On the computation of optimal strategies for large scale control problems

This paper studies the numerical feasibility of the decomposition method of large scale optimal control problems proposed in [3]. A parallel algorithm based on the second order method is developed and compared with the first order gradient method. Since the convergence speed of low level sub-problems plays an important role in the computation time of the overall problem, a heuristics is proposed to improve the low level convergence speed. A numerical example is used to compare the second order method with the first order gradient method as well as the one level gradient method. It is also used to demonstrate the improvement of the computational efficiency for the low level subproblems.