文献类型：期刊文献

英文题名：Guided Explicit Mechanism Property Generation Using WGAN and Integrated Regression Model for Insufficient Gasoline NIR Data Augmentation

作者：Luan, Jingran[1];He, Kaixun[2];Zhong, Weimin[3,4];Peng, Xin[1];Lu, Jingyi[1];Wang, Qiang[5]

机构：[1]East China Univ Sci & Technol, Minist Educ, Key Lab Smart Mfg Energy Chem Proc, Shanghai 200237, Peoples R China;[2]Sch Shandong Univ Sci & Technol, Coll Elect Engn & Automat, Qingdao 266590, Peoples R China;[3]East China Univ Sci & Technol, State Key Lab Chem Engn, Shanghai 200237, Peoples R China;[4]East China Univ Sci & Technol, Sch Informat Sci & Engn, Shanghai 200237, Peoples R China;[5]Naval Aviat Univ, Sch Qingdao Branch, Qingdao 266041, Peoples R China

年份：2025

卷号：21

期号：12

起止页码：9240

外文期刊名：IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS

收录：;EI(收录号：20253619114242);WOS:【SCI-EXPANDED(收录号:WOS:001566828300001)】；

基金：This work was supported in part by the National Key Research and Development Program of China under Grant 2022YFB3304701, in part by the National Natural Science Foundation of China under Grant 62273214, and Grant 62173145, and in part by the Open Research Project of the State Key Laboratory of Industrial Control Technology of China under Grant ICT2024B61 and Grant ICT2024A23.

语种：英文

外文关键词：Data augmentation; sample diversity enhancement; spectral properties generation; Wasserstein generative adversarial network (WGAN); Data augmentation; sample diversity enhancement; spectral properties generation; Wasserstein generative adversarial network (WGAN)

摘要：Near-infrared analysis has demonstrated superiority in constructing calibration models for evaluating gasoline properties during blending. However, the intricacies of sampling and the time-intensive analysis process frequently pose challenges in obtaining sufficient labeled samples, thereby hindering the establishment of calibration model with the satisfactory performance. Generative adversarial networks (GANs) bridge this gap by generating virtual samples. Nevertheless, existing methods neglect the global distribution of samples and lack explicit mechanisms for generating properties. To expand labeled sample sets sufficiently for constructing precise and reliable calibration models, a regression Wasserstein deep convolutional generative adversarial network with divergence (RWDCGAN-div) is proposed. Wasserstein divergence is designed in the RWDCGAN-div framework to broaden the global spatial distribution of labeled samples and consistent convolutional neural networks are utilized to simplify model construction. Regression models are integrated to facilitate the discriminator learn the relationship between spectra and properties, thereby guiding the generator in producing spectra and corresponding properties with explicit mechanisms. The effectiveness of the proposed method is verified through prediction experiments based on gasoline blending process.