Distributed Nested Coevolutionary Architecture and Approach for Multiobjective Scheduling of Agile Earth Observation Satellite

Abstract

As the escalating demands for Earth observation and the increasing diversification of user requirements, agile Earth observation satellites (AEOSs) have been widespread deployed to leverage their superior attitude maneuverability. This leads to a multiobjective AEOS scheduling problem (MO-AEOSSP). The complex time-dependent transition of satellite attitudes between consecutive observations further increases the challenge of the problem. To address the MO-AEOSSP, this article proposes a new approach called the nested distributed coevolutionary algorithm (NDCEA). In particular, an objective-driven heuristic initialization strategy is designed to generate a superior exploration population, and the nondominated individuals are selected to generate an independent population for exploitation. During the evolutionary process, the exploration is executed by a newly devised discrete differential evolution (DDE) algorithm, in which algebraic operators are problem-specifically redefined, and a repair operator is developed to transform the generated infeasible individuals into feasible ones. Exploitation is implemented in a distributed way with the exploration campaign. To fully utilize the superior information of nondominated individuals, the exploitation population is adaptively divided into a convergent population and a diverse population at each generation. They are also coevolved in a distributed way by multiple objective-driven local enhancement strategies. Hence, a nested distributed coevolutionary architecture is constructed and employed in NDCEA. After the exploration and exploitation are completed, a decomposition-based selection strategy is designed to construct superior next-generation populations. Finally, NDCEA is comprehensively assessed against the exact mathematical model on five small-scale instances and compared with state-of-the-art multiobjective algorithms on eight large-scale instances. The experimental results demonstrate the significant superiority of NDCEA over the compared approaches, thereby confirming its efficacy for the considered problem.