An adaptive memetic algorithm for multi-UAV cooperative task assignment under complex constraints

Multiple unmanned aerial vehicle (multi-UAV) cooperative task assignment is a critical technology for large-scale reconnaissance and strike operations. In realistic combat environments, this problem is subject to complex constraints, including task precedence relations, simultaneous arrival requirements, task-specific resource demands, per-UAV resource capacity limits, and capability-requirement matching. However, most existing studies consider only subsets of these constraints or oversimplify their interactions, often resulting in infeasible or suboptimal task assignment solutions. To address these limitations, we formulate a mixed-variable multi-UAV cooperative task assignment model that incorporates all the aforementioned constraints and optimizes a weighted objective function comprising total flight distance, maximum task completion time, and a cost-effectiveness ratio. To efficiently solve the proposed model, we develop an adaptive memetic algorithm (AMA). Specifically, a target-bundled encoding scheme combined with a constraint-aware initialization strategy is introduced to generate high-quality feasible initial solutions. To enhance global exploration capability, a module-based crossover operator strategy and a sparrow-search-inspired mutation operator strategy are designed to effectively increase population diversity. Promising solutions are subsequently refined using knowledge-specific local search operators, with an adaptive operator selection mechanism dynamically choosing appropriate operators and a dual-space adaptive frequency control strategy determining when to activate local search based on feedback from both the decision and objective spaces. Experimental results on 100 test instances with varying numbers of UAVs and tasks demonstrate that AMA consistently outperforms three prevailing methods, achieving a 65%–92% reduction in average relative percentage deviation, thereby confirming its effectiveness and robustness.