Evolving Constructive Heuristics for Agile Earth Observing Satellite Scheduling Problem with Genetic Programming

Abstract

Agile Earth Observing Satellite (AEOS) scheduling problem (AEOSSP) consists in selecting a subset of tasks from a given task set which are then scheduled on the agile satellite with the purpose of maximizing the total reward of scheduled tasks. AEOSSP is strongly NP-hard and therefore existing solution approaches mainly fall in the field of heuristics and metaheuristics. According to the no free lunch theory, it is impossible to find a single heuristic that is well-applied to any problem instance and a problem-tailored heuristic is always needed. In this paper, we propose a genetic programming based evolutionary approach (GPEA) to automatically evolve a best-suited constructive heuristic for any given AEOSSP instance. The programs (individuals) of GPEA are heuristic rules encoded as trees of mathematical functions. The fitness of the program is evaluated through mapping the mathematical function to an AEOSSP solution using a timeline-based construction algorithm. Computational results on a set of well-designed AEOSSP scenarios show that the proposed GPEA leads to a heuristic algorithm that outperforms recently published sophisticated meta-heuristic algorithm (ALNS). Additional experiments were carried out to demonstrate that the timeline based construction algorithm plays a significant role in matching time-related characteristics in comparison to four commonly used heuristic algorithms. Our results also showed that the evolved heuristic rules preserve a certain extent of generality.