基于人工神经网络的工质基础物性预测  

PREDICTION OF PHYSICAL PROPERTIES OF WORKING FLUID BASED ON ARTIFICIAL NEURAL NETWORK

作　　者：林美金 董轩 洪小东 廖祖维[1] 孙婧元[1] 杨遥[1] 王靖岱[1] 阳永荣[1] Lin Meijin;Dong Xuan;Hong Xiaodong;Liao Zuwei;Sun Jingyuan;Yang Yao;Wang Jingdai;Yang Yongrong(College of Chemical Engineering and Bioengineering,Zhejiang University,Hangzhou 310030;ZJU-Hangzhou Global Scientific and Technological Innovation Center)

机构地区：[1]浙江大学化学工程与生物工程学院,杭州310030 [2]浙江大学杭州国际科创中心

出　　处：《石油炼制与化工》2024年第1期180-188,共9页Petroleum Processing and Petrochemicals

基　　金：国家自然科学基金项目(U22A20415);浙江省尖兵领雁计划项目(2022C01SA442617)。

摘　　要：烃类及卤代烃是制冷及余热发电等热力学循环系统潜在的理想工质,但其数量繁多且多数物性参数未知,建立准确的物性预测模型对新型工质的开发至关重要。从多个公开数据库中收集了2500多种烃类及卤代烃分子(含C,H,F,Cl)的基础物性参数,包括正常沸点(T_(b))、临界温度(T_(c))、临界压力(p_(c))、偏心因子(ω),构建了一个工质物性数据库;进一步,通过改进基团贡献-人工神经网络(GC-ANN)的方法,模型的输入参数除基团频率外,还加入相对分子质量、T_(b)、约化维纳指数,建立了预测烃类及卤代烃分子T_(b),T_(c),p_(c),ω的神经网络模型,所开发模型的预测误差小于传统的GC-ANN的误差。Traditional working fluids in thermodynamic cycles,such as refrigeration and waste heat power generation,have been associated with issues such as ozone layer depletion and global warming.The development of efficient and environmentally friendly novel working fluids has become a research focus.Hydrocarbons and halogenated hydrocarbons are ideal candidates,but their large number and many unknown thermophysical properties make it crucial to establish accurate models for predicting these properties in order to screen new working fluids effectively.In this study,the basic thermophysical parameters of more than 2500 hydrocarbons and halogenated hydrocarbons containing C,H,F,and Cl atoms were collected from various public databases,including normal boiling point(T_(b)),critical temperature(T_(c)),critical pressure(p_(c))and acentric factor(ω),and furtherly,by improving the method of group contribution-artificial neural network(GC-ANN),a neural network model for predicting T_(b),T_(c),p_(c)andωof hydrocarbons and halogenated hydrocarbons containing C,H,F,and Cl atoms was established by adding relative molecular mass,T_(b)and approximate wiener index to the input parameters of the model.The prediction errors of the models developed in this study were smaller than those of the traditional GC-ANN.

关 键 词：新型工质 物性预测 基团贡献法 BP神经网络 

分 类 号：TQ201[化学工程—有机化工] TP183[自动化与计算机技术—控制理论与控制工程]