Learning Search Control Knowledge for Deep Space Network Scheduling

Abstract

While the general class of most scheduling problems is NP–hard in worst–case complexity, in practice, for specific distributions of problems and constraints, domain-specific solutions have been shown to perform in much better than exponential time. Unfortunately, constructing such techniques is a knowledge-intensive and time-consuming process that requires a deep understanding of the domain and the scheduler. The goal of our work is to develop techniques to allow for automated learning of an effective domain-specific search strategy given a general problem solver with a flexible control architecture. In this approach, a learning system searches a space of possible control strategies, using statistics to evaluate performance over the expected problem distribution. We discuss an application of the approach to scheduling satellite communications. Using problem distributions based on actual mission requirements, our approach identified strategies that both decrease the amount of CPU time required to produce schedules, and increase the percentage of problems that are solvable within computational resource limitations.