Joint DRL and GCN-based Cloud-Edge-End collaborative cache optimization for metaverse scenarios

Abstract

The rapid emergence of the Metaverse demands deeply immersive and highly responsive virtual experiences, necessitating ultra-high transmission speeds and extremely low latency. Centralized data processing methods are facing increasing constraints in managing large-scale user data due to limitations in computing power, storage capacity, and network bandwidth. To address these challenges, this paper presents a Cloud-Edge-End transmission framework for Metaverse scenarios aimed at optimizing resource allocation, reducing latency, and enhancing rendering efficiency. We propose a distributed trajectory prediction (DTP) algorithm and develop a distributed trajectory prediction cluster system that utilizes the FastDTW algorithm to predict trajectory segments and calculate subscene popularity. Additionally, a real-time collaborative cache optimization scheme (GCNAC), based on GCN and DRL, is introduced to dynamically adjust caching strategies according to subscene popularity, thereby improving cache hit rates and reducing cache replacement frequencies. Simulations demonstrate that the GCNAC scheme markedly outperforms existing methods in target subscene cache hit rates and transmission latency across varying capacities, achieving a 43.49% reduction in edge replacement frequency. This study provides a practical solution for terminal pre-rendering of Metaverse scenes, offering both theoretical support and practical guidance for the development of scene rendering technologies and the generative Metaverse.