基于GP模型和无监督SVM算法的冷水机组的故障检测与诊断

Fault Detection and Diagnosis of Chillers Based on GP Model and Unsupervised SVM Algorithm

介绍了一种基于机器学习的故障检测算法,该算法利用非线性回归器估计被测组件对外部变量的响应,并将其应用于暖通空调冷水机组的故障检测。基于参数先验和观测值的条件似然,采用高斯过程(GP)回归算法建立可预测的概率模型。然后,预测误差与估计方差作为支持向量机(SVM)异常检测器的输入,该检测器能够以无监督的方式检测冷水机组的故障,从而实现在线故障诊断。从实验中可以看出,该算法改进了标准异常检测。

This paper introduces a fault detection algorithm based on machine learning.The algorithm uses a nonlinear regressor to estimate the response of the tested component to external variables,and applies it to the fault detection of HVAC chillers.Based on the prior parameters and the conditional likelihood of observations,a predictable probability model is established by Gaussian process(GP)regression algorithm.Then,the prediction error and estimation variance are used as the input of support vector machine(SVM)anomaly detector,which can detect the fault of chiller in an unsupervised way,so as to realize on-line fault diagnosis.It can be seen from the experiment that the algorithm improves the standard anomaly detection.