化工学报 ›› 2016, Vol. 67 ›› Issue (11): 4689-4695.DOI: 10.11949/j.issn.0438-1157.20160289

• 过程系统工程 • 上一篇    下一篇

基于基团拓扑的遗传神经网络工质临界温度预测

苏文, 赵力, 邓帅   

  1. 天津大学中低温热能高效利用教育部重点实验室, 天津 300072
  • 收稿日期:2016-03-11 修回日期:2016-08-25 出版日期:2016-11-05 发布日期:2016-11-05
  • 通讯作者: 赵力,jons@tju.edu.cn
  • 基金资助:

    国家自然科学基金项目(51276123, 51476110)。

Prediction of refrigerant critical temperature with genetic neural network based on group topology

SU Wen, ZHAO Li, DENG Shuai   

  1. Key Laboratory of Efficient Utilization of Low and Medium Grade Energy, Ministry of Education, School of Mechanical Engineering, Tianjin University, Tianjin 300072, China
  • Received:2016-03-11 Revised:2016-08-25 Online:2016-11-05 Published:2016-11-05
  • Supported by:

    supported by the National Natural Science Foundation of China (51276123, 51476110).

摘要:

用遗传神经网络预测工质的临界温度,网络的输入参数为分子基团和拓扑指数,输出参数为临界温度。所划分的16个分子基团涵盖了制冷、热泵及有机朗肯循环系统中的大部分工质,所选拓扑指数能够分辨工质中所有的同分异构体。通过遗传算法优化得到网络结构及初始参数后,由神经网络对工质临界温度进行预测,同时为了提高网络对临界温度预测的泛化能力,将200种工质划分成训练集、验证集及测试集。所得网络能够区分所有的同分异构体,且与实验值相比,各数据集临界温度的平均相对误差分别为1.18%、1.69%、1.28%,表明该网络对工质临界温度具有很好的预测能力。

关键词: 热力学性质, 临界温度, 工质, 分子基团, 拓扑指数, 遗传算法, 神经网络

Abstract:

A genetic neural network was presented to predict the critical temperature of refrigerants. The inputs of the network included molecular groups and a topological index, and the output was the critical temperature. 16 molecular groups divided can cover most of the refrigerants or working fluids in refrigeration, heat pump and Organic Rankine Cycle research. The chosen topological index was able to distinguish all refrigerant isomers. The critical temperatures of refrigerants were estimated by the neural network after obtaining the optimized network structure and initial parameters by genetic algorithm. At the same time, in order to improve network generalization ability of prediction, 200 data points were divided into three data sets including the training, validation, and test sets. The calculated results based on the developed network showed a good agreement with experimental data. The network can distinguish all refrigerant isomers and compared with the experimental data. The average absolute relative deviations for training, validation and test sets were 1.18%, 1.69% and 1.28%, respectively.

Key words: thermodynamics property, critical temperature, refrigerant, molecular groups, topological index, genetic algorithm, neural network

中图分类号: