Federated learning and digital twins for lifecycle optimization in Urban building renewal

Area of Science:

• Computational intelligence
• Optimization lifecycle structures
• Smart city development

Background:

• Aging Chinese city infrastructure requires advanced renovation methods.
• Current methods lack real-time data integration and predictive lifecycle management, especially concerning privacy and multi-building systems.
• Heterogeneous environments pose challenges for seamless building operation data linkage.

Purpose of the Study:

• To introduce a novel simulation framework for urban renewal in Chinese cities.
• To integrate federated deep reinforcement learning and behavioral digital twins for smart city restoration.
• To address the limitations of current building renovation techniques.

Main Methods:

• A three-layer architecture: physical (IoT sensing), digital twin (BIM-operational sync, LiDAR), and intelligent (federated proximal policy optimization).
• Federated deep reinforcement learning for privacy-preserving distributed decision-making.
• Behavioral digital twins with real-time synchronization and continuous learning capabilities.

Main Results:

• 27.3% reduction in lifecycle operational costs.
• 34.6% improvement in energy efficiency while maintaining thermal comfort.
• 39.7% enhancement in structural integrity prediction accuracy with CFRP optimization.
• Federated learning achieved 5.8% cost savings and 6.2% emission reduction.

Conclusions:

• The proposed framework offers scalable, privacy-sensitive decision support for urban renewal in China.
• It effectively bridges the gap between static 3D models and dynamic behavioral modeling.
• Simulation results demonstrate significant improvements in cost, energy efficiency, and structural health prediction.