Federated Learning assisted Energy-Efficient Digital Twin Edge Networks

Digital Twin (DT) is an emerging technology, which can reshape both the digital space and the physical space. However, the construction of DT models involves continuous data synchronization and incurs high computational overhead, leading to significant energy consumption. To improve the energy efficiency of DT modeling, we propose a two-stage Federated Learning (FL) assisted framework in Digital Twin Edge Networks (DITEN). In the first stage, we select a subset of DTs with strong modeling potential and reliable data using a collaborative and data-aware strategy. In the second stage, we formulate a joint optimization problem of edge association and bandwidth allocation to maximize energy efficiency during training. A greedy tree-expansion algorithm and a Deep Q-Network (DQN) based scheduling method are developed to solve the two-stage problem. The simulation results show that the proposed approach significantly improves energy efficiency while maintaining high model accuracy.