Adaptive Network Slicing Using Multi-Agent Deep Reinforcement Learning

Advances in network technology are expected to significantly impact next-generation communication systems. This research introduces a novel approach to network slicing that addresses workload variability, which is a key challenge in maintaining Quality of Service (QoS) for Internet of Things (IoT) communication. By utilizing a Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm, this approach effectively manages communication complexities. The proposed adaptive network slicing method optimizes deployment and integration across multi-tiered networks. Evaluation metrics reveal an effective slicing paradigm with near-real-time performance, providing a reliable and scalable communication architecture for diverse IoT applications. Additionally, the system model integrates edge and cloud computing to support seamless and efficient communication among interconnected devices. Simulations demonstrate the MADDPG framework’s superior efficiency in minimizing preference-weighted costs and achieving zero delay violations, adapting to changing conditions while maintaining resource efficiency. This approach offers a real-time, practical solution with competitive performance, optimizing resource allocation and management across various communication networks.