Random-direction optimization algorithms with applications to threshold controls (Q5947270)

This work presents an optimization algorithm for solving problems of optimal control where the set of admissible controls is restricted to policies of threshold type. The procedure under discussion uses a random-direction gradient method to optimize the threshold levels. The idea is to collect all random effects of the dynamic system in a process which is supposed to be stationary. At each step, the threshold levels are corrected by a random vector and projected onto a constrained set. The convergence is established and the convergence rate is obtained. The authors emphasize that their method is recursive and easy to implement. An example of production planning in manufacturing is examined.