Automatic Generation of Heuristics for Scheduling Robert A. Morris John L. Bresina Stuart M. Rodgers Florida Institute of Technology Recom Technologies Computer Science Department 150 W. University Blvd. NASA Ames Research Center University of West Florida Melbourne, FL 32901, U.S.A. Mail Stop 269-2 Pensacola, FL 32514, U.S.A morris@cs.fit.edu MofFett Field, CA 94035, U.S.A sturodgers@sprintmail.com bresina@ptolemy.arc.nasa.gov Abstract This paper presents a technique, called GENH, that automatically generates search heuristics for scheduling problems. The impetus for de- veloping this technique is the growing consen- sus that heuristics encode advice that is, at best, useful in solving most, or typical, problem instances, and, at worst, useful in solving only a narrowly defined set of instances. In either case, heuristic problem solvers, to be broadly applicable, should have a means of automat- ically adjusting to the idiosyncrasies of each problem instance. GENH generates a search heuristic for a given problem instance by hill- climbing in the space of possible multi-attribute heuristics, where the evaluation of a candidate heuristic is based on the quality of the solution found under its guidance. We present empirical results obtained by applying GENH to the real world problem of telescope observation schedul- ing. These results demonstrate that G E N H is a simple and effective way of improving the per- formance of an heuristic scheduler. 1 Introduction Employing heuristic methods to solve intractable con- strained optimization problems like scheduling often suf- fers from narrowness in the range of problems to which they can be effectively applied. To overcome these lim- itations, researchers have sought to develop more adap- tive approaches to problem solving; e.g., [Gratch and Chien, 1996]. Adaptive heuristic problem solving delays the selection of an heuristic strategy until some informa- tion can be obtained with respect to which strategy is ex- pected to perform most effectively in solving a problem instance or class of instances. For problems which re- quire a solution to satisfy a set of constraints, the "best" heuristic is typically one which is expected to allow the problem solver to most efficiently converge on a solu- tion. For example, the approach exemplified by SOAR [Laird, et a/., 1986] uses traces of past problem-solving efforts to refine heuristics in order to speed up the search for a solution. However, other criteria besides problem solving efficiency for heuristic selection are possible. For scheduling and other constrained optimization problems, quality of solution may be a more crucial metric with which to compare and select heuristics. This paper proposes a technique, GENH, for automat- ically generating heuristics for solving scheduling prob- lems. G E N H generates a search heuristic for a given problem instance by hill-climbing in the space of pos- sible multi-attribute heuristics, where the evaluation of a candidate heuristic is based on the quality of the so- lution found under its guidance. GENH has been suc- cessfully applied to the problem of scheduling telescope observations using the Associate Principal Astronomer, or APA [Drurnmond, et a/., 1994], a system developed at NASA Ames Research Center. Given a set of telescope observation requests supplied by the user of the APA, GENH solves the problem instance using several "ver- sions" of the APA scheduler, where the versions differ only in the search heuristic employed. Adding GENH to the scheduling process incurs acceptable computational overhead, is accurate (i.e., converges to a better solu- tion more often than previously employed techniques), robust (solves a wide range of problem instances), sim- ple (is based on a simple algorithm easily integrated into the scheduling process), and potentially generalizable to other problem domains. Following a brief summary of the problem domain and the APA scheduler in Section 2, we present a description of the GENH method (Section 3) and a summary of experimental results (Section 4). We then discuss related research (Section 5), future research (Section 5) and conclude (Section 6). 2 Telescope observation scheduling The input to the APA observation scheduler is a set of requests, expressed using the Automatic Telescope In- struction Set, or ATis, [Boyd, et a/., 1993]. Each re- quest is composed of a sequence of telescope movements and instrument commands, as well as scheduling con- straints and preferences. An observation request is said to be enabled on a given night if all its constraints are met. The enablement intervalis the duration of enabled time for a request. The enablement interval is deter- mined by the observation season, as well as factors such as position of the moon on a particular night. For fur- 1260 PLANNING AND SCHEDULING