GenAI-Driven Digital Twin Models for Real-Time Simulation of Edge Retail Infrastructure

Abstract

The proliferation of edge computing in retail introduces significant complexity in managing distributed, heterogeneous infrastructure susceptible to dynamic demand fluctuations and disruptions. Traditional simulation and static digital twins (DTs) lack the adaptability and predictive fidelity required for proactive management. This research presents a novel framework integrating Generative Artificial Intelligence (GenAI) with Digital Twins to create dynamic, real-time simulation models of store-level edge retail infrastructure. We detail the architecture, leveraging Neural Operators, Temporal Graph Neural Networks (T-GNNs), and conditional diffusion models to synthesize high-fidelity system states, predict resource utilization under stochastic conditions, and generate plausible incident scenarios for resilience testing. Implemented using optimized model inference within Apache Flink streaming pipelines and validated against emulated edge environments, our GenAI-DT demonstrates a 32.7% improvement in forecasting accuracy (MAPE) and enables sub-second anomaly impact assessment, significantly enhancing proactive capacity planning and operational resilience. Key challenges around computational overhead and explainability are discussed, alongside future research directions in federated learning and meta-learning for self-evolution.