Industrial Data Anomaly Detection Method Based On Improved MK And Belief Rule Base Inference

When online detecting the abnormal state of complex industrial system,it is necessary to use various sensors to collect the state information about the system operation,and such information is mostly embodied in the form of time series data.Therefore,in case of sudden change(abnormal)of industrial time series data,it is necessary to find the location of sudden change timely and accurately,and take corresponding measures to ensure the normal operation of the system and avoid major losses caused by safety accidents.The traditional time series data mutation detection method can only detect one of the existing change point(CP)offline,and the detection accuracy is low.In practical industrial applications,time series data often has multiple CPs.Therefore,this paper first introduces the memory and forgetting strategies in cognitive science to improve the original Mann-Kendall(MK)method,and makes it suitable for online detection of multiple CPs in time series data;Then,the inference method of the belief rule base(BRB)in expert system is introduced to improve the online detection accuracy;Finally,the anomaly detection problem of multivariate time series data is further explored.The main research work is as follows:(1)Improved MK change point online detection method based on memory and forgetting strategy.Based on the original MK method,memory and forgetting strategies in cognitive science are introduced,and two kinds of statistical constraints are added to improve the original MK method,so as to achieve the purpose of online detection of multiple CPs in time series data.Finally,the effectiveness of the proposed method is verified by an online detection example of abnormal state in blast furnace ironmaking industrial process.(2)Improved online detection method of MK change point based on Belief Rule Base(BRB)reasoning.Based on the improved MK method studied in(1),the BRB reasoning method in the expert system is introduced to flexibly select the significance level in the MK method.Specifically,the BRB system is constructed to dynamically adjust the significance level in the MK method by taking the mean,variance and standard deviation of normal and abnormal data in the time series data as input,so as to reduce the detection error of CPs and improve the detection accuracy.Finally,the effectiveness of the proposed method is demonstrated by the on-line detection experiment of the abnormal state of the motor rotor.(3)Abnormal detection method based on improved MK and qualitative trend analysis(QTA).The improved MK method is used to detect the CPs in the time series data of a each process variable.The QTA method is introduced to obtain the change direction of each process variables,and the anomaly of multivariate time series data is analyzed by combining the CPs and change direction.Finally,the effectiveness of the proposed method is verified by simulation experiments.