Dual time scale network slice resource allocation method based on limited memory and probability learning

Network slicing is crucial for enabling networks to meet the diverse service demands of various verticals. To address the issue of dynamic changes in slice types, a Federated Multi-Agent Reinforcement Learning (F-MALML) algorithm with dualtime scale resource allocation is proposed. At the large time scale, a finite memory learning algorithm allocates resources to each base station. At the small time scale, each base station uses F-MALML to dynamically allocate resources to users. A probabilistic learning mechanism is introduced to adjust the allocation strategy based on previous results and the current network state. Simulation results show that the proposed algorithm achieves significant improvements in slice satisfaction for newly added slices and system spectral efficiency compared to the two benchmark algorithms, while demonstrating better stability.