Distributionally Robust Optimization for Digital Twin Service Provisioning over Edge Computing

TL;DR

This paper introduces a distributionally robust optimization approach for enhancing digital twin service provisioning over edge computing, focusing on minimizing information age and improving service quality under uncertain demand conditions.

Contribution

It proposes a novel Wasserstein DRO method for joint optimization of DT model deployment and selection, addressing demand uncertainties in edge-based DTaaS provisioning.

Findings

Wasserstein DRO outperforms traditional methods in simulation tests.

The approach effectively handles demand uncertainty and improves DTaaS quality.

Simulation results show significant performance gains over existing solutions.

Abstract

Digital Twin (DT) is a transformative technology poised to revolutionize a wide range of applications. This advancement has led to the emergence of digital twin as a service (DTaaS), enabling users to interact with DT models that accurately reflect the real-time status of their physical counterparts. Quality of DTaaS primarily depends on the freshness of DT data, which can be quantified by the age of information (AoI). The reliance on remote cloud servers solely for DTaaS provisioning presents significant challenges for latency-sensitive applications with strict AoI demands. Edge computing, as a promising paradigm, is expected to enable the AoI-aware provision of real-time DTaaS for users. In this paper, we study the joint optimization of DT model deployment and DT model selection for DTaaS provisioning over edge computing, with the objective of maximizing the quality of DTaaS. To address the uncertainties of DT interactions imposed on DTaaS provisioning, we propose a novel distributionally robust optimization (DRO)-based approach, called Wasserstein DRO (WDRO), where we first reformulate the original problem to a robust optimization problem, with the objective of maximizing the quality of DTaaS under the unforeseen extreme request conditions. Then, we leverage multi-level dual transformations based on Wasserstein distance to derive a robust solution. Simulations are conducted to evaluate the performance of the proposed WDRO, and the results demonstrate its superiority over counterparts.