An automatic initialization procedure in parameter estimation problems with parameter-affine dynamic models

This paper proposes an initialization approach for parameter estimation problems (PEPs) involving parameter-affine dynamic models. By using the state measurements, the nonconvex PEP is modified such that a convex approximation to the original PEP is obtained. The modified problem is solved by convex optimization methods yielding an approximate solution to the original PEP. The approximate solution can be further refined by linearizing the original problem around the obtained minimum. An assessment of the distance between the real solution and the one provided by the linearization of the problem around the convex approximation is presented. The optimum obtained by the convex approximation is used to subsequently initialize a simultaneous Gauss–Newton (SGN) approach on the original nonconvex PEP. Comparative results for the SGN with arbitrary initialization and with the proposed approach are presented using three benchmark examples in the chemical and biological fields.