Hybrid digital twin and reinforcement framework for intelligent network slicing and dynamic resource allocation in 5G edge computing environments

Abstract

In the world of rapidly evolving communication networks, network demand and traffic are increasing, which makes efficient network slicing and resource allocation techniques a requirement. The existing models for network slicing and resource allocation lack efficient algorithms, limited generalization, and practical implementations. To overcome these issues, this research introduced a hybrid Digital twin Reinforcement Learning framework with Grey Wolf Harris Hawks Optimization (DRL_GWHHO) to efficiently allocate resources. In the initial stage, real-time data are collected and forwarded to preprocessing. The preprocessed data were then used by the hybrid Multi-Head Attention Vision transformer (MHAV) model for extracting the network features. The features that are extracted are provided as the input to the HDRL model to learn network policy and optimized network states, which are then utilized by the intelligent network slicing module built using the Graph Transformer Kolmogorov-Arnold Network (GTKAN). The optimal network slices from the model are provided for dynamic resource allocation using Temporal Fusion Transformer-Mixture-of-Experts (TFT-MoE). Finally, the network slices and resources are optimized and forwarded to the Explainable AI (XAI) module for ensuring interpretability. The evaluation of the experiment has shown an accuracy of 98.6%, bandwidth utilization of 97.3%, and network reliability of 98.2%. The proposed model showed higher accuracy, packet delivery, resource reliability, and bandwidth utilization with low latency, making it suitable for real-world applications.