AMCATS: An adaptive rule-guided framework for dynamic multi-task allocation in heterogeneous satellite systems

Abstract

Dynamic multitask allocation in heterogeneous environments must reconcile global throughput with the strict protection of urgent tasks under online arrivals, coupled resources, and tight time window constraints. Existing heuristic, learning based, and hybrid schedulers typically address isolated problem facets, degrading when time window pressure and emergency interference act simultaneously. To overcome this, this paper proposes AMCATS, an adaptive rule-guided framework for dynamic multi-task allocation in heterogeneous satellite systems. AMCATS resolves the conflict between throughput and emergency responsiveness through a structural rule layer combining feasibility screening, soft reservation, selective deferral, and bounded preemption, scored by an interpretable candidate-utility function. An optional learning prior may be fused with the rule score but is disabled in all experiments reported here, the results are therefore attributable solely to the rule layer. For reproducible evaluation, a controllable dynamic benchmark named DSMT-Gen is constructed across balanced task flow, window-stressed scheduling, and burst emergency scheduling scenarios. Under a corrected protocol in which all baselines share the same bounded-preemption primitive, experiments over ten random seeds, ablation studies, and transfer validation on EOSSP-MRT instances show that AMCATS attains the best value on every primary metric in all scenarios. It sustains an emergency-task completion rate above 99% under burst stress, exceeding the strongest equally-equipped baseline by more than five percentage points while achieving the lowest average waiting time among the leading methods, and transfers to EOSSP-MRT instances without loss of completion. Coupling structural scheduling knowledge with rule-guided coordination yields a robust, interpretable, and transferable solution for dynamic multi-task allocation.