基于强化学习的无人机全自主电力巡检

针对无人机在电力巡检中的全自主性进行研究,提出全自主电力巡检系统,该系统由无人机智能体、充电桩和待巡检目标构成。借助无线充电技术和强化学习决策算法使无人机拥有全自主执行任务的能力,并在设计的仿真环境中进行了验证,训练后的无人机智能体可以自主路径规划进行电力巡检和自主决策到附近的无线充电桩充电,无需人工介入可完成所有巡检任务。由于传统的近端策略优化算法在本系统中奖励低的问题,因此提出一种基于动作掩码的近端策略优化算法来训练无人机智能体,仿真实验结果表明,动作掩码机制使奖励提高了39%,能耗降低了15.34%。

Autonomous Power Inspection of UAV Based on Reinforcement Learning

Based on the research on the full autonomy of UAV in power line inspection , an autonomous power inspection system is proposed , which is composed of UAV agent , charging pile and targets to be inspected.With the aid of wireless charging technology and reinforcement learning algorithm , UAV has achieved the independent ability to perform a task , which has been verified in the design of simulation environment.Trained UAV agent can dopath planning for electric power inspection and make decision to nearby wireless charging pile autonomously , and all the inspection tasks can be done without human intervention.Due to low reward of traditional proximal strategy optimization ( PPO ) algorithm in this system , therefore , a kind of PPO algorithm based on action mask ( PPOAM ) is proposed to train UAV agent.The simulation results show that the action mask mechanism can increase the reward by 39% and reduce the energy consumption by 15.34%.