多联机系统制热工况下的制冷剂充注量故障诊断模型——基于多层卷积神经网络  

Fault Diagnosis Model for Refrigerant Charge of Varivable Refrigerant Flow System under Heating Conditions Based on Multi-Layer Convolution Neural Network

作　　者：程亨达 陈焕新[1] 李正飞 程向东[2] Cheng Hengda;Chen Huanxin;Li Zhengfei;Cheng Xiangdong(School of Energy and Power Engineering,Huazhong University of Science and Technology,Wuhan,430074,China;Wuhan Textile University,Wuhan,430073,China)

机构地区：[1]华中科技大学能源与动力工程学院,武汉430074 [2]武汉纺织大学,武汉430073

出　　处：《制冷学报》2020年第1期40-47,共8页Journal of Refrigeration

基　　金：国家自然科学基金(51876070,51576074)资助项目~~

摘　　要：本文提出一种基于卷积神经网络的故障诊断模型,并通过正交试验优化了3层网络的卷积核和神经元数目,利用图形化的多联机(VRF)系统制冷剂充注量故障实验数据训练了多层卷积神经网络,评估了本模型的故障诊断性能。结果表明:该"数据图形化-多层卷积神经网络"方法建立的模型能够有效进行多联机制冷剂充注量故障诊断,20个输入特征时,对9类故障诊断总正确率最大为91%,比传统BP神经网络达到更高的诊断精度。该方法首次利用卷积神经网络完成了VRF制冷剂充注量故障诊断,为相关研究的拓展奠定了基础。This paper presents a fault diagnosis model based on a convolution neural network. The kernel size and number of neurons of a3-layer convolution network were optimized by an orthogonal experiment method. The performance of the refrigerant charge fault diagnosis model of variable refrigerant flow(VRF) system was evaluated with graphed experimental data. The results show that the model established by the " data graphing & multi-layer convolutional network" method can be effectively used for the refrigerant charge fault diagnosis of the VRF system. With 20 chosen input features,the accuracy of the 9 level refrigerant charge fault diagnosis reached 91%,surpassing the performance of traditional back propagation neural networks(BPNN). This is the first time to achieve VRF system refrigerant charge fault diagnosis by using a convolutional network,laying a foundation for the expansion of related research.

关 键 词：多联机系统 故障诊断 卷积神经网络 制冷剂充注量故障 正交试验 

分 类 号：TB657.2[一般工业技术—制冷工程] TU831.3[建筑科学—供热、供燃气、通风及空调工程]