针对灾后基础设施受损条件下异构无人机集群覆盖、转发与汇聚协同困难的问题，建立以速度、航向角和发射功率为控制变量的动态路由规划模型，将无人机运动、地面节点覆盖、链路速率、缓存队列、功能性回传能力和能耗等指标统一到同一时域。针对原始L-SHADE随机初始化缺少任务节点空间信息、参数自适应缺少覆盖状态反馈的不足，提出覆盖状态感知L-SHADE：利用覆盖优先个体改善初始种群，并依据精英覆盖违约、可行个体比例及覆盖变化趋势，在覆盖构造、覆盖稳定、性能精修和覆盖恢复状态间切换，约束p、差分缩放因子F和交叉概率CR。在固定异构环境下进行5次独立运行，并与原始L-SHADE和仅采用覆盖优先初始化的L-SHADE进行比较。5次独立运行的描述性结果显示，所提算法平均覆盖率由0.827 4提高至0.894 3，增加了6.69个百分点，相对增加约8.09%；平均覆盖违约降低约37.64%，平均吞吐量增加约64.25%，平均队列负载降低约19.64%，平均归一化能耗降低约6.0%，由0.000 416降至0.000 391。上述差异反映了当前固定场景和小样本条件下的数值改善趋势，尚不能据此建立统计显著优势。在这一证据边界内，结果表明，将覆盖状态反馈显式引入差分进化搜索，有助于在覆盖保持、数据回传和资源消耗之间形成更协调的搜索方向，可为低空经济与灾后应急场景中异构无人平台的持续通信服务规划提供方法参考。

To address the challenges in coverage， forwarding and aggregation coordination of heterogeneous UAV clusters under damaged post-disaster infrastructure， a dynamic routing planning model is established with speed， heading angle and transmit power as control variables， which unifies UAV mobility， ground node coverage， link rate， buffer queue， functional backhaul capability and energy consumption within the same time domain. To tackle the deficiencies of the original L-SHADE algorithm， namely the lack of spatial information of task nodes in random initialization and the absence of coverage state feedback in parameter adaptation， a coverage-state-aware L-SHADE is proposed. The coverage-prioritized individuals are adopted to optimize the initial population. According to elite coverage violations， the proportion of feasible individuals and the trend of coverage variation， the algorithm switches among coverage construction， coverage stabilization， performance refinement and coverage recovery states， and constrains the p value， differential scaling factor F and crossover probability CR. Five independent runs are carried out in a fixed heterogeneous environment， and comparisons are made with the original L-SHADE and the L-SHADE only with coverage-prioritized initialization. Descriptive results from the five independent runs show that the average coverage rate of the proposed algorithm rises from 0.827 4 to 0.894 3， an increase of 6.69 percentage points or a relative increment of approximately 8.09%； the average coverage violation is reduced by about 37.64%， the average throughput increases by roughly 64.25%， the average queue load decreases by approximately 19.64%， and the average normalized energy consumption drops by about 6.0%， from 0.000 416 to 0.000 391. The above differences reflect the trend of numerical improvement under the current fixed scenario and small-sample conditions， and a statistically significant superiority cannot be concluded from these results. Within this evidence boundary， the results demonstrate that explicitly introducing coverage state feedback into differential evolution search facilitates a more coordinated search direction among coverage maintenance， data backhaul and resource consumption， and can provide methodological references for continuous communication service planning of heterogeneous unmanned platforms in low-altitude economy and post-disaster emergency scenarios.