Math 4/5779: Mathematics Clinic (Spring 2006)

Satellite Payload Scheduling with Dynamic Tasking

sponsored by: Raytheon

Problem Description: 

The Satellite Mission Scheduling problem with Dynamic Tasking (SMS-DT) involves scheduling tasks to be performed by a satellite, where new task requests can arrive at any time,  non-deterministically, and must be scheduled in real-time.  In this project, we focus on satellite data collection systems, such as those involved in remote sensing.  Here, a task involves  taking a picture of a certain location (target) on the Earth's surface.  Performing a task involves moving a camera into position, and then taking the picture.  Both of these operations take time, and since there are typically many task requests, not all requests can be serviced.  Moreover, the time required to move the camera between two successive targets depends on the relative positions of the targets.  Thus, the order in which tasks are performed greatly influences the efficiency of the schedule.   This report investigates algorithmic approaches for determining an optimal or near-optimal sequence of tasks, allocated to a satellite payload over time, with dynamic tasking considerations.   A detailed description of the problem can be found [1].

The SMS-DT is, both practically and theoretically, an extremely difficult optimization problem.  Even without the dynamic tasking considerations, the (static) Satellite Mission Scheduling  problem (SMS) is an NP-hard combinatorial optimization problem.  NP-hard problems are intractable to solve exactly (except for small problem sizes), so must be tackled with heuristic  techniques.  Many such heuristic approaches have been studied in the literature, but the best choice of method is highly problem dependent; what works well for one class of problems may  perform poorly for another.  In short, the algorithmic approach must be carefully designed to exploit any known structure of the problem at hand.

By adding the dynamic tasking requirements, this NP-hard combinatorial optimization problem becomes even more complex because it also involves optimization under uncertainty.  In particular, scheduling decisions have to be made before all information is known. Therefore the schedule needs to be flexible enough to accomodate dynamic task requests when they  become known.

The goal of this research effort is the creation of an algorithm to determine an optimal or near-optimal sequence of tasks, allocated to a satellite payload over (discrete) time, with dynamic tasking considerations.