摘要
数据驱动的冷水机组模型通常只用于专有对象,当涉及到一种不同类型的冷水机组时,需要大量的正常数据和故障数据训练一个新的模型,这既耗时又依赖大量数据,不利于冷水机组故障诊断技术在实际应用中的推广。本文的研究中,介绍了处理不平衡数据技术,探索将离心式冷水机组训练成一个能够诊断螺杆式冷水机组故障的新模型的可能性,即只需使用少量新数据即可。采用合成少数类过采样技术(SMOTE)对故障样本进行过采样,不平衡率为5%,然后使用支持向量机(SVM)用于故障诊断。通过对50%~400%的过采样率的研究,发现螺杆式冷水机组四种故障的过采样率为100%,诊断准确率达到96.70%。
Data driven diagnostic model for refrigeration systems is often used exclusively to a dedicated object. When it comes to a different type of chiller, a new model must be trained with large among of normal and faulty data, which is both time-consuming and heavily data-depending,and accordingly, curbs its application. In this study, the technology for the tackling of imbalanced data was introduced to probe the possibility of extrapolating an old model trained for a centrifugal chiller to a new one that can diagnose the faults of a screw chiller, by just using small amount of new data. Synthetic Minority Oversampling Technique(SMOTE) is used to oversample the fault sample set with an unbalance ratio of 5% and support vector machine(SVM) is employed for fault diagnosis.By investigating oversampling ratios between 50% and 400%, it was found that the ratio of 100%was the best and the diagnostic accuracy reached 96.70% for the four types of faults of the screw chiller.
作者
范雨强
崔晓钰
韩华
陆海龙
武浩
徐玲
FAN Yu-Qiang;CUI Xiao-Yu;HAN Hua;LU Hai-Long;WU Hao;XU Ling(School of Energy and Power Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China)
出处
《工程热物理学报》
EI
CAS
CSCD
北大核心
2019年第6期1219-1228,共10页
Journal of Engineering Thermophysics
基金
国家自然科学基金资助项目(No.51506125)
