Digital transformation in the built environment offers new opportunities to improve building maintenance through data-driven approaches. Smart monitoring, predictive modeling, and artificial intelligence can enhance decision-making and enable proactive strategies. The preservation of historic buildings is an important scenario where preventive maintenance is essential to ensure long-term sustainability while protecting heritage values. This thesis presents a comprehensive solution for data-driven smart maintenance of historic buildings, integrating Internet of Things (IoT), cloud computing, edge computing, ontology-based data modeling, and machine learning to improve indoor climate management, energy efficiency, and conservation practices.

To enable long-term environmental monitoring, a scalable digitalization solution is developed in Paper I, integrating an IoT-based sensing system with edge and cloud computing. Field deployments confirm the long-run reliability of the system in supporting real-time and historical data analysis for maintenance decisions. Papers II and III further introduce the concept of parametric digital twins, incorporating ontology-based data models to ensure a consistent representation of building structures, systems, and environmental conditions. Case studies at the City Theatre of Norrköping and Löfstad Castle in Östergötland, Sweden, validate the effectiveness of digital twins in identifying indoor climate risks and guiding conservation strategies.

Based on the collected data, Papers IV and VI explore deep learning methods for building energy forecasting. Paper IV evaluates state-of-the-art deep learning architectures for point and probabilistic multi-horizon forecasting, showing that incorporating future exogenous factors improves prediction accuracy. It also highlights how different building operating modes impact forecasting performance. Paper VI integrates deep learning with digital twins to identify energy-saving opportunities and optimize operations.

Papers V and VII focus on predictive modeling for indoor climate management. Paper VII proposes an edge-centric approach as an alternative to cloud-centric solutions, ensuring low latency and data privacy. Paper V explores federated deep learning as a privacy-aware solution for decentralized indoor climate forecasting. A comparative study of federated learning algorithms demonstrates that federated models can achieve prediction accuracy comparable to centralized learning while preserving data privacy. These findings offer practical insights for managing heterogeneous, distributed environmental data to support sustainable building operations.

This thesis advances data-driven conservation of historic buildings by combining smart monitoring, digital twins, and artificial intelligence. The proposed methods enable preventive maintenance and pave the way for the next generation of heritage conservation strategies.