局部广义矩阵学习向量量化在多联机系统阀类故障诊断中的应用  

Application of Localized Generalized Matrix Learning Vector Quantization in Fault Diagnosis of Valves in Variable Refrigerant Flow System

作　　者：张鉴心 李正飞 陈焕新[1] 刘倩 王誉舟 ZHANG Jianxin;LI Zhengfei;CHEN Huanxin;LIU Qian;WANG Yuzhou(School of Energy and Power Engineering,Huazhong University of Science and Technology,Wuhan 430074,Hubei,China)

机构地区：[1]华中科技大学能源与动力工程学院,湖北武汉430074

出　　处：《制冷技术》2022年第1期8-16,共9页Chinese Journal of Refrigeration Technology

基　　金：国家自然科学基金(No.51876070,No.51576074)。

摘　　要：多联机(VRF)系统结构复杂,采用传统方法难以识别设备故障,本文首次将特征选取和多目标进化算法优化框架集成于局部广义矩阵学习向量量化(LGMLVQ)故障诊断模型中,并应用于多联机系统的电子膨胀阀和四通阀故障诊断研究。运行数据来自多联机系统实验平台。实验结果表明,特征选择后的模型对多联机阀类故障的诊断准确率从91.40%提升到95.73%,而进化算法进一步将准确率提升至99.50%,耗时从14.109 s缩短为11.724 s。分析最优模型的正常数据自适应相关性矩阵可以对故障发生部位进行模糊定位,可以对实际故障维修提供一定的帮助。Due to the complicated internal construction of variable refrigerant flow(VRF)system,it is difficult to identify equipment faults by traditional method.The local generalized matrix learning vector quantization(LGMLVQ)fault diagnosis model with feature selection and multi-objective evolutionary algorithm optimization framework are integrated for the first time,which is applied to faults diagnosis for electronic expansion valve and four-way valve.The operation data are extracted from the VRF system experiment platform.The results show that the diagnosis accuracy of the model after feature selection is risen from 91.40%to 95.73%,and up to 99.50%with evolutionary algorithm optimization.Meanwhile,the operation time consumption is reduced from 14.109 s to 11.724 s.The analysis of features adaptive correlation matrix of the optimal model can locate fault roughly,which can provide some help for the equipment repair.

关 键 词：多联机 故障诊断 随机森林 多目标进化算法 局部广义矩阵学习向量量化 

分 类 号：TU834[建筑科学—供热、供燃气、通风及空调工程] TP306.3[自动化与计算机技术—计算机系统结构]