Robust predictor for nonlinear systems based on bounding-error methods

A new robust predictor for nonlinear systems is proposed. The predictor uses a set of system input-output measurements and a local linearization method based on bounded-error to return an interval that bounds the system output. The midpoint of the prediction interval is the optimal solution of an optimization problem which minimizes a quadratic prediction-error functional cost with a regularization term. The width of the prediction interval can be used as a reliability index of this central prediction. Bounded-error methods use an unique error bound applied to all measurements. The main idea of this work is to use a reliability index that provides a different error bound for each measurement. This allows us to apply the proposed method to measurements with outliers or different error bounds. The main contribution of the paper is the explicit expression that provides the prediction interval and assures a low computational effort.