Nonlinear distributed dynamic optimization based on first order sensitivities

A method for the distributed optimization of dynamic nonlinear systems is presented. The method is based on partial goal-interaction operators. Partial goal-interaction operators provide gradient information of non-local objective functions. Hence, these operators are used to modify the objective functions of infimal optimization problems in order to take non-local information into account and to achieve an optimum for the overall objective, i.e. the objective of the entire process. However, that optimum is achieved by a decentralized but cooperative optimization, while communication between the different infimal optimization units is limited. An important part of the method is the decentralized calculation of sensitivities. The method is applied to a nonlinear differential-algebraic simple-toy system and compared to the dual-optimization method as well as to the solutions of fully centralized and fully decentralized optimizations.