Optimal control of a class of discrete-continuous nonlinear systems -- decomposition and hierarchical structure (Q5950249)

A decomposition-coordination technique for solving a nonlinear optimal control problem (NOP) is derived. The state-space NOP is decomposed into a set of linear quadratic (LQ) subproblems. The parameters and modifiers of subproblems are updated at each decomposition-coordination iteration. It is shown that the process converges to the solution of the original NOP in spite of LQ simplifications. The calculation of modifiers is based on the derivatives of model variables, which are difficult to estimate during iterations when intermediate solutions are infeasible. It is demonstrated how this limitation can be alleviated by relevant approximation methods. The effectiveness of the technique is demonstrated on a multivariable continuous-state NOP example with a discrete-time non-separable and partly non-quadratic performance index.