| With the rapid development of communication technologies,the variety and number of services for wireless communication have increased dramatically,leading to an increasingly complex communication network structure.These services have different requirements in terms of latency,reliability,and mobility,and therefore,wireless network resource management is facing unprecedented challenges.A prerequisite for on-demand deployment of resources is the ability to sense service requirements in advance,so accurate resource prediction is critical to the deployment of functions and resource allocation for the entire network.In addition,there is a demand for multiindustry and multi-service shared network resources in the 6G air-space-space-sea scenario,and how to ensure on-demand provision of differentiated service resources to meet the ultimate service demand,secure isolation between services,and reduce network operation and maintenance costs and energy consumption to achieve a high level of network autonomy is an urgent problem to be solved.To address the above issues,this thesis investigates wireless network resource prediction and management based on deep learning algorithms to achieve reasonable and efficient utilization of network resources and meet the diverse needs of different terminal services.The main research contents of this thesis are as follows:(1)Aiming at the limitation of the traffic resource point prediction model in wireless network environment,by introducing error factor and interval prediction method on the basis of LSTM algorithm,a wireless access network resource interval prediction algorithm combining error prediction and LSTM model is proposed.Firstly,the model parameters are optimized through a large number of training processes.Then,the trained model is used to predict the wireless network traffic and traffic error respectively.Then,from the Angle of error,the capacity demand interval of wireless network is constructed by using the prediction results of both.Finally,the prediction interval of the proposed algorithm is evaluated by simulation.Simulation results show that the proposed algorithm is superior to the traditional interval prediction method in terms of prediction accuracy and coverage rate.(2)Aiming at the problems that the performance requirements of various services are difficult to be guaranteed due to the diversified service requirements of terminals,the indiscriminate sharing of network resources,and the interrelation of services,a wireless network resource management algorithm based on joint network slice selection and slice resource allocation is proposed.Firstly,an optimization problem was established,and sliced communication,computing and storage resources were allocated according to the characteristics of service delay and energy consumption.Then QLearning method was used to solve the problem,and the optimal sliced service access strategy was obtained.Then,based on the optimal slice access strategy,MA-DDPG method was used to obtain the optimal slice resource allocation strategy.Finally,the proposed algorithm is evaluated by simulation.Simulation results show that the proposed method is superior to the traditional deep reinforcement learning method in terms of service access success rate and optimal decision time.(3)In order to realize the function of terminal services accessing slicing base station on demand,an experimental system based on open source software Free5 GC and UERANSIM was built to realize the function of network slicing arrangement and precise access slicing of services.Firstly,the network environment including a core network and two base stations is simulated,two kinds of slices of SMF and UPF functions of the core network are created,and related configurations of the core network,base station and terminal are completed.Then,the communication process is analyzed from two aspects of base station and terminal service.The experimental results show that the user plane data bearing no services can be connected to the corresponding network slicing resources according to the slicing orchestration results. |
