Intelligent Resources Orchestration Mechanism In Space-air-ground Integrated Networks

The space-air-ground integrated networks(SAGIN)are the important infrastructure for building up "network power" and "aerospace power",which is regarded as a complex giant system and consists of interconnected satellite networks and ground networks.Moreover,SAGIN exhibits typical characteristics,such as high dynamics,large dimensions,multiple layers and strong time-space spans.Hence,SAGIN can fully exploit the advantages of various communication network technologies to provide real-time,on-demand and reliable transmission services for various businesses.Aerospace,disaster relief,national security and remote communication services can be applied into SAGIN and enable higher demands for multi-dimensional resources,such as network bandwidth,computing and caching.The traditional network resource scheduling mechanism is difficult to meet efficient bearing demands for differentiated businesses,leading to lower resource utilization.Therefore,faced with pervasive intelligent service scenarios,it is urgent to break through key technologies such as intelligent computation offloading,highly efficient resource allocation,reliable energy harvesting and privacy protection to provide technical support for various services.Focused on highly efficient orchestration mechanism in SAGIN,this paper explores related network resources allocation and interference management,computation offloading and energy harvesting and resources scheduling and privacy protection,respectively,which further optimize system’s energy consumption and delay,network throughput,channel interference,satellite cloud servers’ profits and privacy protection overhead.The main works and contributions of this paper can be presented as follows:Firstly,focused on resource allocation and interference management problems in SAGIN,considering dynamic network environment(such as stochastic application tasks and time-varying channel gains),we establish deep edge network models and propose novel deep reinforcement learning frameworks for the sake of task scheduling,transmission power and the frequency of central processing units(CPU).Compared with traditional baseline methods,it can reduce the system’s delay and energy consumption overhead and improve the system’s task migration capabilities in the single edge network scenario.Considering the channel interference in multiple edge networks,we propose a multi-agent deep deterministic policy gradient method to allocate the optimal wireless channel and computational resources for each user.Compared with traditional baseline methods,it can severely reduce the system’s energy consumption,delay and channel interference in multiple edge networks.Secondly,focused on computation offloading and energy harvesting problems in SAGIN,we consider time-varying channel gains,massive aerial platforms coverage and battery energy storage constraints,and establish five-sphere SAGIN structure,centralized and distributed cooperative learning framework and system’s throughput model,propose Lyapunov-based stability theory multi-agent proximal policy optimization algorithms to solve task scheduling and aerial platforms selection for computation-intensive and delay-sensitive tasks,conceive a convexoptimization and linear-programming method,jointly optimize CPU cycle frequency,transmission power and battery energy.Compared with traditional baseline methods,it can notably improve the system’s throughput,reduce system’s energy consumption and guarantee battery service levels.Thirdly,focused on network resource scheduling and privacy protection problems,considering the stochastic task arrivals,time-varying channel gains and privacy protection for terrestrial users,we propose a SAGINdigital twin integrated blockchain model and Lyapunov-aided multi-agent deep federated reinforcement learning framework for optimizing CPU cycle frequency,the size of block,the number of digital twins and battery energy to achieve highly efficient resources scheduling and privacy protection.Compared with traditional baseline methods,it can prominently reduce the system’s energy consumption and privacy protection overhead.On the other hand,considering the profits of satellites cloud servers and the processed number of bits,we establish a blockchain-integrated Stackelberg game model and propose a Lyapunov-based stability theory multi-agent meta learning task scheduling mechanism to cooperatively optimize the selection of channel,task scheduling and the size of block,which further achieves the deep integration of communication,computation and block resources.Compared with traditional baseline methods,it can prominently improve the satellites cloud servers’ profits and the processed number of bits,reduce privacy protection overhead and mutual channel interference among multiple edge networks.