基于改进深度Q网络的无人机巡视三维路径规划方法研究

Research on UAV Patrol 3D Path Planning Method Based on Improved Deep Q-Network

针对传统无人机路径规划算法存在的算法维度高、建模困难、效率低等问题,研究了一种基于改进深度Q网络的无人机三维路径规划算法。在该算法中,基于卷积神经网络构建了深度Q网络;为提高网络对关键地形信息的提取,设计了注意力增强模型;为实现综合优化飞行路程与能耗,设计了奖励函数。针对传统深度强化算法存在的网络收敛困难等问题,设计了组合探索策略。将该算法与A*算法进行对比,从定性和定量角度验证了该算法可以实现权衡路程与能耗的无人机路径规划,并显著提高规划效率。

To address the problems of traditional UAV path planning algorithms,such as high algorithm dimensionality,difficult modelling and low efficiency,this paper proposes a UAV 3D path planning algorithm based on an improved deep Q-network(DQN).In this algorithm,DQN is constructed based on convolutional neural networks,an attention enhancement model is designed to improve the extraction of key terrain information by the network,and a reward function is designed to achieve comprehensive optimisation of flight distance and energy consumption.To address the problems of traditional deep reinforcement algorithms such as network convergence difficulties,a combined exploration strategy is designed in this paper.This paper compares the algorithm with the A*algorithm,and verifies that the algorithm can achieve UAV path planning with a trade-off between distance and energy consumption from both qualitative and quantitative perspectives,and significantly improves the planning efficiency.