Combinatorial optimization techniques for spacecraft scheduling automation

Abstract

Most spacecraft have extremely limited resources compared with the state of the art in computer science and most missions have ambitious scientific goals, such as in the case of fly-bys like Voyager and Ulysses where there are limited windows of opportunity for achieving these goals. As a result, the scheduling of spacecraft experiments is a complete NP-hard problem for which an efficient solution procedure producing acceptable results is yet to be found. We describe the application of combinatorial optimization techniques to the automatic spacecraft scheduling problem. The sequencing problem is the search over the candidate sequences of experiments for a sequence that maximizes thevalue of the science conducted while minimizing constraint conflicts. A voluminous literature exists on planning and scheduling problems. Early approaches to solving these problems focused on analytical models, where the models were often based on techniques from mathematical programming and stochastic processes. While numerous advances have been made, many researchers have looked towards less traditional approaches to problems of this nature in order to solve the large-scale problems often encountered in practical applications. These approaches have still been based on the formulation of a mathematical model; however, heuristic procedures have been used to generate schedules that are considered good but not guaranteed to be optimal. There are several solution approaches that we believe can be successfully integrated with mathematical programming techniques to create a new solution paradigm addressing large-scale spacecraft scheduling optimization problems. These are Simulated Annealing (SA), Random Search, Tabu Search, and a diversification technique for Random Hill Climbing termed Strategic Oscillation. The power of these techniques lies in their ability to treat the objective function as a black box that returns a measure(s) of the goodness of the sequence; that is, these algorithms do not require a closed-form analytical description of the objective function nor any function derivatives. The algorithms we developed were evaluated by testing scheduling problems. Such testing on reduced size problems indicates the practicality of the proposed algorithms from a performance and timing perspective, while eliminating the need for sophisticated new sequence evaluation software to be developed or the need to integrate the new automation techniques into current software. Our exploratory computational results indicate that pseudo-random search techniques can generate viable sequences in reasonable times. Although the spacecraft scheduling problems used for testing were not based on actual experiment request data, they did match the data structures of actual problems with regard to size and constraint types. Therefore, we believe that the techniques described in this paper present viable approaches to spacecraft sequencing problems.