Molecular Reconstruction of Naphtha based on Physical Information Neural Network. Issue 7 (2022)

• Record Type: Journal Article
• Authors: Ma, Fangyuan; Zheng, Xin; Han, Chengyu; Wang, Jingde; Sun, Wei
• Is Part Of: IFAC-PapersOnLine. Volume 55:Issue 7(2022)
• Page Start: 186
• Page End: 191
• Publication Date: 2022
• Subjects: Artificial Neural Network -- Mixing Rules -- Generalization Ability; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2022.07.442
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: A molecular reconstruction method based on physical information neural network is proposed for predicting the molecular composition of naphtha. By embedding physical information utilized in typical molecular reconstruction methods, such as mixing rules, into the loss function of the neural network, the model tends to converge to the state conforming to physical rules in training stage. The neural network model obtained by the method contains certain physical information, which can improve the generalization ability of the model. The results show that the prediction performance and application range of the proposed method are better than those of the typical ANN-based molecular reconstruction method.