Early Identification of Process Deviation Based on the Statistical Feature of Prediction Residuals

With the application of well-developed distribution control system (DCS) in process industry, large amount of data have been collected. Under a specific normal operating condition, correlation among variables are relatively stable. Once the correlation changes, it indicates that the process has deviated from its original operating state.

Residual, refers to the difference between the predicted value and the actual value of a regression model. For a stead state operation, the system can be well approximated by a mathematical model, usually linear regression model adopted for calculation efficiency, with a modelling residuals, which show a statistical distribution of Gaussian, and is considered as signal noise. Once the residuals are no longer in accordance with Gaussian distribution, it indicates that the prediction residual is contributed by both data noise and process deviation. In other words, due to the nonlinearity of process dynamics, the correlation among the variables described for the specific steady state is not valid any more. Therefore, it can be reasonably expected to identify the change of operation state by monitoring statistical feature in prediction residuals.

In this work, a process monitoring method based on the statistical feature of prediction residuals is proposed. Based on the topology of a process, variables are selected to establish a regression model to extract the correlation for a steady state operation, especially those with spatial correlations regarding certain unit or unit group. By monitoring statistical feature change in prediction residuals, early identification of process deviation can be achieved, which can effectively avoid the impact of data noise on monitoring results. Data from Tennessee Eastman process and a pre-reforming reactor for hydrogen production are investigated to validate the proposed methods. The results show that the process deviation can be detected at its early stage.