基于传感器关联网络的燃气轮机异常检测方法

Anomaly detection method based on gas turbine sensor associated network

通过分析燃气轮机4种典型异常形式的产生机理以及特征表现,得到了不同异常形式与传感器关联网络结构特征之间的映射关系。在此基础上,得到传感器关联网络的异常特征模式,提出了基于传感器关联网络的燃气轮机异常检测策略。通过实例分析,证明了融合多源信息基础上建立的传感器关联网络模型,可以过滤掉节点间关联指标低于阈值0.37的相关性,有效的实现燃气轮机的稳态异常检测,并可以检测出转速上升过程中,节点间相关性的正常线性变化趋势,存在大于12%的异常凹陷非线性趋势,从而有效的实现燃气轮机的动态异常检测。

The mechanism and characteristics of four typical abnormal forms of gas tur- bine were analyzed, and the mapping between the different abnormal features and the sensor network characteristics were acquired. On this basis, the abnormal characteristics of sensor network were obtained, and the detection strategy of gas turbine anomaly based on sensor correlation network was proposed. The results of experiments indicated that sensor related network model based on multi source information fusion could filter out the correlation be- tween nodes when the correlation indicator was below the threshold of 0.37, and achieve steady-state anomaly detection of gas turbine effectively, showing that there was a non-linear trend of abnormal depression greater than 12 % during the speed up; although the correlation between nodes under normal circumstances should be a linear trend, this method could a- chieve dynamic anomaly detection of gas turbine effectively.