Introduction
This project will be developed in collaboration with BetterData, a synthetic data company whose industrial cases highlight a practical barrier to sustainable smart buildings: the data needed for better energy decisions are often too sensitive to share.
In one motivating scenario, an international construction firm collected sensor data from a Singapore office, including signals related to employee movement, location, workplace behaviour and environmental conditions. These data could support energy and workplace analytics at overseas headquarters, but direct transfer of employee-related data was restricted. Synthetic data therefore becomes a possible medium for converting local, sensitive building-use data into a shareable resource for modelling, testing and decision-making.
Buildings are central to the net-zero transition. Offices, campuses and commercial estates consume substantial energy through heating, cooling, ventilation, lighting, plug loads and local renewable integration. Many avoidable emissions arise because building operation does not reflect actual human use: spaces are conditioned when under-used, peak loads are forecast from coarse historical profiles, retrofit decisions lack realistic occupancy scenarios, and control strategies struggle to balance carbon, cost and comfort.
AI can help through occupancy prediction, demand forecasting, fault detection, adaptive HVAC control, retrofit simulation and energy-flexibility optimisation. However, these methods require granular data about people, spaces and environmental conditions. Modern smart buildings can generate traces from IoT sensors, access systems, cameras, wearables, desk sensors and building management systems. Such data are valuable for modelling the human side of energy demand, but they can also reveal sensitive behavioural patterns and are frequently constrained by employment, privacy and cross-border data-transfer rules.
The core hypothesis is that synthetic data can act as an enabling layer for building decarbonisation, not merely as a privacy-preserving substitute for real data.
Properly designed synthetic datasets could help organisations train energy-demand models, test occupancy-aware control policies, simulate hybrid-working and heatwave scenarios, benchmark building performance across sites, and collaborate with technology providers without exposing raw employee-level records. This matters because sustainable smart-building AI must learn across seasons, building types, working cultures, control systems and regulatory contexts.
The scientific challenge is that generic synthetic data quality metrics are insufficient for sustainable buildings. A synthetic dataset may match marginal distributions while failing to preserve the temporal, relational and behavioural structures that matter for carbon reduction, such as occupancy-energy coupling, weekday cycles, room-level movement, comfort-energy trade-offs and rare peak-load conditions. Conversely, a highly realistic synthetic dataset may still leak sensitive behavioural patterns. The project will therefore ask what makes synthetic smart-building data useful for decarbonisation, and when it is safe and trustworthy enough to be used.
Research Aims and Outcomes
The PhD will develop a task-oriented framework for generating and evaluating synthetic smart-building data. It will model representative variables such as time, location, occupancy, environmental conditions, energy use, HVAC state and comfort proxies. Candidate methods may include time-series generative models, relational synthetic data models, vertical federated synthetic data generation, constraint-aware generation and hybrid approaches combining building-physics priors with data-driven learning. Evaluation will focus on downstream sustainability tasks, including occupancy forecasting, energy-demand prediction, peak-load estimation, retrofit scenario analysis and cost-comfort-carbon optimisation. Privacy and governance assessment will examine re-identification, trajectory leakage, rare-pattern disclosure, provenance, watermarking and usage restrictions.
The expected outcome is a practical framework showing when synthetic data can unlock privacy-preserving collaboration for sustainable smart buildings, and when it should not be trusted. The project will contribute to data-centric AI for building decarbonisation, privacy-preserving machine learning, federated synthetic data generation and sustainable operations.