Anomaly Detection from Online Monitoring of System Operations Using Recurrent Neural Network

摘要

In production systems, the operation data is multivariate time series data including internal state of the system, control variables, control parameters and the like. As monitoring centre collects data intensively, monitoring time differs for each system. The predetermined frequency of data recording per day may not be protected. In this study, we decided to use a RNN which can learn data with missing values. The neural network learns diagnosis of system abnormality from the operation data of the system and the data of the maintenance record. Then we examine the usefulness of prediction of abnormal occurrence of learned neural network.