Automatic Database Monitoring for Process Control Systems

In industrial plants, soft sensors are widely used to predict difficult-to-measure process variables for process control. One of the problems of soft sensors is the degradation of the soft sensor models. The predictive accuracy decreases when the states in plants change. Many adaptive soft sensor models have been developed to reduce the degradation. Since the database management is required for those adaptive models, we previously proposed the database monitoring index DMI and the database managing method with the DMI. By judging whether new data should be stored in database or not compareing the DMI values, the amount of information can increase while controlling the number of data in the database. In this study, we proposed the automatic method determining the threshold of DMI with only training data. The model construction and data deletion are repeated while checking the DMI values. Through the analysis of simulation data and real industrial data, we confirmed that the proposed method can monitor the database appropriately and the predictive accuracy of the adaptive soft sensor models improve.