In chemical plants, soft sensors have been widely used to estimate difficult-to-measure process variables online. The predictive accuracy of soft sensors decreases due to changes in the state of chemical plants, and soft sensor models based on time difference (TD) have been constructed for reducing the effects of deterioration with age such as the drift. However, details on models based on TD (TD models) remain to be clarified. In this study, therefore, TD models were discussed in terms of noise and variance in data, auto-correlation in process variables, degree of model accuracy, and so on. Then, we theoretically clarified and formulated the difference of predictive accuracy between normal models and TD models. The relationships and the formulas of TD were verified through the analysis of simulation data. Furthermore, we analyzed dynamic simulation data with considering observed disturbances and unobserved disturbances, and confirmed that predictive accuracy of TD models increased by setting appropriate intervals of TD.
View full abstract