基于机器学习模型的制冷剂燃爆特性预测  

Prediction of Refrigerant Ignition and Detonation Characteristics Based on Machine Learning Model

作　　者：费腾 杨昭[1] 陈裕博 张勇[1] 李杰[1] FEI Teng;YANG Zhao;CHEN Yubo;ZHANG Yong;LI Jie(Key Laboratory of High Efficiency Utilization of Low and Medium Temperature Heat Energy,Ministry of Education,Tianjin University,Tianjin 300072,China)

机构地区：[1]天津大学,中低温热能高效利用教育部重点实验室,天津300072

出　　处：《工程热物理学报》2024年第3期667-673,共7页Journal of Engineering Thermophysics

基　　金：国家自然科学基金资助项目(No.51936007)。

摘　　要：本文以常见可燃工质为对象,基于Gaussian 16W的M06-2X/6–311+G(d, p)优化计算,得出工质分子结构在微观模型下的分子描述符数据。同时采用多元线性回归(MLR)、随机森林(RF)、人工神经网络(ANN)三种不同的机器学习方法,将微观数据与宏观实验数据关联起来,从而预测这些工质的最小点火能,总体R2分别达到了0.853、0.782和0.906,表明预测有着良好的准确度和鲁棒性,该预测结果可以为新工质的实用性和安全性提供理论依据。In this paper,common flammable working medium is taken as the object.Based on M06-2X/6-311+G(d,p)optimization calculation of Gaussian 16 W,molecular descriptor data of working medium molecular structure under microscopic model are obtained.At the same time,three different machine learning methods,namely Multiple Linear Regression(MLR),Random Forest(RF)and Artificial Neural Network(ANN),were used to correlate the micro data with the macro experimental data,so as to predict the minimum ignition energy of these working media.The overall R^(2) reached 0.853,0.782 and 0.906,respectively.The results show that the prediction has good accuracy and robustness.The prediction results can provide theoretical basis for the practicability and safety of the new working medium.

关 键 词：燃爆特性 机器学习 最小点火能 可燃工质 

分 类 号：TK16[动力工程及工程热物理—热能工程]