基于机器学习的煤油液滴蒸发模型探索  

Investigation of kerosene droplet evaporation model based on machine learning

作　　者：王方[1,2,3] 韩琪炜 蔡江涛[1] 李典望 甘甜 金捷 WANG Fang;HAN Qiwei;CAI Jiangtao;LI Dianwang;GAN Tian;JIN Jie(Aeroengine Numerical Simulation Research Center,School of Energy and Power Engineering,Beihang University,Beijing 100191,China;Jiangxi Research Institute,Beihang University,Nanchang 330096,China;Chengdu Innovation Research Institute on Aircraft Power,Beihang University,Pengzhou Sichuan 611930,China)

机构地区：[1]北京航空航天大学能源与动力工程学院航空发动机数值仿真研究中心,北京100191 [2]北京航空航天大学江西研究院,南昌330096 [3]北京航空航天大学成都航空动力创新研究院,四川彭州611930

出　　处：《航空动力学报》2023年第8期1956-1964,共9页Journal of Aerospace Power

基　　金：国家自然科学基金(91741125,12172345,92041001)。

摘　　要：基于厚交换层液滴蒸发理论以及煤油的实验数据,通过机器学习理论中的线性回归、随机森林、支持向量机、极致梯度提升回归和全连接神经网络方法构建煤油液滴的蒸发模型,检验新构建模型的适用范围和精度。对比实验数据和传统模型、机器学习蒸发模型的预测结果,发现随机森林方法和极致梯度提升回归方法生成的蒸发模型不能合理外推,支持向量机方法的外推效果欠佳。线性回归的厚交换层模型和全连接神经网络模型的整体效果更好,与训练数据的均方误差分别为2.71×10^(−2)和1.81×10^(−3)。基于深度学习模型良好的预测效果,可以构建基于实验数据的、可以合理外推的“数字蒸发模型”,可能有更好的现实适应能力。机器学习液滴蒸发模型丰富了现有液滴蒸发模型,为机器学习液滴蒸发模型研究打基础。Based on the droplet evaporation theory of Thick Exchange layer and experimental data of kerosene,a kerosene droplet evaporation model was constructed by using linear regression,random-forest,support vector machine,extreme gradient boosting,and fully-connected neural network methods in machine learning theory to test the applicability and accuracy of the newly constructed model.Comparing the experimental data with the prediction results of traditional models and machine learning evaporation models,it was found that the evaporation models generated by the random-forest method and extreme gradient boosting method cannot be reasonably extrapolated.The extrapolation effect of support vector machine method was not good.The overall effect of the linear regression thick exchange layer model and the fully-connected neural network model was superior to others,with a mean square error of 2.71×10−2 and 1.81×10−3 compared with the training data,respectively.Based on the prediction consequences of the deep learning model,it is feasible to construct a“digital evaporation model”stemmed from experimental data that can be reasonably extrapolated,and may have better realistic adaptability.Machine learning droplet evaporation model enriched extant droplet evaporation models,laying the foundation for machine learning droplet evaporation model research.

关 键 词：液滴蒸发模型 煤油液滴 机器学习 神经网络 拟合能力 外推能力 

分 类 号：V231.2[航空宇航科学与技术—航空宇航推进理论与工程]