制冷系统故障可由多种模型进行模拟诊断。为了提高其诊断性能，将包括K近邻模型（KNN）、支持向量机（SVM）、决策树模型（DT）、随机森林模型（RF）及逻辑斯谛回归模型（LR）在内的5种成员诊断器，通过绝对多数投票方法集成为一个集成模型，并采用美国采暖、制冷与空调工程师学会（ASHRAE）故障数据对1台90冷吨的离心式冷水机组进行建模及验证，数据包含制冷系统的7类典型故障及一类正常运行。结果表明：集成模型在所选数据集上总体诊断正确率达到99.58%，较各成员诊断器（94.55%~99.05%）均有显著提升，对正常运行、局部故障及全局故障的诊断性能亦有改善。此外，对比分析了不同集成模型及成员诊断器的诊断性能，从中找到诊断正确率与时间成本最佳的集成模型（99.41%，1.34 s）。可见，集成模型较单一模型性能更佳，在制冷系统故障诊断中具有更好的应用前景。

Faults of refrigeration system can be simulated and diagnosed by various models.In order to improve the diagnostic accuracy,this paper integrates several ensemble members into an integrated diagnostic model by means of multiple voting,including KNearest Neighbor (KNN),Support Vector Machine (SVM),Decision Tree (DT),Random Forest (RF) and Logistic Regression (LR).The ASHRAE (American Society of Heating,Refrigerating and AirConditioning Engineers) fault simulation data is used to establish and verify the model.The data includes normal operation and seven typical faults of a 90ton centrifugal chiller under 27 operating conditions.The results show that the overall diagnostic accuracy of the integrated model reaches 99.58%,which is significantly improved from that of the ensemble members (94.55%~99.05%).The diagnostic performance of the integrated model is also improved for normal operation,local faults and system faults.In addition,different integrated models and ensemble members are analyzed and compared,and the best integrated model with the best compromise between diagnostic accuracy and time cost is found (99.41%,1.34 s).It is demonstrated that the integrated model has better performance than the single model and has better application prospects in the fault diagnosis of refrigeration system.

TB657

国家自然科学基金(51506125)