基于物理混合神经网络的涡流管性能研究  

Research on the Performance of Vortex Tube Based on Physical Hybrid Neural Network

作　　者：李申申 韩志宏 刘蜀阳 黄志远 甘德俊 LI Shenshen;HAN Zhihong;LIU Shuyang;HUANG Zhiyuan;GAN Dejun(School of Mechanical and Electronic Engineering,Jingdezhen Ceramic University,Jingdezhen 333403,China;New Engineering Industry College,Putian University,Putian 351100,China)

机构地区：[1]景德镇陶瓷大学机械电子工程学院,江西景德镇333403 [2]莆田学院新工科产业学院,福建莆田351100

出　　处：《现代信息科技》2025年第8期194-198,共5页Modern Information Technology

基　　金：国家自然科学基金(52066006);景德镇市科技局项目(2019GYZD008-13)。

摘　　要：文章通过添加伯努利方程和尼古拉兹公式物理约束条件构建了混合神经网络模型,探索了涡流管冷端温度变化规律并进行了相应预测。网络采用多层前馈模型和Levenberg-Marquardt学习算法,选择双曲正切函数作为传递函数。此外,利用决定系数(R~2)、均方根误差(RMSE)来确定所开发模型的统计有效性,并分析了模型不确定性和鲁棒性。混合模型的指标R~2为0.993 6,RMSE为0.339 2,在不确定性和鲁棒性方面也具有良好的表现。这些数据表明文章构建的模型成功地预测了涡流管冷端温度的变化,并且具有良好的精度。In this paper,a hybrid neural network model is constructed by adding the physical constraint conditions of the Bernoulli equation and the Nicolas formula,exploring the temperature change law of the cold end of the vortex tube and making corresponding predictions.The network adopts a multi-layer feedforward model and the Levenberg-Marquardt learning algorithm,and the hyperbolic tangent function is selected as the transfer function.In addition,the coefficient of determination(R2)and the Root Mean Square Error(RMSE)are used to determine the statistical validity of the developed model,and the model's uncertainty and robustness are analyzed.The hybrid model has an index R2 of 0.9936 and an RMSE of 0.3392,and also has a good performance in terms of uncertainty and robustness.These data indicate that the model constructed in this paper successfully predicts the changes in the temperature of the cold end of the vortex tube and has good accuracy.

关 键 词：涡流管 预测模型 混合神经网络 温度性能 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP399[自动化与计算机技术—控制科学与工程]