摘要
针对获取化合物临界温度(T_(c))的传统实验和计算方法成本高的问题,将图对比学习(GCL)算法应用于原油组分T_(c)的预测中,结合现有的Tc数据集与补充的原油组分相关数据比较了GCL算法和传统计算模型区别。计算结果表明,GCL算法可捕捉图结构中的节点和边特征,同时对训练数据量要求较小,适用于分子性质预测;GCL算法具有更高的预测准确度,同时调整分子二维和三维结构编码可对GCL的预测性能起到提升的效果。
In response to the high cost of traditional experimental and computational methods for obtaining the critical temperature(T_(c))of compounds,the graph comparative learning(GCL)algorithm was applied to predict the T_(c) of crude oil components.The differences between the GCL algorithm and traditional computational models were compared by combining the existing Tc dataset with supplementary related data of crude oil components.The calculation results indicate that the GCL method can capture the characteristics of nodes and edges in the graph,while requiring less training data,making it suitable for predicting the properties of molecules.The GCL method shows higher prediction accuracy,and the encoding method of adjusting the 2D and 3D of molecular can improve the predictive performance of GCL.
作者
蔡一涵
崔乐雨
李欣
苏智青
何秀娟
李应成
CAI Yihan;CUI Leyu;LI Xin;SU Zhiqing;HE Xiujuan;LI Yingcheng(Sinopec Shanghai Research Institute of Petrochemical Technology Co.,Ltd.,Shanghai 201208,China;State Key Laboratory of Green Chemical Engineering and Industrial Catalysis,Shanghai 201208,China)
出处
《石油化工》
CAS
CSCD
北大核心
2024年第4期518-524,共7页
Petrochemical Technology
基金
中国石化集团公司重点研发项目(KL22055)。
关键词
分子性质预测
图对比学习
碳捕集利用与封存
CO_(2)驱油
molecular property prediction
graph contrastive learning
carbon capture,utilization and storage
carbon dioxide flooding