Reinforcement Learning-based Dynamic Admission Control with Resource Recycling for 5G Core Network Slicing

5G core network slicing is an important part of 5G end - to-end slicing, which can provide customized services by tailoring network functions for diverse application scenarios. In order to implement core network slicing efficiently and make full use of network resources, slice admission control that selectively accepts or rejects slice establishment requests is crucial. However, existing related works mainly focus on optimizing the revenue of mobile operators, and lack consideration of dynamic resource scheduling and recycling under limited resources. To this end, we propose a dynamic slice admission control mechanism with a warm-up resource recycling method, which uses the adaptability of reinforcement learning to improve the slice admission rate and ensure the efficient utilization of resources. Also, considering the differentiated demands of typical application scenarios, a slice request dataset construction rule is designed and a dataset is established to evaluate the effectiveness of the proposed mechanism. Experimental results demonstrate the superiority of the proposed core network slice admission control mechanism in guaranteeing high slice admission rate under various simulation settings.