改进FL-APF算法的无人机群无线多跳通信网协同飞行控制方法

Collaborative flight control method for unmanned aerial vehicle fleet wireless multi hop communication network based on improved FL-APF algorithm

摘要:

为了提升无人机群无线多跳通信网协同飞行控制效果,设计改进FL-APF算法的的无人机群无线多跳通信网协同飞行控制方法。构建处理器与传感器协同结构,利用扩展卡尔曼滤波对传感器数据进行去噪处理,结合加权平均融合算法整合视觉传感器、毫米波雷达与 IMU 多源数据,生成高精度环境感知数据集。对人工势场函数进行优化,引入局部极小值规避机制与多机协同信息,结合姿态调整逻辑实现指令动态修正,构建编队误差与避障距离误差的联合控制函数。设计联邦学习分层抗丢包聚合策略,将无人机群按通信跳数划分子集群,通过分层聚合架构与多权重补偿模型,适配无线多跳通信的丢包特性,保障参数聚合有效性。通过参数分发、本地训练、分层聚合与参数反哺的流程动态优化控制参数,实现联邦学习与人工势场函数的深度融合,并通过稳定性分析确保系统在动态障碍物、通信干扰等场景下可靠运行。测试结果表明,设计方法实际飞行航迹能够达到目标飞行点位,途中的飞行航迹与总体航迹一致；避障效果良好,成功避开全部障碍物；CDMCI整体高于0.94,说明应用设计方法后,无人机群在决策过程中可以更快速地达成高度一致；DEAE整体高于0.96,说明设计方法控制下无人机群具备更强的环境感知和快速响应能力。

Abstract:

In order to improve the collaborative flight control effect of unmanned aerial vehicle swarm wireless multi hop communication network, an improved FL-APF algorithm for unmanned aerial vehicle swarm wireless multi hop communication network collaborative flight control method is designed. Build a collaborative structure between processors and sensors, use extended Kalman filtering to denoise sensor data, and integrate multi-source data from visual sensors, millimeter wave radar, and IMU using weighted average fusion algorithm to generate a high-precision environmental perception dataset. Optimize the artificial potential field function, introduce local minimum avoidance mechanism and multi machine collaborative information, combine attitude adjustment logic to achieve dynamic command correction, and construct a joint control function for formation error and obstacle avoidance distance error. Design a federated learning layered anti packet loss aggregation strategy to divide the drone swarm into sub clusters based on the number of communication hops. Through a layered aggregation architecture and a multi weight compensation model, adapt to the packet loss characteristics of wireless multi hop communication and ensure the effectiveness of parameter aggregation. By dynamically optimizing control parameters through parameter distribution, local training, hierarchical aggregation, and parameter feedback, the deep integration of federated learning and artificial potential field functions is achieved, and stability analysis is used to ensure reliable operation of the system in scenarios such as dynamic obstacles and communication interference. The test results indicate that the actual flight path of the design method can reach the target flight point, and the flight path along the way is consistent with the overall trajectory; Good obstacle avoidance effect, successfully avoiding all obstacles; The overall CDMCI is higher than 0.94, indicating that after applying design methods, the drone swarm can achieve a high degree of consensus more quickly in the decision-making process; The overall DEAE is higher than 0.96, indicating that the drone swarm under the control of the design method has stronger environmental perception and rapid response capabilities.