UAV Path Planning Based On Deep Learning In The Internet Of Things

Recently,the method of UAV-assisted data collection in the Internet of Things(Io T)has been widely used.Compared with traditional static data collection methods,the method of UAV-assisted data collection not only overcomes the limitations of ground transportation,but also reduces the energy consumption of sensor nodes and prolongs the service life of devices in Io T.However,UAV-assisted data collection still faces some key issues that need to be solved urgently.Path planning is an important research topic in the field of UAV.For different types of execution tasks,different path planning problems are established to improve the performance of UAVs in terms of service energy efficiency and time delay.UAV path planning can be divided into global path planning and local path planning.In this paper,a global optimal path is firstly planned through global path planning,and then a single target node is navigated and searched through local path planning.This subject of this paper is divided into two parts.The first part mainly models and solves the problem of UAV’s global path planning.The second part solves the singletarget node navigation of UAV’s local path planning.The specific research content is as follows:(1)Design of the global path planning of UAV based on pointer networksConsidering the energy constraints of the UAV,the revenue of the sensor nodes,etc.,in this paper,we establish the mode of global path planning for UAV data collection based on the orienteering problem.The orienteering problem is a combination problem of vertex selection and determining the shortest Hamiltonian path between selected vertices.The optimization goals are maximizing the total reward of collecting data during the mission period and minimizing the flying distance of UAV.This paper uses the strategy of pointer network and random search to solve the established orientation problem model,and obtains the set of UAV service nodes and order.We also compare the planning effects of different optimization objective functions,and designs the strategy of greedy reward to compare the planning performance with the strategy of pointer network and random search.(2)Design of UAV navigation learning mechanism based on Deep Q networksTaking the cost issue into account,it may not be feasible to configure the navigation of the UAV,and the navigation accuracy is affected by the terrain and wireless environment.In this paper,according to the received broadcast reference signal strength from the sensor node,the UAV performs navigation search on the single target nodes.This paper considers the channel model of free space loss and line-of-sight loss,and uses the deep Q network to learn the action selection strategy.By continuously evaluating and updating network parameters,the local path is planned,and the target node is approached and served.Then,the learning performance of deep Q learning under different channel models is verified.