon fuels and their mixtures in a wide temperature range under supercritical pressure are established In the prediction of density and C-p, BP neural network has a good prediction effect. The results show that the MAPE is lower than 2% in the prediction of density and C-p, but the MAPE of viscosity prediction is slightly higher than 5% using BP. Furthermore, MEA and GA are used to optimize the prediction of viscosity. The optimization effect and computation of the MEA is better than that of GA because MEA does not have the local optimization and prematurity problems. The present work offers an efficient tool to predict the thermophysical properties of hydrocarbon fuels over a wide range of temperatures under supercritical pressure which can be easily extended to other fuels of interest. It will be beneficial to the experiment and numerical simulation studies of supercritical sprays.

深入了解碳氢燃料在超临界压力下的热物理性质对于燃料超临界喷雾的实验研究和数值模拟至关重要。由于极高的压力和温度，直接进行实验测量非常困难。在本研究中，建立了反向传播（BP）神经网络、思维进化算法优化的BP（MEA - BP）以及遗传算法优化的BP（GA - BP）神经网络，用于确定碳氢燃料在超临界压力下密度、粘度和等压比热容（\(C_p\)）随温度变化的非线性热物理性质。同时，初步提出了使用多项式拟合来预测这些性质的近似公式。本文建立了能够预测三种碳氢燃料及其混合物在超临界压力下较宽温度范围内的三类物理性质的模型。在密度和\(C_p\)的预测中，BP神经网络具有良好的预测效果。结果表明，在密度和\(C_p\)的预测中，平均绝对百分比误差（MAPE）低于2%，但使用BP预测粘度时，MAPE略高于5%。此外，使用MEA和GA对粘度预测进行优化。由于MEA不存在局部最优和早熟问题，其优化效果和计算性能优于GA。本研究提供了一种有效的工具，用于预测碳氢燃料在超临界压力下较宽温度范围内的热物理性质，该工具可轻松扩展到其他感兴趣的燃料。这将有利于超临界喷雾的实验和数值模拟研究。