Research On Intelligent Configuration Mechanism In Radio Access Network For Complex Environment

In recent years,with the widespread commercialization of 5G mobile communication,the initiation of 6G technology research and development,and the emergence of large-scale artificial general intelligence models,the advancement of the information and communication industry and artificial intelligence technology has accelerated significantly.The emergence of diversified service types and various intelligent applications has led to extreme communication performance demands that exceed the capacity of existing mobile communication networks.In order to meet the growing communication requirements and support diversified services,the future wireless access network architecture will exhibit characteristics of flexibility,intelligence,density,and customization.At the same time,due to the coexistence of multiple wireless communication technologies,multiple base stations,and various application scenarios,the future wireless access network will become more distributed,heterogeneous,and dynamically complex.This will result in model-driven network planning and optimization,as well as resource coordination and scheduling methods,being insufficient to effectively cope with such a volatile environment.To support highly dynamic and elastic networks and service requirements,network configuration mechanisms such as deployment,collaboration,optimization,and self-healing have become an important research topic in the field of wireless communication.Currently,advancements in artificial intelligence theory,breakthroughs in computing power,and support from big data technology have facilitated the shift of wireless network configuration tasks from relying on expert experience to a data-driven approach.Machine learning algorithms,a major component of artificial intelligence,have been applied in various aspects of radio access networks due to their exceptional data feature extraction and efficient inference capabilities.However,when applying machine learning algorithms designed for computer domain tasks(such as natural language processing,computer vision,speech recognition,and prediction and recommendation)to wireless communication domain configuration tasks,issues may arise,including lack of robustness and adaptability,high complexity,state-action space explosion,and absence of convergence guarantees.Thus,designing intelligent self-configuration mechanisms for radio access networks in complex environments has become a significant and challenging issue in future wireless communication technology.This dissertation employs Transformer,reinforcement learning,and federated learning machine learning algorithms along with classical optimization algorithms to investigate adaptive configuration mechanisms for future radio access networks.The main research issues include:(1)intelligent on-demand deployment of moving base stations in radio access networks：(2)efficient collaboration mechanism based on federated learning in radio access networks：(3)distributed intelligent interference coordination mechanism in radio access networks：(4)autonomous self-healing mechanisms for radio access networks based on Pareto optimization.The contributions and innovations of this dissertation can be summarized as follows:First,this dissertation investigates the on-demand deployment problem in radio access networks,using traffic prediction-based pre-deployment and resource allocation to offload traffic in hotspot areas.To address the on-demand deployment problem,this dissertation designs an autonomous learning framework and solves it through the following three sequential stages.In the demand prediction stage,a dual Transformer network is designed to learn from historical traffic demand to periodically predict traffic distribution and identify hotspot areas.In the proactive deployment stage,an improved Multi-cut General Benders Decomposition(DMGBD)algorithm based on semi-supervised learning is proposed to address the on-demand deployment of mobile base stations and the pre-allocation of wireless resources.Moreover,this thesis theoretically proves that the proposed DMGBD can converge to a Γ-optimal solution.Finally,in the resource allocation fine-tuning stage,real-time resource adjustments enhance the transmission rate for offloaded users,thereby mitigating the reduction in resource utilization caused by traffic prediction deviations.Experimental results demonstrate that the proposed deployment mechanism can significantly increase the system’s total throughput and effectively prevent severe degradation of users’ service of quality in hotspot areas.Then,this dissertation investigates the efficient collaboration mechanism in radio access networks based on federated learning,aiming to enhance the learning efficiency by designing adaptive collaboration strategies.However,due to the limitation of wireless resources,there is a complex relationship of cooperation and competition among multiple heterogeneous collaborating nodes.Therefore,the problem can be modeled as a Markov game process.To address this problem,a multi-agent reinforcement learning based collaborative strategy is designed for the fully distributed collaborative learning framework.Moreover,this dissertation theoretically proves that the strategy can converge to the Nash equilibrium.Furthermore,to deal with the high-dynamic heterogeneous wireless environment,an mean field representation method and a dual neural network are designed in the multi-agent reinforcement learning algorithm to address the issues of action space explosion and overfitting.Experimental results show that the proposed algorithm significantly outperforms other four traditional algorithms in terms of accuracy,training time,learning efficiency,and communication cost on open-source datasets.Next,this dissertation investigates the distributed interference management problem in ultra dense radio access networks,aiming to minimize inter-cell interference while ensuring user transmission rate by minimizing transmit power.The interference coordination problem is modeled as a decentralized partially observable Markov decision process,and a fully distributed interference management mechanism is proposed based on a model-free multi-agent reinforcement learning framework.This mechanism does not require extra signaling interaction between base stations,allowing telecom operators to configure the network in a plug-and-play manners.To reduce the dimensionality of joint actions,this dissertation uses mean field theory to approximate the action-value function,effectively avoiding complex information exchanges between agents.Simulation results show that the proposed distributed intelligent interference coordination mechanism can effectively mitigate interference,reduce power consumption,and improve network performance while ensuring user service quality compared to other traditional algorithms.Finally,this dissertation investigates the self-healing problem in sliced radio access networks,with the goal of seeking the Pareto solution for the utilities of multiple slices through wireless resource reconfiguration,enabling multiple wireless access network slices to maximize their utilities while ensuring their service requirements.To quickly reconstruct broken slices in radio access networks,this dissertation proposes a generative adversarial network based Pareto optimization algorithm and theoretically proves that the algorithm can converge to Pareto solution with a probability of 1.Simulation results demonstrate that the proposed GPO algorithm effectively reduces the inverse generation distance of the optimal Pareto solution set and improves the utility and isolation level of radio access network slices.In summary,this dissertation systematically investigates artificial intelligence-enabled radio access network self-configuration mechanisms in complex heterogeneous environments,addressing the challenges faced in radio access network configuration.The research achieves efficient autonomous network configuration functions,providing useful theoretical foundations and technical approaches for the development of a new generation of intrinsically intelligent mobile communication networks.