Dynamic Data Reconciliation for Enhanced Control Performance of Systems with Data-Driven Model

Effective control of chemical processes is an important prerequisite for ensuring production benefits. Traditionally, the influence of feedback measurement noise is not considered in the control performance assessment (CPA). In this work, a dynamic data reconciliation (DDR) filtering method is develop in a system with unknown process model to enhance the CPA results impaired by measurement noise. Firstly, the interference of measurement noise on CPA results is derived. Then, for the unknown process model, as a typical data-driven model, the long short-term memory network is used for process modeling to obtain model information. Finally, based on the measurement feedback information and model information, DDR uses Bayesian inference to obtain an optimal feedback information to enhance the reliability of CPA results. The control performance enhancement effect is demonstrated on the control process of a continuous stirred tank reactor process.