基于蝙蝠算法优化的BP神经网络估算工质沸点温度

BP neural network based on bat algorithm optimization to estimate the boiling point temperature of working medium

为了能够对计算机辅助分子设计过程中产生的分子实现精确的热物性预测,采用工质的沸点为重要预测参数,将现有的烷烃、烯烃、卤代烃、醇、醚、胺等6类热力循环所使用的工质划分为16类基团,并引入拓扑指数来区分同分异构体。提出了一种基于蝙蝠算法优化过的BP人工神经网络来预测工质的沸点温度。最后,通过仿真实验将优化后得到的BA-ANN与典型的GA-ANN模型进行误差对比,结果表明新构建的BA-ANN模型的各类预测误差均低于GA-ANN模型,其新模型能够显著提高工质沸点温度的预测精度。

In order to achieve accurate prediction of thermal properties of molecules generated in computer aided molecular design process,the boiling point of working fluid was taken as an important prediction parameter,and existing working fluids commonly used in thermal cycle was divided into 16 groups,and the topological index was introduced to distinguish isomers.The BP artificial neural network optimized by bat algorithm was proposed to predict the boiling point temperature of working fluid.Finally,the error comparison between the optimized BA-ANN model and the typical GA-ANN model was carried out through simulation experiments.The results show that all kinds of prediction errors of the developed BA-ANN model are lower than that of GA-ANN model,and the new model can significantly improve the prediction accuracy of boiling point temperature.