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

The restructuring of aging Chinese city infrastructure requires new approaches based on computational intelligence and optimization lifecycle structures. Current building renovation methods are limited by the lack of seamless linkage between real-time operation data and predictive lifecycle management, particularly regarding privacy and multi-building management in heterogeneous environments. This paper introduces a simulation framework that combines federated deep reinforcement learning and behavioural digital twins to restore Chinese buildings into smart cities. The architecture involves a three-layer system: a physical layer with IoT-enabled sensing networks in distributed building clusters, a digital twin layer with real-time BIM-to-operational model synchronization and LiDAR-enhanced geometric precision, and an intelligent layer with privacy-reflecting federated proximal policy optimization for distributed decision-making. The framework addresses the gap between fixed 3D representations and variable behavioural modelling by integrating continuous learning processes that respond to changing occupancy, energy consumption, and structural decay. Simulation studies of Chinese urban residential communities show better performance: 27.3% lower lifecycle operational costs, 34.6% improved energy efficiency with the same thermal comfort, and 39.7% better structural integrity prediction with CFRP-optimized improvements. The federated learning architecture results in 5.8% cost savings and 6.2% emission reduction, offering scalable, privacy-sensitive urban renewal decision support for China's modernization efforts.