A Modified Extended Recursive Least-Squares Method for Closed-Loop Identification of FIR Models

摘要

Performing a plant test under closed-loop conditions is desirable for model identification because production loss and safety problems may result when control loops are opened during plant testing. However, identification of models from closed-loop data is more difficult compared to identifying models from open-loop data because of the correlation between the colored noise and the process inputs created by the feedback. A novel method for identifying models using closed-loop data is proposed, which employs a time-varying bias term with a moving average dynamic component as the model structure. Then identification for this process model is performed using a modified extended recursive least-squares algorithm to eliminate the bias from the process parameter estimates. Evaluation of the proposed algorithm is performed using simulation case studies involving multivariable processes controlled by either diagonal PI controllers or a model predictive controller (DMCPlus). The simulation results showed that the proposed method is robust with respect to measurement noise and is able to identify high quality models from closed-loop data.