Planning and Scheduling: Dynamic Assignment and Scheduling with Contingencies

摘要

Planning and scheduling under uncertainty is important in many important AirForce operations. When real time monitoring of the progress of operations is available, it is possible to replan missions in response to contingencies. In this research, we investigated a class of planning and scheduling problems under uncertainty which captured important features of risky multiplatform Air Force missions. Planning and scheduling problems under uncertainty can be solved in principle by stochastic dynamic programming techniques. These techniques require large computations for moderate problems, as they must anticipate all of the possible future events. In this research, we investigated the use of approximate stochastic dynamic programming techniques to obtain near optimal schedules which anticipate future contingencies, and which can replan in response to contingencies. These approximations are based on techniques for obtaining estimates of the future costs associated with current decisions, using techniques such as rollout of heuristic strategies, off line training of approximations, or approaches such as neuro dynamic programming. The results indicate that, in the context of the mathematical problems investigated, the performance of some approximate dynamic programming algorithms is near that of the optimal performance. Further development and evaluation will establish the viability of these techniques for Air Force mission planning problems.