Abstract: The traditional 3σ criteria is widely used in the identification of abnormal data in monitoring sequences,but this criteria is only applicable to data sequences satisfying normal or approximate normal distribution.However,since the quality of monitoring data is easily affected by the environment,equipment and external forces,monitoring sequences with outliers will inevitably appear.For "step" and "shock" shapedsequences with outliers,traditional methods are prone to make misjudgment.Therefore,by introducing the membership cloud generator and combining with the 3b criteria of the subordinate cloud,this paper preliminarily screened the monitoring data,and then the control function was constructed by the expectation of the monitoring data sequence and the mean value of the bandwidth sequence.Therefore,an anomaly recognition algorithm based on membership cloud was proposed and applied to the displacement and seepage monitoring of Gongzui Hydropower Station.The application results showed that this method was significantly improved in the accuracy and reliability of anomaly recognition compared with the traditional criterion,and the relevant experience could be used as a reference for similar engineering safety monitoring.