Data Liquidity Optimization for Digital Twin Empowered Internet of Things: A Federated Learning Approach

Widespread applications based on the Internet of Things (IoT) are driving the adoption of digital twins (DTs) to optimize operations and increase efficiency. However, the development of DT-empowered IoT systems has put forward a series of new requirements for the in-depth mining of comprehensive data. The first is how to ensure that the fusion results of data from a large number of connected physical and virtual entities scattered in heterogeneous networks are always better than data from the physical or virtual world alone. The second is how to effectively coordinate the interaction between physical entities and virtual entities to optimize the operation and management of IoT services. This article proposes a novel DT architecture to build various feature-rich virtual entities, and introduces a new concept of DT data liquidity to orchestrate physical-to-physical, physical-to-virtual, and virtual-to-virtual interactions of intelligent IoT functions. A federated learning method based on a two-phase optimization scheme is also developed to optimize the data liquidity of the DT system through orchestrated cloud-edge-device DT collaboration to provide high fault tolerance and reliability IoT services. Performance evaluation results on the CIRFAR10-LT benchmark and two real-world datasets show that the proposed method outperforms the state-of-the-art methods in terms of convergence, accuracy and generalization ability.