On the design of a continuous‐time model estimator for bias distribution

Modelling of physical processes always involves approximations. This undermodelling, coupled with noisy process data, results in biased frequency response estimates of models. The bias error cannot be eliminated, but can be distributed over frequencies by careful design of the estimator. This paper aims at analysing the bias error in the estimates from an equation-error algorithm based on τ-time-ahead prediction using a continuous model of the process with filtered input/output signals. From the analysis, the roles played by the design variables of the estimator in affecting the bias distribution are identified and the choice of these is discussed. Finally, a procedure to optimize the design variables is indicated. Numerical simulation results are presented.