Research On Collaborative Offloading Algorithm Of Computing Tasks For Satellite Internet

Satellite Internet is a multi-level three-dimensional network composed of ground network,air network and space satellite network,which provides seamless internet access services for global equipment and supports key military and commercial applications such as battlefield operations,disaster relief,maritime transportation,etc.Satellite Internet ground equipment resources are limited,and it is difficult to deal with computing intensive tasks.It is urgent to design an effective computing task offloading algorithm to solve this problem.The existing research on the offloading of computing tasks in Satellite Internet mainly focuses on reducing the offloading delay,energy consumption and the total cost of tasks in LEO satellite networks,ignoring the impact of the change of coverage area on the task offloading performance when LEO satellites move at high speed.On the one hand,when a single-layer multi mobile LEO satellite moves,the change of the coverage area makes the remaining coverage time of the equipment providing services uncertain,and the change of the dynamic network environment makes it difficult to predict the task transmission time.The existing task offloading algorithm is difficult to be applied to high-speed mobile satellite scenarios.On the other hand,due to the different remaining coverage time of multi-layer mobile LEO satellites with different orbits and the dynamic changes of the communication environment,the task offloading decision space is complex,and the existing offloading algorithm is difficult to make the optimal offloading decision.Aiming at the problem that the uncertainty of the remaining coverage time of single layer multi mobile satellite constellation and the difficulty in estimating the task transmission time lead to the inapplicability of the existing task offloading algorithm,a task offloading algorithm based on multi-agent Soft Actor Critic(SAC)is proposed.Firstly,under the change of satellite remaining coverage time,a collaborative computing architecture of ground equipment,UAV,satellite and cloud server is designed,and the collaborative computing rules with remaining coverage time constraints are constructed under this architecture.Secondly,considering the dynamic information such as the computing resources required by the ground equipment task,the dynamic distance between the ground equipment and UAV,the bandwidth and interference between the ground equipment,constructing the computational cost model of the task offloading algorithm.Finally,under the constraints of single-layer multi mobile satellite remaining coverage time,it analyzed the offloading problem of minimizing the computational cost of all tasks is solved by minimizing the computational cost of each equipment task,and the problem is modeled as a Markov decision process(MDP).In this thesis,simulation experiments are carried out on the task offloading process under the condition of the remaining coverage time of single layer multi satellites and the number of ground equipment.The experimental results show that the multi-agent SAC based offloading algorithm can reduce the computational cost of all task offloading under the condition that singlelayer multi mobile satellites provide different remaining coverage time constraints for each ground device,and is more suitable for ground equipment task offloading in Mobile Satellite Internet scenarios than other algorithmsSecondly,considering the computing resources required by each ground equipment task,the dynamic distance between ground equipment and UAV,the bandwidth,interference and other dynamic information between ground equipment,a computing cost model of task offloading algorithm is constructed.Finally,under the constraint of the remaining coverage time of single layer multiple mobile satellites,the offloading problem of minimizing the computing cost of all tasks is solved by minimizing the computing cost of each device task,and the problem is modeled as a Markov Decision Process(MDP).In this thesis,the task offloading process of single layer multi satellite with different remaining coverage time and number of ground equipment is simulated.The experimental results show that under the condition that single layer multiple mobile satellites provide different remaining coverage time for each ground device,compared with the single agent DDPG algorithm,the offloading algorithm based on multi-agent SAC reduces the task offloading calculation cost by an average of 24.7%,which is more suitable for the task offloading of ground devices in the mobile satellite Internet scene.Aiming at the problem that the decision space of task offloading is complex in the scene of multiple layers,different orbits and multiple mobile satellite constellations,which makes it difficult to solve the optimal decision,a task offloading algorithm based on Twin Delayed Deep Deterministic Policy Gradient(TD3)of multi-agent is proposed.Firstly,the dynamic network environment such as the distance between each ground equipment and multiple UAVs,different orbits and different satellites,the remaining coverage time of satellites,and the problem of minimizing the task computing cost of each ground equipment are analyzed,and the computing cost model of the task offloading algorithm is constructed.In order to solve the dimensional disaster problem caused by the complexity of the decision-making space of each ground equipment,each ground equipment is regarded as an agent to observe the local environmental information,and the multi-agent task offloading process modeling is realized through the distributed part of the observable MDP.Finally,after each agent perceives the dynamic complex environment in real time,it learns to improve the offloading decision by interacting with the environment to obtain reward values.In this thesis,the task offloading process of multiple mobile satellites in different orbits with different remaining coverage time and number of ground equipment is simulated.The experimental results show that when multiple mobile satellites in multiple layers and different orbits provide different remaining coverage time for each device,compared with the single agent DDPG algorithm,this algorithm significantly reduces the decision solution space and the average computing cost of 45.7%in the task offloading decision process,and can make the optimal offloading decision for ground equipment tasks.The collaborative offloading algorithm of computing tasks for Satellite Internet proposed in this thesis can offload computing for resource constrained device tasks,significantly reduce the cost of task offloading computing,and achieve effective task offloading computing under the condition that single layer or multi-layer multiple mobile satellites provide different remaining coverage time for each ground equipment.