| The timely data transmission in emergency scenarios is very important to improve the rescue efficiency.However,the basic communication facilities in disaster areas are severely damaged.Traditional emergency communication methods,such as emergency communication vehicles can not be well applicated due to their poor mobility and low flexibility.Therefore,it is necessary to use unmanned aerial vehicles(UAVs)to assist emergency communications.However,there are two main problems need to considered.First,given the limited capabilities of UAVs,how to achieve the optimal deployment of UAVs to maximize the communication capacity is an issue that needs to be concentrated on.Second,the stability of information transmission in the data center networks(DCNs)also needs to be studied.Hence,this paper focuses on the two aspects in UAV-assisted emergency scenario,i.e.,the deployment of UAVs and the stability of DCNs.Specifically,this paper mainly includes the following three aspects:.First,in the large-scale user access emergency scenarios such as earthquakes,facing the problem of insufficient ground communication infrastructure,UAVs are used as temporary base stations to provide communication services for rescuers.Specifically,taking the differences of user distribution into account,a heuristic algorithm based on particle swarm optimization(PSO)is proposed to optimize the three-dimensional positions of UAVs,and the number of UAVs is optimized accordingly.In order to further improve the system performance and convergence speed,an initialization method based on fuzzy clustering and an inertia factor update scheme are applicated.Further,a power allocation scheme is proposed to meet the minimum service requirement of all users,considering the limited transmission power of UAV.Finally,simulation results verify the rationality and efficiency of the deployment mechanism.Then,considering the particularity of the forest scenes,there is a lack of adequacy of the traditional urban and rural channel models.Hence,the deployment of UAVs based on the forest dedicated channel model is studied.This paper proposes a deployment optimization scheme based on deep reinforcement learning(DRL)algorithm.In order to reduce the action and state space,the UAVs are divided into multiple clusters.Each cluster is regarded as an independent agent,then jointly optimizes the UAV positions in the cluster.Simulation results show that the proposed scheme achieves a better compromise between the algorithm complexity and the system performance.Finally,this paper analyzes the network stability and round-trip time(RTT)in a simplified Data Center Transmission Control Protocol(DCTCP)fluid model to better control DCN stability.Numerical results demonstrate that the network stays stable if RTT is smaller than a critical value which is the Hopf bifurcation point.This paper further analyzes the effect of other parameters to stability,while shedding a light on ways to adjust and control the stability for stability-sensitive services. |
