Approximate Q-learning-based (AQL) network slicing in mobile edge-cloud for delay-sensitive services

Mobile Edge Computing is an innovative network architecture in the 5G era that aims to provide low-latency services to end users.By utilizing network slicing technology, infrastructure providers can create separate slices for each service and offer them to service providers (SP). In order to ensure the effectiveness of the network, InP must accept slice requests from SP, thus increasing their revenue and avoiding degradation of the services when scaling up is necessary. Reinforcement learning methods have been increasingly used to address this challenge, particularly Q-learning. However, one major limitation of this method is its lack of compatibility with increasing state-action spaces. To overcome this limitation and enable control over slice acceptance in continuous environments, we propose an approximate Q-learning method that addresses the weaknesses of the Q algorithm. This method achieves fast convergence of the algorithm through modifications to crucial algorithmic functions, including defining and weighting important network features. Extensive evaluations were conducted to validate the effectiveness of our proposed approach. Metrics such as coverage, cumulative rewards, resource utilization, slice acceptance, penalty ratio, and revenue were considered. The results demonstrate significant improvements in terms of network efficiency, revenue generation, and resource allocation.