Path Planning And Computation Offloading In Unmanned Aerial Vehicle Assisted Mobile Edge Networks

Unmanned aerial vehicle(UAV)is an autonomous aircraft equipped with multi-sensors and controlled by ground station and onboard computer.Because of its airborne storage,communication and computing capabilities,it has become a potential choice for the new generation of mobile edge computing(MEC)node.In order to push edge computing services closer to users through UAV,this thesis studies from the following four aspects of UAV path-planning-based obstacle avoidance and auxiliary computation offloading in view of the dynamic airspace environment and network topology.First,airborne Li DAR-assisted UAV obstacle recognition under dynamic trajectory and attitude is studied.The existing works did not comprehensively consider the datastore space and UAV attitude.The obstacle recognition schemes based on computer vision had a high occupancy rate of airborne storage resources,while the Li DAR-based recognition schemes did not consider the dynamic change of UAV trajectory and attitude.Therefore,in this thesis,the contour time-domain accumulation of multi-obstacles is completed by coordinate transformation of 2D Li DAR point cloud data and mapping image noise preprocessing.Based on above operation,this thesis puts forward an optimal time-domain cumulant graph clustering and motion state detection algorithm based on dynamic trajectory and attitude.Experimental results show that,compared with other clustering algorithms,this algorithm can significantly improve the detection accuracy of the number of obstacles,and can locate and identify the movement state and speed of each object.Next,UAV obstacle avoidance based on joint optimization of global and local path planning is studied.Previous works did not comprehensively consider the global nature of path planning and the dynamic nature of uncertain multi-obstacles.Existing obstacle avoidance schemes by using single path planning algorithm are difficult to take into account the real-time performance of global map updating and the local dynamics of air proximity obstacle avoidance.Thus,this thesis considers the optimality of global path planning solution,and proposes a global path planning algorithm based on improved particle swarm optimization through adaptive inertia weight optimization,particle swarm selective mutation and optimal solution particle mutation.Meanwhile,considering the timeliness of local path planning solution,a local path planning algorithm based on improved artificial potential field is proposed by optimizing repulsion function and setting virtual target points.Moreover,this thesis proposes a joint optimization algorithm of global and local path planning for UAV obstacle avoidance through planning mode switching and path smoothing.The experimental results show that,compared with single path planning algorithm,the proposed obstacle avoidance algorithm can reduce the path length by 10.94% and its fitness value is only 17.1% of the original algorithm.Then,multi-level computation offloading problem of UAV-assisted MEC network is studied.The existing works rarely considered UAV as a relay node to provide multi-layer computation offloading service.However,the independent-node scheme is not conducive to the joint optimization of communication,computing and energy resources of heterogeneous nodes in MEC network.Therefore,in this thesis,the optimal hovering position of UAV with the objective of minimizing communication distance is solved through the joint optimization of UAV flight altitude and antenna half power beam width angle.Besides,combining with UAV position,a three-layer computation offloading strategy among Io T devices,UAV and MEC server is proposed by solving the problem of minimizing energy consumption under the constraint of task demand delay.Experimental results show that,compared with the random offloading strategy with different efficiency,the proposed strategy can dynamically plan the offloading scheme under different task demand delay,data volume and UAV hovering height,and reduce the average UAV energy consumption by at least 13.85%.Finally,ground-air controlled joint path planning for multi-UAV-assisted MEC offloading is studied.In the current numerous literatures,the ground-station-controlled cluster scheduling and impact of user mobility on airborne-controlled path planning and computation offloading are not considered comprehensively.Thus,this thesis establishes a multiUAVs,multi-users and multi-areas energy efficiency optimization model with a total time delay constraint of computing tasks.Considering the impact of flight trajectory on the quality of offloading services,the above model is decomposed into two subproblems of energyefficiency-coupled cluster path planning and computation offloading.The optimal cluster scheduling strategy and the shortest global path are obtained by considering the priority of service area,the residual energy of UAV and the flight distance between them.The optimal waypoint location and the shortest local path are determined based on the communication coverage limited by user mobility and task delay.Moreover,by maximizing the access number and minimizing the task completion delay,the energy efficiency of the proposed strategy is obtained.Experimental results show that,compared with other path-planning-based offloading algorithms,the proposed strategy can provide more computational offloading services,obtain shorter flight path length,reduce the total task delay and achieve greater energy efficiency.To sum up,the researches on path planning and computation offloading in UAVassisted mobile edge network can comprehensively consider the uncertainty of airspace environment,the mobility of heterogeneous edge computing nodes and the dynamics of network topology.While ensuring the safety of UAV flight,this thesis can make full use of the surplus airborne resources to provide users with closer computation offloading service and improve the quality of user experience.