People
Thermal protection and human variability
We model how hot or cold people feel and how likely each sensation is, accounting for human heat generation, physiology, and context.
Research and selected work
People, systems, climate.
Research on how people, building systems, and climate interact to shape thermal protection and energy demand. We ask how building decisions can account for the likelihood and consequences of unacceptable discomfort across diverse occupants and uncertain climate futures.
Three interacting areas
Research programme
Nine highlighted papers. Select a contribution to read its citation below.
Systems
Integration and energy uncertainty
We connect human-response models to building simulation and control, using predicted thermal sensations to assess discomfort risk, energy demand, and operational uncertainty.
Climate
Future weather and resilience
Weather generation, changing exposure, and climate stress-testing reveal how uncertain futures affect building performance and the people buildings serve.
People / Selected work
How can building decisions account for differences in human heat generation and the likelihood of discomfort?
Human heat defaults and contextual factors shape how we model thermal response. Predicting the probability of each sensation makes uncertainty in how people feel visible to design and control.
Human heat defaults. Re-examines the 120 W/person heat-generation assumption using demographic-aware metabolic rates and their implications for energy demand and thermal comfort.
Correcting the 120-Watt Assumption: Demographic-Aware Metabolic Rates for Energy Savings and Thermal Comfort Equity in Buildings.
Hongshan Guo, Ruiji Sun, Youmin Xu
Energy and Buildings · 2025
Contextual drivers. Examines sensitivity to thermal comfort drivers and shows how missing-data choices change their apparent importance in learned models.
Toward Smarter HVAC Control: Machine Learning Reveals Hidden Drivers in Thermal Comfort Databases.
Hongshan Guo, Ilaria Pigliautile, Yu Chang, Qingyao Qiao, Yichun Li
Energy and AI · 2026
Thermal sensation probabilities. Estimates the probability of each response on the seven-point cold-to-hot scale, making uncertainty in thermal sensation available to design and control decisions.
From Seven Points to Probabilities: Ordinal Learning for Risk-Aware Thermal Comfort Prediction.
Hongshan Guo, Dorit Aviv
Building and Environment · 2026
Systems / Selected work
How can building control weigh the likelihood of discomfort alongside energy demand and peak loads?
Co-simulation connects models of human response to building operation. Model benchmarking and probability-informed supervision then examine the energy and thermal trade-offs that emerge when those models inform system decisions.
Co-simulation. Links data-driven thermal sensation models to building energy control, creating a bridge from human-response prediction to testing how systems operate.
A Co-Simulation Methodology for Integrating Data-Driven Thermal Sensation Models with Building Energy Control.
Hongshan Guo, Kanxuan He, Youmin Xu, Yue Lei
Energy and Buildings · 2026
Model benchmarking. Benchmarks machine-learning and deep-learning sensation models against predicted mean vote across five climates, comparing energy, comfort, and peak-demand trade-offs.
Lower Peaks, Similar Growth: Isolating the Value and Limits of AI-Driven Thermal Sensation Modeling for Building–Grid Resilience Under Future Climates.
Hongshan Guo, Kanxuan He
Engineering Applications of Artificial Intelligence · 2026 · Accepted 15 September 2026.
HVAC supervision using sensation probabilities. Uses predicted sensation probabilities to inform HVAC supervision and examine delivered-energy and operative-temperature trade-offs across a weather stress grid.
Delivered-Energy and Operative-Temperature Trade-offs for Learned-Probability-Informed HVAC Supervision in a Weather Stress Grid.
Hongshan Guo
Energy and Buildings · 2026 · Accepted 27 September 2026.
Climate / Selected work
How should buildings be tested when future weather and the people they serve are changing?
The paired sAMY studies examine how climate-adapted weather is generated and what determines its fidelity. The cooling-and-ageing study connects future weather with differences in the thermal outcomes people experience.
sAMY: weather generation. Generates scenario-conditioned weather from multi-decadal observations for building performance analysis under climate uncertainty.
Scenario-Conditioned Actual Meteorological Years (sAMY): A Stochastic Weather Generator Using Multi-Decadal Observations.
Hongshan Guo, Kanxuan He
Energy and Buildings · 2026
sAMY: input fidelity. Examines how input quality and statistical complexity affect climate-adapted weather fidelity through a decomposition of degree-day errors.
Input Quality, Not Statistical Complexity, Determines Climate-Adapted Weather File Fidelity: A Causal Decomposition of Degree-Day Errors.
Hongshan Guo, Kanxuan He
Energy · 2026
Cooling and ageing. Examines how future weather and population ageing produce different residential thermal outcomes even when cooling setpoints are the same.
Equal Cooling Setpoints, Unequal Thermal Outcomes: Residential Cooling Service Under Future Weather and Ageing.
Hongshan Guo
Energy · 2026 · In press
Methods and applications
Thermal comfort, human heat, and MRT
- Ordinal and probabilistic thermal comfort prediction from seven-point sensation votes
- Physics-informed neural networks with physiological constraints
- Mean radiant temperature sensitivity analysis for HVAC control
- Demographic-aware metabolic-rate modeling and ASHRAE standards revision
- Fan-assisted cooling, gender differences, and cross-cultural comfort evidence
Climate forcing, weather futures, and uncertainty
- Climate-adapted weather file evaluation using causal decomposition of degree-day errors
- Scenario-conditioned and stochastic weather generation from long observational records
- ERA5 / CMIP6 forcing chains for climate stress-testing
- Monte Carlo propagation and uncertainty benchmarking across future scenarios
- Building performance evaluation under climate-stressed futures through 2100
Control, co-simulation, and interpretable learning
- EnergyPlus-based co-simulation for occupant-aware building control
- Risk-aware control framing from probabilistic comfort outputs
- Interpretable ML, feature attribution, and sensitivity tracing
- Indoor temperature forecasting and competition-tested predictive workflows
- Digital twins that connect learned models to operational testing
Ventilation, public institutions, and design-facing AI
- Wet-market ventilation retrofits and IEQ-energy trade-off analysis
- Field measurement in mixed-mode and public-building contexts
- Design pedagogy systems such as Socratic Oracle and AI Design Coach
- Evidence-based generative design workflows for architecture education
- Research legible to design studios, standards work, and public-sector collaboration
Explore further.
Browse publications since joining HKU, or use the full CV for the complete career record.