Abstract
In this paper, a Prediction Error based algorithm is developed for the identification of a Wiener system, a linear dynamic subsystem followed by a static non-linearity, in the presence of a non-stationary disturbance added before the static non-linearity. This structure represents a non-stationary process disturbance, as is common in chemical process control applications. In the proposed method, the disturbance part is restructured to be stationary by differencing the input signal and implicit differencing of the unmeasured intermediate signal. First, an approximate but linear in the variables model is fitted, and used to construct an initial estimate of a parametric system model. This is then refined using a quasi-Newton optimization. Finally, Monte-Carlo simulation is used demonstrate the performance of the algorithm.