Route Planning For Fire Rescue Of Unmanned Aerial Vehicles Based On Reinforcement Learning

For scenes of fire rescue,post-disaster rescue is subject to more complex site and terrain restrictions.Post-disaster rescue often falls into the inability to enter the disaster-affected area and it is difficult to realize the early investigation work.The emergence of drones can effectively break this dilemma.Through reasonable path planning and obstacle avoidance training,UAVs can effectively shorten the detection time and improve the success rate of detection tasks.But for now,the premise of traditional UAV path planning is that the environment needs to be modeled first.This process is relatively complicated.In the real world,due to the complexity of the flight environment,it is difficult to directly apply the modeling method.In addition,the current path planning for multi-tasks and multi-objectives is basically offline planning,and users need to upload the planned path to the drone in advance.This means that once the surrounding environment changes dynamically,the existing path planning will become invalid.In this way,UAVs need to rely on a higher level of overall decision-making to complete tasks,and the ability of UAVs to make autonomous decisions and adapt to the environment is weakened.Based on the above problems,this paper proposes a research on UAV path planning based on reinforcement learning.It is expected that in the rescue process of small-scale disaster scenes and large-scale disaster scenes,UAVs can better complete the reconnaissance tasks and be more efficient.Quickly complete path planning.In the research of this paper,the DDPG method is mainly proposed in the post-disaster response for small-scale disasterstricken sites and relatively concentrated rescue targets.The DDPG method is a simplified Single UAV Multi Objective Path Planning(SUMOPP)method,which has a very significant effect on improving the detection efficiency of a single UAV.In view of the characteristics of large-scale disaster-stricken sites,this paper mainly introduces the multi-UAV multi-objective reconnaissance collaborative path planning(Multi UAV Multi Objective Collaborative Assignment and Path Planning,MUMOCAPP)using the multi-agent DDPG method,which can be used in a wide range of disasters,In the disaster relief scene where there are many disaster sites,multiple drones are used to coordinate reconnaissance,so as to realize the work distribution and mutual cooperation among multiple drones.This paper mainly does the following research work:For the detection problem of a single UAV facing multiple target points,multiple indicators need to be considered,such as the number of obstacles,the location and distance of sub-targets,etc.The separation structure and the dispersion of state space and operation action space allow a single UAV to choose a reasonable task completion sequence to complete the reconnaissance task while being able to dodge and overcome obstacles.In the multi-UAV multi-target reconnaissance problem,we specify a multi-intelligence collaborative decision-making model for multi-UAV cooperative target assignment and path planning problems.Integrate the multi-target assignment and path planning stages,model the competitive behavior of multi-UAVs for multi-targets,design multi-UAV cooperation incentive structures and coordinate state space,and make multi-UAV clusters control the number of targets and reconnaissance tasks.Smart selection also supports intelligent selection tasks in multiple drone reconnaissance missions to avoid collisions and flight conflicts between drone groups.This paper proposes a UAV route planning method based on structural simulation environment verification.In the disaster relief scene of a single UAV,the success rate of 1000 reconnaissance missions is 94.7%,and the reconnaissance process takes an average of 5seconds.In the multi-drone survey,the success rate is 96.8%,and the survey can be completed in an average of 15 seconds.All these confirmed the validity and reliability of the method proposed in this paper.