Smarter Cities: AI Models Predict Comfort and Energy Use

Author: Denis Avetisyan

A new framework leverages artificial intelligence to rapidly assess thermal comfort and building energy performance in hot urban environments.

This review details an agentic AI system combining large language models with physics-based simulations for improved sustainable urban design in tropical climates.

Addressing the escalating challenges of urban heat islands and building energy consumption, this study introduces an ‘Agentic AI-Enabled Framework for Thermal Comfort and Building Energy Assessment in Tropical Urban Neighborhoods’ that integrates the reasoning capabilities of large language models with streamlined physics-based simulations. The framework rapidly evaluates thermal comfort metrics-such as physiological equivalent temperature [latex]PET[/latex]-and building energy demands by autonomously interpreting design tasks and activating appropriate computational models. This closed-loop system enables exploration of climate-resilient strategies like green façades, offering a pathway to reduce wall heat gain and energy use. Could this agentic approach unlock more responsive and sustainable urban design solutions for rapidly changing climates?

The Burden of Heat: Understanding Urban Thermal Stress

The Urban Heat Island (UHI) effect describes the marked temperature difference between urban and rural areas, with cities consistently experiencing warmer temperatures – often several degrees Celsius higher. This phenomenon isn’t merely a matter of discomfort; it significantly elevates energy consumption as increased demand for air conditioning strains power grids, particularly during peak hours. Simultaneously, the UHI effect poses substantial risks to public health, exacerbating heatstroke and respiratory illnesses, and disproportionately impacting vulnerable populations like the elderly and those with pre-existing conditions. Prolonged exposure to elevated temperatures can also contribute to increased air pollution, creating a feedback loop that further degrades air quality and intensifies health concerns. Consequently, understanding and mitigating the UHI effect is becoming increasingly critical for sustainable urban development and public well-being.

Conventional urban planning often reacts to the Urban Heat Island (UHI) effect after it manifests, relying on analyses of existing temperature patterns and infrastructure. This post-hoc approach limits the ability to implement preventative measures, as solutions are typically designed to mitigate existing problems rather than anticipate future thermal stress. Furthermore, many traditional models lack the predictive power necessary to accurately forecast UHI development in response to planned urban changes – such as building density or material choices. Consequently, cities find themselves continuously adapting to increasing temperatures instead of proactively shaping cooler, more sustainable environments; a shift towards predictive modeling and preemptive design strategies is crucial for effective UHI management.

Singapore’s equatorial location and high humidity amplify the Urban Heat Island effect, creating a particularly challenging thermal environment. Unlike temperate cities where cooler ambient temperatures offer some respite, Singapore consistently experiences high baseline temperatures and moisture levels, meaning even small increases due to urbanization have a disproportionate impact on perceived heat and energy consumption. This necessitates a shift beyond conventional mitigation strategies – such as increasing green spaces – towards integrated, technologically-driven solutions. Research focuses on advanced materials for building facades, optimized urban ventilation strategies leveraging prevailing monsoon winds, and the deployment of smart cooling systems that dynamically respond to microclimate variations, all aimed at maintaining thermal comfort while minimizing reliance on energy-intensive air conditioning and ensuring a sustainable, livable urban environment.

An Intelligent Framework for Thermal Analysis

The Agentic AI Framework integrates Large Language Models (LLM) with lightweight physics-based models to perform comprehensive urban thermal analysis. This combination allows for the automated processing of complex urban environments by leveraging the LLM’s capacity for understanding and interpreting design parameters, coupled with the efficiency of physics-based models in simulating thermal behavior. The framework moves beyond traditional simulation workflows by enabling a data-driven approach to urban design, facilitating detailed assessments of thermal comfort and energy performance across various scenarios and building configurations. This integrated system allows for rapid iteration and evaluation of design options, providing a more holistic understanding of urban microclimate dynamics than either LLMs or physics-based models could achieve in isolation.

The Agentic AI Framework employs Intent Analysis to streamline the urban thermal simulation process by converting high-level design objectives – such as maximizing pedestrian comfort or minimizing building energy consumption – into specific, quantifiable parameters for the simulation engine. This is achieved through natural language processing of design briefs, identifying key performance indicators, and mapping these to relevant simulation inputs like building material properties, geometry modifications, and environmental settings. By automating this parameterization step, the framework significantly reduces manual effort and potential for human error, allowing for rapid prototyping and evaluation of multiple design alternatives without requiring specialized expertise in simulation software.

Parameter Governance within the Agentic AI Framework establishes data integrity through automated validation and consistent application of design parameters across simulations. This enables prioritized evaluation of design alternatives based on pre-defined performance criteria related to both thermal comfort – measured via metrics like Predicted Mean Vote (PMV) and Predicted Percentage Dissatisfied (PPD) – and building energy assessment, utilizing key performance indicators (KPIs) such as Energy Use Intensity (EUI). The system operates as a closed-loop, iteratively refining designs based on simulation results and feedback, ensuring optimized performance across multiple objectives and facilitating sustainable urban design strategies.

Rapid Assessment Through Physics-Based Modeling

Lightweight physics-based models, specifically Computational Fluid Dynamics (CFD) and Radiative Heat Transfer Modeling, provide a computationally efficient means of predicting key microclimate variables such as temperature, humidity, and airflow velocity. These models utilize simplified governing equations and numerical methods, allowing for rapid simulations without sacrificing substantial accuracy compared to more complex, high-fidelity simulations. CFD focuses on fluid motion, enabling the prediction of air currents and pollutant dispersion, while Radiative Heat Transfer Modeling calculates heat exchange due to electromagnetic radiation. The combination of these techniques facilitates the assessment of thermal comfort, ventilation effectiveness, and the impact of building design on localized environmental conditions, offering a practical approach for iterative design optimization and performance evaluation.

Geometry analysis within rapid assessment workflows utilizes Stereolithography (STL) files as the standard input for converting three-dimensional design concepts into formats suitable for computational simulation. STL files define the surface geometry of a design using a mesh of triangles; this tessellated representation is then processed to create a solid model for airflow and thermal transfer calculations. This direct translation from design to simulation eliminates the need for manual model reconstruction, significantly reducing modeling time and potential errors. The process allows for iterative design modifications to be rapidly evaluated through simulation, streamlining the design process and facilitating performance optimization before physical prototyping.

Integration of the physics-based modeling framework with tools such as EnergyPlus-LLM and Contam facilitates detailed multizone airflow and energy performance simulations. These simulations allow for granular analysis of thermal behavior within complex geometries, leading to quantifiable improvements in energy efficiency. Specifically, testing has demonstrated measurable reductions in peak cooling power; for example, a reduction from approximately 784 W to 740 W per element was observed at solar noon under simulated conditions. This level of detail enables designers to optimize building elements for reduced energy consumption and improved thermal comfort.

Towards Cooler Cities: Impact and Optimization

The design process benefits from a novel integration of artificial intelligence, specifically leveraging the generative capabilities of GPT-4o to explore a wider range of potential solutions for mitigating urban heat. This Large Language Model doesn’t simply offer suggestions; it actively proposes innovative designs tailored to specific environmental challenges. Crucially, these AI-generated concepts aren’t accepted at face value. Each proposal undergoes rigorous evaluation through physics-based simulations, ensuring that theoretical improvements translate into tangible benefits in real-world conditions. This iterative cycle – generation by AI, validation by simulation – allows for a rapid and efficient exploration of the design space, identifying optimal strategies that balance performance, feasibility, and cost-effectiveness, ultimately paving the way for more resilient and comfortable urban environments.

Albedo modification presents a promising avenue for mitigating urban heat islands through strategic alterations to surface reflectivity. This approach focuses on implementing materials like cool roofs and cool pavements, engineered to reflect a greater proportion of incoming solar radiation back into the atmosphere, thereby reducing the amount of heat absorbed by built environments. These materials, often light-colored and possessing high solar reflectance, effectively counteract the tendency of traditional dark surfaces – such as asphalt and conventional roofing – to trap heat. Research demonstrates that widespread adoption of these techniques can lower ambient temperatures, lessen energy demands for cooling, and improve overall urban thermal comfort, offering a localized climate solution with potentially significant impacts on public health and sustainability.

The simulation framework relies on comprehensive IWEC climate data to establish realistic environmental parameters, thereby guaranteeing the accuracy and dependability of its findings. Analysis of Districts A and B revealed peak Physiological Equivalent Temperatures (PET) reaching 52.18°C and 52.35°C, respectively, highlighting areas particularly vulnerable to extreme heat. Notably, the simulations also identified a localized ‘albedo penalty’ effect – a slight increase of 1°C in PET at specific hotspots – demonstrating how certain surface materials can inadvertently contribute to heat retention despite broader cooling initiatives. This granular level of detail, derived from precise climate modeling, is crucial for targeted interventions and validating the effectiveness of proposed urban cooling strategies.

The pursuit of comprehensive urban assessment often yields baroque complexity. This work, however, prioritizes streamlined understanding. It demonstrates how agentic AI, integrating large language models with simplified physics-based simulations, can yield rapid insights into thermal comfort and building energy performance. This echoes Donald Davies’ sentiment: “Simplicity is prerequisite for reliability.” The framework avoids exhaustive modeling, instead focusing on essential relationships-a structural honesty yielding a byproduct of informed sustainable design. The reduction to core principles allows for faster iteration and broader application, acknowledging that silence – the data not collected – can be as informative as exhaustive documentation.

What Lies Ahead?

The presented framework, while demonstrating a convergence of agentic AI and established physics-based modelling, does not resolve the fundamental tension between computational expediency and representational fidelity. The ease with which lightweight models are integrated into an agentic system comes at a cost: a necessary simplification of the very phenomena under investigation. Future iterations must confront the question of acceptable error – not as a purely technical metric, but as a philosophical constraint. How much of reality can be shed before the resulting model ceases to be meaningfully connected to the world it purports to represent?

A more pressing limitation resides in the inherent opacity of large language models. The ‘reasoning’ that underpins agentic decision-making remains largely inscrutable. While performance metrics may indicate success, understanding why a particular design intervention is deemed optimal requires further investigation. The field must move beyond simply observing that an answer is produced, and towards elucidating the internal logic – or lack thereof – driving the process.

Ultimately, the true measure of this work will not be its predictive accuracy, but its capacity to provoke better questions. The complexities of urban microclimate and human thermal perception are not amenable to simple solutions. The most valuable outcome may be not a perfected model, but a framework that reveals the limits of modelling itself, and directs attention towards the irreducible ambiguities of the built environment.

Original article: https://arxiv.org/pdf/2604.21787.pdf

Contact the author: https://www.linkedin.com/in/avetisyan/

The Burden of Heat: Understanding Urban Thermal Stress

An Intelligent Framework for Thermal Analysis

Rapid Assessment Through Physics-Based Modeling

Towards Cooler Cities: Impact and Optimization

What Lies Ahead?

See also: