Abstract
In this paper, a piecewise-affine direct virtual sensor is proposed for the estimation of unmeasured outputs of nonlinear systems whose dynamical model is unknown. In order to overcome the lack of a model, the virtual sensor is designed directly from measured inputs and outputs. The proposed approach generalizes a previous contribution, allowing one to design lower-complexity estimators. Indeed, the reduced-complexity approach strongly reduces the effect of the so-called “curse of dimensionality”, and can be applied to relatively high-order systems, while enjoying all the convergence and optimality properties of the original approach.