(92c) A Modified Extended Recursive Least-Squares Method for Identification of Fir Type Models from Closed-Loop Data

Performing a plant test under closed-loop conditions can be desirable for model identification. Production loss, resulting in loss of profits, and safety problems may result when control loops are opened for plant testing. However, identification of models from closed-loop data is more difficult compared to models that are identified from open-loop data due to the correlation between the colored noise and the process inputs created by the feedback. This correlation creates a bias in the estimated process model parameters when the normal least-squares estimator is used. The batch solution for the least-squares estimate assumes the prediction error to be zero mean white noise. This assumption fails for closed-loop data. A new method for identifying FIR type models using closed-loop data is proposed. A time-varying bias term with a moving average process is introduced into the model structure and identification is performed using a modified extended recursive least-squares algorithm which eliminates the bias from the process parameter estimates. Once this bias model is introduced, the open-loop step response model (SRM) can be estimated from the closed-loop data which can be used for control purposes. Simulation case studies on a fluidized catalytic cracking (FCC) unit with DMCPlus© as the controller were conducted to support the proposed method.