Fuel type recognition of ester isomer additives in flames by optical diagnostics coupled with machine learning method  

作　　者：HE JiaYing CHEN MengFei WU BingKun YING YaoYao YAN WeiJie LI TianJiao LIU Dong 

机构地区：[1]MIIT Key Laboratory of Thermal Control of Electronic Equipment,School of Energy and Power Engineering,Nanjing University of Science and Technology,Nanjing 210094,China [2]Advanced Combustion Laboratory,School of Energy and Power Engineering,Nanjing University of Science and Technology,Nanjing 210094,China

出　　处：《Science China(Technological Sciences)》2024年第11期3431-3442,共12页中国科学（技术科学英文版）

基　　金：supported by the National Natural Science Foundation of China(Grant Nos.52076110,52376115,and 52106160)。

摘　　要：Based on machine learning models,an approach for the type recognition of oxygenated additives(ester isomers,i.e.,methyl butyrate,methyl crotonate,ethyl acrylate,and ethyl acrylate)via optical diagnostics was proposed.By utilizing optical diagnostic methods flame features were extracted,and three models including random forest(RF),artificial neural network(ANN),and support vector machine(SVM),were employed to establish the relationship between flame images and oxygenated additives.Moreover,the impact of multiple factors on model performance,including image compression,dataset size,and feature number was also investigated.The images of flame obtained from inverse diffusion flame under four different oxygenated additives and various combustion conditions were used as examples to examine the effectiveness of the proposed approach.Results indicated that the accuracy of the recognition of ester isomers by the proposed approach exceeded 90%.Furthermore,it is observed that image compression had minimal impact on prediction accuracy but significantly reduced processing time.Different types of features contributed to predicting the type of ester isomers variously,and all models exhibited improved accuracy with an increased number of features.The number of samples significantly affected model accuracy.The investigation of feature missing and insufficient training samples suggested that ANN and RF models were more suitable for cases with many missing features,while SVM was more suitable for dealing with small samples.

关 键 词：fuel recognition machine learning feature extraction optical diagnostics ester isomers 

分 类 号：TQ517[化学工程] O657.3[理学—分析化学] TP181[理学—化学]