基于先验概率更新机制的冷水机组故障检测与诊断分析

Analysis on fault detection and diagnosis of chiller based on prior probability update mechanism

在现有的利用贝叶斯网络(BN)进行冷水机组故障检测和诊断(FDD)中,先验概率更新机制的特点没有被使用。提出了一种先验概率更新机制(UP),将其融入BN模型,用于冷水机组的FDD。UP通过设置更新样本容量对先验概率实时更新,逐步挖掘样本的先验信息,提高FDD性能,同时UP-BN模型能根据现场实际进行模型修正,增加BN模型对现场实际的普适性。使用ASHRAE RP-1043试验数据对所提方法的有效性进行验证,结果表明:UP-BN模型的虚警率降到了1.8%;最大漏检率为制冷剂泄漏(RL)的10%,最大误诊率为RL的3.8%,均控制在可接受的范围内;正常状态和各类故障的FDD正确率基本在90%以上,最低的是RL,但也有86.2%。

In the existing fault detection and diagnosis of chiller based on Bayesian network(BN), the characteristics of prior probability update mechanism have not been used. In this paper, a prior probability update mechanism(UP) is proposed and incorporated into the BN model for fault detection and diagnosis(FDD) of chiller. UP updates the prior probability in real time by setting the updated sample size, gradually mining the prior information of the samples, and improving the FDD performance. Meanwhile, the UP-BN model can be modified according to the actual site, and the universality of the BN model to the actual site can be increased. ASHRAE RP-1043 experimental data were used to verify the effectiveness of the proposed method. The results show that the false alarm rate of UP-BN model is reduced to 1.8%. The maximum missed detection rate was 10% of RL, the maximum misdiagnosis rate was 3.8% of RL. The correct rate of FDD in normal state and all kinds of faults is basically above 90%, the lowest is RL, but also has 86.2%.