Resource Allocation and Scheduling For Training

Pedro Szekely

University of Southern California, Information Sciences Institute (USC/ISI)

28 May 28, 2004

USC/ISI offers novel resource allocation and scheduling technology that gives users the ability to explore options and balance multiple concerns. We have successfully applied this technology to scheduling both training and combat sorties for AV8B jets. The technology has been fielded with the Marines Air Group 13 in Yuma, Arizona, and on board several amphibious attack ships conducting operations in the Middle East and Japan.

Conventional scheduling systems force users to specify an objective function that measures the quality of a schedule in terms of a single number calculated as a function of multiple features of a schedule. These systems produce a schedule that maximizes the value of the objective function. The problem with this approach is that it is difficult, if not impossible, to boil down multiple concerns into a single number. Suppose that the requirements are to train Jones on night systems and Smith on low altitude tactics by the end of the month. If limitations on resources don’t allow both to be accomplished how should the scheduler decide what to do? In the conventional objective function approach one would have to give weights to each requirement, say 80% for Jones and 20% for Smith, which would lead the scheduler to give Jones more priority, but does not address the real world situation. Is it better to delay both by a week, or finish Jones by the end of the month and Smith by the 15th of the next month? Given many requirements, the objective function approach does not give the user a sense of what is possible, and what trade-offs can be achieved.

In contrast, the USC/ISI technology provides an interactive decision support environment where the system gives users information about feasible schedules and thus gives the user the ability to define and explore meaningful trade-offs. With the USC/ISI technology the user would be able to explore different options for Jones and Smith and see the consequences on other training and operational requirements. Perhaps delaying Smith for 2 weeks is feasible because he would still be able to meet the training requirements for his deployment, whereas Jones has other training requirements to meet before his deployment. Our experience with the Marines is that this ability to explore trade-offs enables users make better decisions and allows them to present viable options to commanders when the situation makes it necessary to do so.

The USC/ISI technology has a rich model that enables it to reason about extensive details that affect scheduling decisions in a training situation. The model encompasses the following concerns:

  • Training manual: the system has a detailed understanding of training codes, pre-requisites, qualifications, and currency. The training manual enables the system to reason about all the tasks that must be accomplished to fulfill a requirement and the sequencing decisions that must be observed.
  • Resources: the system models all resources required to perform a task. This includes equipment (aircraft, simulators, ranges, etc.) as well as instructors. The training manual enables the system to determine what type of instructor is required for each task depending on the current capabilities of the trainee and the particular task to be performed.
  • Constraints: the system models an extensive number of constraints including constrains on when and how tasks can be accomplished (e.g., LUX levels to perform night sorties), constraints on the use of equipment (e.g., type of aircraft or simulator required for each task, turn around times for equipment), crew day and crew rest constraints for all personnel involved, and preferences on the most appropriate type of equipment for each task.

The system has been used to produce detailed squadron-level schedules (24 pilots, 16 aircraft, 30+ ranges, 3 simulators) with planning horizons of a week or a month, producing schedules at 1 minute resolution.

We propose to adapt our existing technology to other training domains as appropriate, and thus give users the ability to solve complex scheduling problems, to explore multiple alternatives, to make informed trade-offs and to offer commanders the justification of proposed options.

1