Device Digital Twin Construction Based on Random Matrix Theory and Deep Auto-Encoder Networks

Modeling user behavior and evaluating demand response potential at the device level remain critical challenges in the construction of digital twins for power systems. These challenges stem from the high-dimensional nature of user data and the dynamic interactions between users and environmental factors. To address these issues, this paper proposes a framework that integrates Random Matrix Theory (RMT) and Deep Auto-Encoder (DAE) networks. RMT is employed to uncover complex coupling relationships in high-dimensional datasets, while DAE networks perform dimensionality reduction and provide interpretable visualizations of user demand response potential. Case studies demonstrate the effectiveness of the framework in identifying user behavior patterns, evaluating demand response metrics such as peak shaving and valley filling, and generating 3D visualizations to support decision-making. The proposed approach enhances the adaptability and precision of device-level digital twins, offering practical solutions for optimizing power system operations.