基于LS-SVM的制冷系统故障诊断

Fault Diagnosis for Refrigeration System Based on LS-SVM

为了提高制冷系统故障诊断速度及准确性,提出了基于最小二乘支持向量机(LS-SVM)的制冷系统故障诊断模型,并采用ASHRAE制冷系统故障模拟实验数据进行模型训练与验证.对一台90冷吨(约316 kW)的离心式冷水机组的7类制冷循环典型故障进行了实验.研究结果表明,LS-SVM模型对制冷系统七类故障的总体诊断正确率比支持向量机(SVM)诊断模型、误差反向传播(BP)神经网络诊断模型分别提高0.12%和1.32%;尽管对个别局部故障(冷凝器结垢、冷凝器水流量不足、制冷剂含不凝性气体)的诊断性能较SVM模型的略有下降,但对系统故障的诊断性能均有较大改善,特别是对制冷剂泄漏/不足故障;诊断耗时比SVM模型减少近一半,快速性亦有所改善.可见,LS-SVM模型在制冷系统故障诊断中具有良好的应用前景.

In order to improve the fault diagnosis speed and accuracy for refrigeration system, a fault diagnosis model based on least squares support vector machine （LS-SVM） was proposed. American Society of Heating, Refrigerating, and Air-conditioning Engineering （ASHRAE） refrigeration system fault simulation data was used for the model training and validation. The experiments of a centrifugal chiller of 90 tons with seven types of typical faults were conducted. The results showed that the overall diagnostic accuracy of LS-SVM model for seven types of faults increased by 0. 12% and 1.32% respectively, compared with support vector machine（SVM） diagnosis model and error back-propagation （BP） neural network model. Although diagnostic performance of LS-SVM model for individual component-level fault（ConFoul/ReduCF/NonCon） was low slightly compared with SVM model, the diagnosis performance for system-level were greatly improved, especially for refrigerant leakage or lack of refrigerant. The diagnosis time of LS-SVM model reduced nearly half than that of SVM model. At the same time, its rapidity improved. Therefore, LS-SVM diagnostic model had good application in the fault diagnosis of refrigeration system.