A differential geometric approach for parametric nonlinear estimation

A new approach for nonlinear parameter estimation is presented. The paper describes an algorithm which is constructed on differential geometric properties of a nonconvex objective function in an extended space, composed by parameters and time. The parameter search mechanism uses a gradient based law, together with tensorial information provided by a polynomial model which captures the geometry of the cost functional in a relative broad region. The proposed algorithm shows ability for climbing hills of existing folded slots to reach the straight slot of the cost functional. A selected example exhibits the features of the proposed approach