冷水机组多故障耦合研究

Research on multiple faults coupling of water chilling packages

为了提高冷水机组的运行效率、设备可靠性和能源利用率,本研究将对冷水机组的多故障耦合进行检测和诊断。首先,本研究使用RP-1043(Research Promotion)项目的故障数据,对冷水机组几种典型故障数据进行分析对比。其次,本研究使用3种树模型对数据进行训练,发现随机森林在准确率和训练预测的效率上综合表现最好。使用随机森林模型,结合专家知识,对故障等级为1的冷水机组的运行数据进行特征选取,然后建立贝叶斯网络故障诊断模型。最后,使用该模型对实际故障案例进行诊断与分析,对比附加信息层对故障诊断的影响。结果表明,该模型仅使用故障特征节点便可以对故障进行有效的诊断,合理利用附加信息层可以进一步提高故障诊断的可靠性。

In order to improve the operating efficiency,equipment reliability and energy efficiency of the water chilling packages,the detection and diagnosis on the multiple faults coupling of the water chilling packages is conducted.First,using the failure data of the RP-1043 project,several typical failure data of the water chilling packages are analyzed contrastively.Secondly,three tree models are used to train the data,and it’s found that the random forest has the best overall performance in terms of accuracy and training prediction efficiency.Using the random forest model,combined with expert knowledge,the feature selection of water chilling packages operating data with fault level 1 is conducted,and then the Bayesian network fault diagnosis model is set up.Finally,the model is used to diagnose and analyze actual fault cases,and the influence of additional information layer on fault diagnosis is compared.The results show that the model can effectively diagnose the fault using only the fault characteristic nodes,and the reasonable use of the additional information layer can further improve the reliability of the fault diagnosis.