⚡ electrical engineering

Multi-objective Optimization of Energy Performance, Thermal Comfort, and Retrofit Cost for Underground Commercial Spaces Using a Surrogate-Based Approach

This study employs a surrogate-based multi-objective optimization framework integrating DesignBuilder, EnergyPlus, BPNN, and NSGA-II to identify optimal retrofit strategies for underground commercial spaces that balance energy use intensity, thermal comfort, and investment costs while revealing key trade-offs and dominant influencing factors.

Original authors: Zhongcheng Duan, Zhou Hao, Shibo He, Gang Yao

Published 2026-09-28

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Zhongcheng Duan, Zhou Hao, Shibo He, Gang Yao

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Deep beneath the surface of university campuses, a quiet revolution in how we think about buildings is taking place. These are not the sun-drenched offices or residential towers we are used to, but underground commercial spaces: bustling food courts, retail shops, and gathering spots buried below the earth. Because they lack windows and natural light, these environments rely entirely on machines to keep the air moving, the lights on, and the temperature comfortable. This heavy dependence creates a difficult balancing act. If you turn down the machines to save electricity, the space can become stuffy or too cold. If you crank them up to ensure comfort, the energy bill skyrockets. For decades, engineers have tried to fix these problems by looking at one factor at a time, but a new study suggests that the solution lies in understanding how these factors fight and work together. The goal is no longer just to save energy, but to find a sweet spot where comfort, cost, and efficiency can coexist without one destroying the others.

Researchers at the China University of Mining and Technology turned their attention to a specific underground cafeteria on their Xuzhou campus, a space filled with food stalls, a small supermarket, and dining areas. They wanted to see if they could redesign how this space operates to be better for the planet, the wallet, and the people inside. To do this, they built a detailed digital twin of the building, a virtual model that could run thousands of simulations to see what would happen if they changed everything from the thickness of the walls to the temperature settings on the air conditioners. They did not just guess; they fed the model real data about how many people visited, what equipment was used, and how the building actually behaved. Then, they used a powerful computer algorithm to test millions of different combinations of changes, looking for the best possible outcomes.

The study focused on three main goals that often pull in opposite directions. The first was energy use, measured by how much electricity the building consumed per square meter over a year. The second was thermal comfort, which they tracked by counting the hours when the temperature was too hot or too cold for people to feel at ease. The third was the cost of the changes themselves, calculated as the annualized investment required to make the upgrades. The researchers found that the original, unmodified building used about 248 kilowatt-hours per square meter annually and left people uncomfortable for nearly 2,813 hours a year. When they let the computer search for the perfect solution, it revealed a stark reality: you cannot have everything. The best possible scenario for saving energy resulted in a building that was significantly less comfortable, while the scenario that made people most comfortable required a massive jump in energy use.

Instead of picking a winner, the researchers mapped out a landscape of compromises, known as a Pareto frontier. This map showed that for every step toward lower energy bills, there was a corresponding step toward more hours of discomfort, and vice versa. They identified that the most powerful levers for change were not the walls or the roof, but the way the building was managed. The temperature settings on the heating and cooling systems, the amount of fresh air brought in from outside, and the number of people inside were the dominant factors. Surprisingly, simply adding more insulation to the underground walls had a much smaller effect than expected, because the earth surrounding the building already provides a stable temperature. The study showed that the biggest gains came from adjusting the internal operations: lowering the power of the lights and equipment, tuning the air conditioning to be slightly less strict, and carefully managing how much fresh air was pumped in.

From this complex web of options, the researchers selected a single "overall optimal" solution that offered the best balance. This recommended plan would increase the energy use slightly to about 285 kilowatt-hours per square meter, but it would reduce the time people felt uncomfortable by nearly nine percent, bringing the total down to roughly 2,563 hours. The cost to implement these changes would be modest, amounting to about 22 yuan per square meter per year. This solution did not achieve the absolute lowest energy use or the absolute highest comfort, but it avoided the extreme costs and poor performance of the other options. It proved that by making small, coordinated adjustments to how the building runs, rather than overhauling its physical structure, managers can create a space that is efficient, comfortable, and affordable.

The findings challenge the common assumption that fixing an underground building requires expensive, heavy construction. The study suggests that the most effective path forward is to prioritize how the space is operated. By fine-tuning the temperature settings, reducing the power of lighting and equipment, and managing the flow of air, operators can achieve a significant improvement without breaking the bank. This approach offers a practical blueprint for universities and other institutions that rely on these underground spaces. It shows that the future of green building in these unique environments is not about building thicker walls, but about thinking more carefully about the invisible systems that keep the space alive. The research provides a clear, data-driven guide for making these difficult decisions, ensuring that the underground spaces of tomorrow are not just energy-efficient, but also places where people genuinely want to be.

Technical Summary: Multi-objective Optimization of Energy Performance, Thermal Comfort, and Retrofit Cost for Underground Commercial Spaces Using a Surrogate-Based Approach

Problem Statement
Underground commercial spaces, particularly within university campuses, present a unique set of challenges for energy retrofitting. Unlike above-ground buildings, these spaces suffer from limited daylight, restricted natural ventilation, and a heavy dependence on mechanical HVAC systems. They are also characterized by high internal heat gains from concentrated occupant activities and equipment, leading to complex interactions between energy use, thermal comfort, and economic feasibility. Existing retrofit strategies often focus on single objectives (e.g., energy reduction) or are derived from above-ground building contexts, failing to address the specific trade-offs inherent in underground environments. There is a lack of systematic frameworks that simultaneously optimize energy use intensity (EUI), thermal comfort (measured by indoor discomfort hours, IDH), and annualized retrofit investment cost (AIC) for these specific building typologies.

Methodology
The study employs a surrogate-based multi-objective optimization framework applied to the underground commercial space of the Taoyuan Cafeteria at the China University of Mining and Technology (approx. 2,745 m²). The methodology proceeds through six sequential stages:

Building Performance Simulation: A calibrated 3D model was developed using DesignBuilder and EnergyPlus, incorporating field data on geometry, envelope properties, internal loads, and operational schedules. The model was validated against field measurements using Normalized Mean Bias Error (NMBE) and Coefficient of Variation of Root Mean Square Error (CV(RMSE)).
Parametric Sampling: Using jEPlus, 1,600 parametric samples were generated via Latin Hypercube Sampling (LHS) from a high-dimensional design space (originally containing over 10 billion scenarios). Input variables included envelope insulation (material and thickness), lighting and equipment power densities, HVAC setpoints, outdoor air supply rates, air infiltration, and occupancy density.
Sensitivity Analysis: Standardized Rank Regression Coefficient (SRRC) analysis was conducted on the simulation database to identify dominant influencing factors on EUI, IDH, and AIC, filtering out variables with low sensitivity.
Surrogate Modeling: A Back-Propagation Neural Network (BPNN) was trained to predict EUI and IDH based on the input variables. The model utilized 11 input neurons and 2 output neurons, achieving high predictive accuracy ( $R^2 > 0.99$ for EUI and IDH) while avoiding the computational cost of repeated EnergyPlus simulations. Annualized Retrofit Investment Cost (AIC) was calculated deterministically using a cost function rather than being predicted by the neural network.
Multi-Objective Optimization: The trained BPNN surrogate model was coupled with the Non-dominated Sorting Genetic Algorithm II (NSGA-II) to generate a Pareto-optimal solution set. This approach explored the trade-offs between minimizing EUI, IDH, and AIC simultaneously.
Decision Making: The Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) was applied to the Pareto front to select a single "overall optimal" solution that balanced the three conflicting objectives.

Key Results

Baseline Performance: The baseline scenario exhibited an EUI of 248.17 kWh/(m²·a) and an IDH of 2,813.00 hours.
Sensitivity Findings: SRRC analysis identified occupancy density, heating setpoints, and outdoor air supply rates as the dominant factors influencing EUI. For IDH, occupancy density, heating setpoints, and lighting power density were most influential. AIC was primarily driven by physical retrofit measures such as insulation thickness, material type, and lighting system upgrades, showing low sensitivity to operational setpoint adjustments alone.
Pareto Trade-offs: The optimization revealed clear conflicts between objectives. Reducing EUI generally increased IDH (thermal discomfort), while improving thermal comfort (reducing IDH) often required higher energy consumption or increased investment costs. The Pareto front showed EUI ranging from 158.86 to 402.89 kWh/(m²·a), IDH from 2,487.81 to 2,822.63 hours, and AIC from 21.45 to 35.52 CNY/(m²·a).
Optimal Solutions:
- Energy-Prioritized: Achieved a 34.50% reduction in EUI but only a 0.90% reduction in IDH.
- Comfort-Prioritized: Reduced IDH by 11.30% but increased EUI by 60.60%.
- Cost-Prioritized: Achieved the lowest AIC (21.73 CNY/(m²·a)) with moderate improvements in energy and comfort.
- Overall Optimal (TOPSIS): Selected as the recommended strategy, this solution achieved an EUI of 285.21 kWh/(m²·a) (a 14.93% increase over baseline, reflecting the cost of comfort), an IDH of 2,563.10 hours (an 8.88% reduction), and an AIC of 21.94 CNY/(m²·a).

Significance and Claims
The paper claims to provide a methodological basis for the green retrofit of university underground commercial spaces by integrating building performance simulation, surrogate modeling, and evolutionary algorithms. Its primary contributions include:

Framework Integration: Establishing a workflow that combines DesignBuilder/EnergyPlus, jEPlus, BPNN, SRRC, and NSGA-II to efficiently handle the computational complexity of multi-objective optimization in complex underground environments.
Trade-off Clarification: Quantifying the specific trade-offs between energy savings, thermal comfort, and economic costs in underground commercial spaces, demonstrating that single-objective optimization is insufficient for these contexts.
Strategic Insight: Identifying that for underground spaces, operational control (HVAC setpoints), internal load management (lighting/equipment), and ventilation strategies are more critical for performance improvement than envelope insulation alone. The study suggests a sequential retrofit pathway prioritizing operational parameters and internal loads before moderate envelope improvements.
Decision Support: Offering a quantitative, balanced retrofit strategy (via the TOPSIS-selected solution) that avoids the extremes of single-objective optimization, providing a practical reference for stakeholders aiming to balance low-carbon goals with economic feasibility and occupant comfort.

The authors acknowledge limitations, noting that the study simplified certain dynamic occupant behaviors and did not include measured thermal sensation feedback or indoor air quality metrics in the optimization. They suggest future work could integrate stochastic occupant models and broader environmental indicators.

More like this