Automatic Initial Value Generation with Interval Arithmetic for Nonlinear Process Models

Abstract We present a novel approach for the initialization of nonlinear and non-convex algebraic equation systems. The method reduces the solution space spanned by initial variable bounds through a box consistency test and removes its discontinuities. Subsequently, a numerical solver is started from the midpoint(s) of the reduced search space to find all problem solutions. The proposed approach is applied on three different examples related to chemical engineering processes with the dimensions of 3, 28 and 95. While the method already solves the smallest example, it provides good initial points for the other two in the order of a few minutes based on simple initial bound selections.