Closed-loop identification of an unmeasured disturbance model for control performance assessment and improvement

摘要

This invention defines a calculation method to identify a class of unmeasured disturbance models of any process variable under closed-loop control. The calculation uses only normal closed-loop data. It does not require any identification test to dither the set point or the manipulated variable of the underlying controller. The method first identifies a Finite Impulse Response (FIR) model of the control errors over a window of data set. For a specific class of disturbance models, these FIR model weights, after a number of time lags, have a uniform property that is independent of closed-loop control. The existence of such a property is owing to the fact that any well-performing controller compensates for its past control errors over a fractional number of the FIR weights. This property is then converted to model parameters of the unmeasured disturbance. This method repeats itself with a moving window of data set to identify a time-varying disturbance model.