A parallel algorithm for temporal decomposition of long horizon optimal control problems

This paper presents a solution methodology for the temporal decomposition of a class of discrete time optimal control problems. By applying the temporal decomposition, a long horizon optimal control problem is converted into a two-level optimization problem where the high-level problem involves parameter optimization and low-level subproblems are optimal control problems with shorter time horizon. A two-level optimization algorithm with parallel processing structure is developed. It adopts Newton's method for the high level and the Differential Dynamic Programming (DDP) for the low level. Issues for integrating the two optimization techniques into one efficient two-level parallel algorithm are investigated. Numerical testing results show potential of this approach in solving long-horizon problems with parallel processors.