Navigating the Trade-Off Between Decarbonization and Thermal Comfort: A Simulation-Driven Optimization for Office Buildings Under Health Constraints

Abstract

Office buildings are significant contributors to energy consumption and carbon emissions due to high occupancy density and prolonged operation. To balance decarbonization with indoor environmental quality, this study proposes a simulation-driven multi-strategy optimization framework for a three-story office building in Jinan. This study integrates EnergyPlus 23.2, jEPlus+EA 2.3.2, and the NSGA-II algorithm to co-optimize building performance. We evaluate the synergistic effects of roof photovoltaic coverage ratio, night ventilation turn-on temperature difference, and HVAC control strategies on carbon emissions and thermal comfort, while ensuring that CO₂ concentrations remain within health thresholds. The results indicate that the night ventilation temperature turn-on temperature difference is the most influential parameter. It yields standardized regression coefficients (SRCs) of 0.7456 for carbon emissions and 0.5325 for thermal discomfort. The Pareto-optimal solution achieves a carbon footprint of approximately 477 tCO₂, with only 8.8% indoor discomfort hours. This framework provides a robust, practical approach for the low-carbon and healthy operation of office buildings.

1. Introduction

The construction industry is central to global carbon neutrality goals. Office buildings, with dense occupancy, high internal heat gains, and long operating hours, contribute substantially to energy use and carbon emissions [1]. With twentieth-century industrialization and population growth, office cooling demand has risen sharply, challenging grid stability and near-zero-energy targets [2]. At the same time, healthy-office standards require higher IAQ, including sufficient outdoor air and strict indoor CO₂ control to protect health and productivity [3]. However, increased ventilation can significantly raise heating and cooling energy consumption [4]. Therefore, balancing decarbonization with health-driven ventilation via efficient design and control remains a pressing optimization challenge.

The current literature predominantly tackles this strategy through two isolated pathways: active system optimization and passive architectural strategies. On the one hand, active control strategies for chilled water systems have shifted toward intelligent coordination. For instance, deep reinforcement learning frameworks can improve operational energy efficiency by approximately 15% compared to traditional variable-frequency control schemes, and coupling pump frequencies with end valves reduces hydraulic imbalance [5]. On the other hand, passive strategies, particularly night ventilation, have proven effective in decoupling cooling demand from peak grid consumption [6,7], heavily relying on the building’s thermal mass [8,9,10]. However, treating these pathways independently overlooks their dynamic synergistic potential. As global warming intensifies, the cooling potential of isolated passive strategies like night ventilation is becoming increasingly limited [11,12]. Optimizing active chilled water systems without integrating passive building pre-cooling often leads to suboptimal partial-load inefficiencies. Therefore, a critical need exists to dynamically couple passive ventilation thresholds with active HVAC control mechanisms to address extreme conditions [13,14].

Furthermore, when multi-objective optimization is applied to building performance, the vast majority of existing studies focus exclusively on the dichotomy between energy consumption and thermal comfort, such as PMV or discomfort hours [15,16]. Health metrics, particularly strict CO₂ concentration thresholds, are frequently marginalized. Relevant studies indicate that conventional ventilation standards are no longer sufficient to meet occupants’ health needs, and there is an urgent need to improve air quality by increasing the supply of fresh air [17]. Yet, strictly maintaining indoor CO₂ concentrations below 800 ppm across different climate zones can incur a massive 20% to 40% energy penalty for HVAC systems [18,19]. Most current optimization frameworks fail to incorporate IAQ as a hard boundary condition [20,21]. They either treat CO₂ as a soft penalty or optimize building envelopes and shading, without addressing the massive carbon footprint induced by health-oriented fresh air demands [22].

Building energy savings and indoor comfort often conflict. Reducing energy consumption typically lowers thermal comfort. Optimizing air-conditioning settings to balance these two factors is highly complex. Thus, traditional empirical and trial-error methods rarely yield optimal solutions [23]. To address this gap, this study proposes a co-simulation optimization framework. This framework couples active and passive building strategies. The specific contributions of this study are detailed below:

(1) Conventional energy optimizations often treat air quality as a flexible penalty rather than a primary requirement. To address this limitation, the proposed multi-objective framework integrates dynamic CO₂ limits as a strict health constraint. This method enables a rigorous evaluation of the Pareto boundary between operational carbon emissions and thermal comfort.

(2) The study reveals the dynamic thermodynamic response mechanisms of passive night ventilation. It demonstrates how optimized activation thresholds utilize building thermal mass to suppress daytime peak indoor temperatures. This approach effectively reduces active cooling loads while preventing the risks associated with subsequent overcooling.

(3) Comprehensive evaluations assess the synergistic effects of combined active and passive strategies. The analysis identifies an optimal night ventilation threshold and validates the energy-saving benefits of decoupling the variable speed pump from the chiller under extreme low-load conditions characterized by a part-load ratio.

This study utilized EnergyPlus (U.S. Department of Energy, Washington, DC, USA) as the core dynamic simulation engine, combined with jEPlus+EA (De Montfort University, Leicester, UK), to enable parametric batch simulation calculations. To achieve optimization under health constraints, this study determined the Pareto-optimal frontier for the dual objectives of decarbonization and thermal comfort, while strictly limiting CO₂ concentrations, thereby realizing health-oriented architectural design.

2. Materials and Methods

2.1. Research Framework

EnergyPlus calculates building loads using thermal balance algorithms and is unable to perform design optimization. Therefore, this study utilized jEPlus+EA to perform optimization calculations. This study proposes an optimization framework that integrates building performance simulation with multi-objective optimization algorithms. Figure 1 illustrates the technical workflow of this study. Firstly, a building model was established in OpenStudio (National Renewable Energy Laboratory, Golden, CO, USA) based on the building’s physical and geometric characteristics. This model was equipped with HVAC and renewable energy systems, with meteorological data from a typical meteorological year used as weather input. Four parameters were defined in the building model. These included roof photovoltaic coverage ratio, night ventilation turn-on temperature difference, indoor CO₂ concentration control upper limit, and chilled water variable frequency pump pressure difference set point. During the optimization process, these four parameters were configured, and the NSGA-II algorithm was employed to perform multi-objective optimization, incorporating various objective functions and constraints to obtain the optimal solution.

2.2. Building Model

This study focused on a three-story office building in Jinan, China, with a building height of 10.5 m and a floor area of 3470 m². A detailed 3D building model is depicted in Figure 2. The simulation calculations utilized hourly meteorological data for Jinan from the typical meteorological year (TMY) within the China Standard Climate Data (CSCD) data-set; this data-set is widely recognized and is embedded within the EnergyPlus software.

To ensure the representativeness of the baseline model, the thermal parameters of the building envelope and internal heat gain parameters—such as occupancy density, equipment power density, and lighting power density—are detailed in

Table 1. Design parameters for internal heat gains.

Parameters	Design Value	Unit
People load	10	m2/People
Lighting load	6.8	W/m2
Equipment load	10	W/m2
Occupant metabolic rate	66.67	W/m2
Summer clothing insulation	0.58	Clo

. The core metrics also complied with ASHRAE Standard 90.1, which is widely adopted internationally [24]. With regard to the HVAC system, cooling was provided by water-cooled chillers, with a design supply and return water temperature of 7/12 °C and an annual average coefficient of performance set at 4.5; heating was supplied via the district heating network, with fan coil units and a dedicated fresh air system used at the terminal units.

This study utilized the EnergyPlus engine to simulate dynamic annual energy consumption and environmental responses, incorporating the solar radiation, temperature, and humidity conditions shown in Figure 3. The assessment of indoor thermal comfort was conducted in accordance with the internationally recognized ASHRAE Standard 55 [25].

2.3. A Physics-Driven Co-Simulation Platform for Parametric Optimization

This study employed EnergyPlus, developed by the US Department of Energy, as its core physical calculation engine. Unlike traditional load calculation software, this engine utilizes thermal equilibrium methods and heat transfer functions. It performs a synchronous coupled analysis of the thermal dynamic response of building envelopes and the transient behavior of HVAC systems [26]. This mechanism eliminates time lag errors between load calculations and system responses. Therefore, it is particularly suitable for assessing the dynamic energy-saving potential of passive strategies such as night ventilation in buildings with high thermal inertia [27].

The simulation platform integrates a mathematical model to calculate photovoltaic power output based on irradiance and coverage area. It also accurately simulates the nonlinear efficiency degradation caused by photovoltaic backsheet temperature. To achieve precise hysteresis control for night ventilation thresholds, the model implements a customized energy management logic system. This approach offers greater flexibility than conventional thermostats and effectively manages natural cooling timing [28]. In addition, the simulation continuously monitors indoor pollutant levels in real time. It triggers demand-controlled ventilation to strictly maintain carbon dioxide concentrations below the defined health safety threshold [29].

To address optimization in high-dimensional parameter spaces, this study utilized a Java-based batch processing middleware. This tool effectively mapped the four decision variables and drove the calculation engine to execute thousands of parallel simulation tasks. The extracted simulation results were then used to construct response surfaces for carbon emissions and environmental quality, providing data support for the subsequent multi-objective optimization algorithms [30]. The flowchart is shown in Figure 4.

The reliability of the simulation outcomes relies on the physical engine and standardized baseline inputs rather than specific physical measured data. This study focused on evaluating high-dimensional parametric combinations during the strategic planning phase. This research involved thousands of virtual configurations, making it unfeasible to obtain physical measurements for all control scenarios. The core calculation engine is recognized globally and validated by standard industry test procedures for building energy simulation. Furthermore, the baseline parameters of the three-story office building in Jinan, including envelope thermal properties, internal heat gains, and occupancy schedules, complied strictly with national public building energy efficiency design standards. The multi-objective optimization framework evaluated the relative performance variations among different operational strategies. This comparative approach ensured the validity and engineering applicability of the Pareto optimal solutions, independent of a specific physical calibration process.

2.4. Formulation and Solution of Multi-Objective Optimization Problems

This study formulated the low-carbon and healthy retrofitting and operation of office buildings as a constrained, non-linear, multi-objective optimization problem. The core objective was to identify a set of Pareto-optimal solutions within the high-dimensional decision space Ω, thereby achieving the optimal balance between carbon emissions from building operations and indoor environmental quality. The complete mathematical model of this optimization problem comprised four modules: decision variables, objective functions, constraints, and solution algorithms.

The optimization framework formulated a multi-objective problem designed to simultaneously minimize annual operational carbon emissions and thermal discomfort hours. The mathematical construction set these two performance metrics as competing objectives, while strictly enforcing the indoor carbon dioxide concentration limit as a boundary constraint. This formulation ensured that the genetic algorithm evaluated fitness based on energy efficiency and thermal acceptability without violating basic respiratory health requirements.

2.4.1. Decision Variables

To achieve close coordination between active systems and passive strategies, this paper constructed a five-dimensional decision vector X, which encompassed four dimensions: renewable energy utilization, natural ventilation, hydraulic distribution control, and indoor health environment benchmarks. The expression for X is shown in Equation (1):

$$\mathrm{X} = {\left[{\mathit{x}}_1,{ \mathit{x}}_2,{ \mathit{x}}_3,{ \mathit{x}}_4,{ \mathit{x}}_5\right]}^{\mathrm{T}}$$

(1)

where the physical definitions of the variables are as follows: x₁: roof photovoltaic coverage R_pv, representing the proportion of the total net roof area occupied by the active photovoltaic array, which directly determines the energy production potential of the building itself; x₂: the temperature difference threshold ΔT_on for night ventilation, representing the minimum indoor–outdoor temperature difference required to trigger the natural cooling mechanism, which determines when the passive cooling source is activated; x₃: the night ventilation shutdown temperature difference threshold ΔT_off, which indicates the lower limit of the temperature difference at which ventilation is terminated and is used to prevent excessive cooling that could reduce comfort levels the following day; x₄: the set point for the variable-frequency chilled-water pump (ΔT_set represents the closed-loop control target for the variable-frequency driven pump; this parameter delivery determines the hydraulic distribution efficiency and energy consumption under partial load conditions); x₅: the upper limit for indoor CO₂ concentration control, where upper-limit C_CO2,lim serves as the threshold for demand-controlled ventilation strategies, reflecting the building’s response level to healthy air quality.

Table 2. Optimization operation parameters.

Categories	Parameters	Variable Types	Range Values	Step Size	Unit
Night ventilation turn-on temperature difference	P0	discrete variable	[2, 6]	7	°C
Chilled-water variable-frequency pump pressure difference set point	P1	discrete variable	[80, 150]	10	Kpa
Indoor CO2 concentration control upper limit	P2	discrete variable	[700, 1200]	100	Ppm
Roof photovoltaic coverage ratio	P3	discrete variable	[0, 80]	7	%

summarizes the key optimization operation parameters.

With regard to night ventilation strategies, this study focused on optimizing the activation temperature difference ΔT_on. This is because the activation time determined the upper limit of the potential for introducing natural cooling. To simulate the hysteresis in real control systems, the deactivation temperature difference was set as a fixed value. This avoided frequent equipment start–stops and excessive cooling in the early morning. The night ventilation strategy was automatically disabled when the indoor–outdoor temperature difference dropped below this threshold. Consequently, ΔT_on was selected as the optimization decision variable.

2.4.2. Objective Function

• Minimization of operational carbon emissions

To accurately reflect the dynamic carbon footprint of energy systems, J₁ did not simply sum energy consumption but incorporated dynamic grid carbon emission factors while deducting the instantaneous consumption contribution from roof photovoltaic systems. The model employed a Max function to address net-zero energy boundary conditions, as shown in Equation (2).

$$\mathit{min}{\mathit{J}}_1\left(\mathit{X}\right)={\sum }_{\mathit{t}=1}^{8760}\left[\mathit{max}\left(\left({\mathit{E}}_{\mathit{HVAC},\mathit{t}}\left(\mathit{X}\right)+{\mathit{E}}_{\mathit{Lig}\mathit{h}\mathit{t},\mathit{t}}+{\mathit{E}}_{\mathit{Equip},\mathit{t}}\right)-{\mathit{E}}_{\mathit{PV},\mathit{t}}\left({\mathit{x}}_1\right),0\right)\times {\mathit{\psi }}_{\mathit{grid}}\right]$$

(2)

where t denotes the simulation time step; E_HVAC,t(X) represents the total energy consumption of the chillers, pumps, and fans within the HVAC system at time t, which is a highly non-linear implicit function of vector X; E_pv,t(X) depends on the coverage factor x₁ and local irradiance for roof photovoltaic power generation; ψ_grid denotes the equivalent carbon emission factor for the local grid, expressed in tCO₂/GJ. To accurately reflect the local power generation mix, the carbon emission factor was set at 0.1584 tCO₂/GJ.

The operational carbon emissions were calculated based on the total electrical energy consumption of the HVAC system multiplied by the regional grid emission factor. To accurately reflect the local power generation mix, the carbon emission factor was determined according to the Standard for Building Carbon Emission Calculation of China.

• Minimization of dissatisfied hours of the indoor environment

To avoid a single linear weighting obscuring the physical reality, J₂ represented the cumulative discomfort hours. This metric, based on ASHRAE 55-2023, penalized periods of both thermal discomfort and deteriorated air quality. The formula for indoor environmental dissatisfaction hours is shown in Equation (3):

$$\min {\mathit{J}}_2\left(\mathrm{X}\right) = \sum \left(\mathsf{\alpha } \times {\mathrm{II}}_{\mathrm{thermal}}\left(\mathit{X},\mathit{t}\right)+\mathsf{\beta } \times {\mathrm{II}}_{\mathrm{IAQ}}\left(\mathit{X},\mathit{t}\right)\right)$$

(3)

where II(·) denotes a binary indicator function describing system state transitions, taking the value 1 when environmental parameters exceeded the set comfort range and 0 otherwise; α and β represent weighting factors balancing the priority of thermal comfort and air quality. This study assumed that both were equally important; hence, α = β = 1.

Thermal comfort status was assessed based on Fanger’s predicted mean thermal sensation index, whose indicator function was defined as shown in Equation (4):

$${\mathrm{II}}_{\mathrm{thermal}}\left(\mathit{X},\mathit{t}\right) = \{\begin{array}{c}1,\mathrm{w}\mathrm{h}\mathrm{e}\mathrm{n} \left|{\mathrm{PMV}}_{\mathrm{t}}\left(\mathrm{X}\right)\right| | 0.5\\ 0, \mathrm{o}\mathrm{t}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{w}\mathrm{i}\mathrm{s}\mathrm{e} \end{array}$$

(4)

where PMV_t(X) denotes the predicted mean thermal discomfort index at time t. According to ASHRAE Standard 55, when |PMV| ≤ 0.5, the environment was classified as a Grade I comfort zone. This value was influenced by both the cooling effect from night ventilation strategies and the active cooling capacity.

IAQ was primarily characterized by CO₂ concentration, with its indicator function defined as shown in Equation (5):

$${\mathrm{II}}_{\mathrm{IAQ}}\left(\mathit{X},\mathit{t}\right)=\{\begin{array}{c}1, \mathrm{w}\mathrm{h}\mathrm{e}\mathrm{n} {\mathrm{CO}}_{2,\mathrm{t}}\left(\mathrm{X}\right) | {\mathrm{CO}}_{2,\lim }\\ 0, \mathrm{o}\mathrm{t}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{w}\mathrm{i}\mathrm{s}\mathrm{e} \end{array}$$

(5)

where CO₂,t(X) denotes the dynamic CO₂ concentration value within the room at time t.

Metrics like degree-hours mainly capture short-term thermal extremes. In contrast, calculating the percentage of discomfort hours directly aligns with ASHRAE Standard 55. This binary approach simplifies the evaluation. It clearly verifies if a building maintains acceptable thermal comfort over a full year.

2.4.3. Constraints

(1)

Boundary constraints

To ensure the physical feasibility of the control strategy and the stability of system operation, boundary constraints needed to be satisfied. That is, decision variables needed to remain within the physical adjustment capabilities of the equipment and geometric limitations. The boundary constraint adjustment is shown in Equation (6):

$$L_{B,i}<x_i<U_{B,i} x_i\in (1, 5)$$

(6)

where L_B,i and U_B,i denote the lower and upper bounds of the ith decision variable, respectively. The choice of xi from the integer set reflected the discrete nature of equipment parameter settings such as fan speed levels.

(2)

Logical deadband constraints

To prevent the night ventilation strategies system from cycling on and off frequently near the critical temperature, which could damage the actuators, a control deadband needed to be introduced. The temperature difference threshold for activation needed to be significantly greater than that for deactivation. The logical operating deadband constraints are shown in Equation (7):

$$x_2-x_3\ge \delta$$

(7)

where x₂ represents the indoor and outdoor temperature thresholds for activating night ventilation strategies, x₃ represents the indoor and outdoor temperature difference threshold for deactivating nighttime ventilation, and δ represents the set minimum deadband width.

(3)

IQA safety hard constraints

x₅ is a variable decision variable. At any given moment, the instantaneous peak concentration of indoor CO₂ must not exceed the safety limit of 1500 ppm stipulated by health standards, thereby safeguarding basic human health. The safety hard constraint for IAQ is expressed in Equation (8).

$$C_{CO2,t}\left(X\right)\le 1500\mathrm{p}\mathrm{p}\mathrm{m}$$

(8)

Using CO₂ concentration as both a hard constraint and an objective metric creates a clear, two-level control logic: First, the hard constraint acts as a basic safety baseline. It forces the algorithm to reject any control strategy that exceeds 1500 ppm. Second, the objective function acts as an air quality optimizer. It penalizes CO₂ levels within a much stricter range to minimize poor air quality over time.

In short, the constraint guarantees absolute safety, while the objective actively drives the system toward optimal ventilation. This dual approach ensures that the algorithm cannot sacrifice basic human health to achieve lower energy consumption.

2.4.4. Optimization Algorithms and Configuration

To address the aforementioned multi-variable, non-linear, constrained optimization problem, this study proposes a fast, non-dominance sorting genetic algorithm incorporating an elite retention strategy. The Non-dominated Sorting Genetic Algorithm II (NSGA-II) algorithm was proposed in 2002 based on genetic algorithms and NSGA [31]; NSGA-II utilizes fast, non-dominance sorting; a crowding comparison operator; and an elite strategy. The detailed optimization process, from parametric modeling to Pareto optimal solution screening, is illustrated in Figure 5.

The NSGA-II algorithm was employed for multi-objective optimization to improve the adaptive fitness of the candidate population. The population continuously reproduced and evolved to generate individuals representing optimal solutions. The population size was set to 30, with parallel simulations conducted and a maximum of 100 iterations.

Based on the established building energy optimization literature, the crossover and mutation probabilities were set at 90% and 20%, respectively [31]. This specific combination maximized global search quality while maintaining computational efficiency.

Table 3. Initialization parameters of the NSGA-II algorithm.

Parameters	Value
Population size	30
Crossover	0.9
Mutation	0.2
Maximum generations	100
Total simulations	3000

summarizes the initialization parameters for NSGA-II. Specifically, a high crossover rate of 90% ensured that effective control parameters, such as optimized temperature set points, were rapidly inherited by subsequent generations, thereby accelerating algorithmic convergence. Concurrently, the moderate mutation rate of 20% introduced necessary random variations. This maintained population diversity within the highly discrete building parameter space, effectively preventing the simulation from getting trapped in local optima.

2.4.5. Optimal Solution Selection Method from the Pareto Front

While the Pareto front provided a set of non-dominated trade-off solutions, practical engineering applications necessitated the selection of a single deterministic optimal design. To systematically identify the most balanced compromise between decarbonization and thermal comfort, the method was employed.

Since the two objectives—annual carbon emissions J₁ and discomfort hours J₂—had entirely different units and numerical magnitudes, they first needed to be min–max normalized to a [0, 1] scale:

$$J_{\mathrm{i},\mathrm{k}}^{\mathrm{\prime }}={\displaystyle \frac{J_{\mathrm{i},\mathrm{k}}-J_{\mathrm{i},\mathrm{m}\mathrm{i}\mathrm{n}}}{J_{\mathrm{i},\mathrm{m}\mathrm{a}\mathrm{x}}-J_{\mathrm{i},\mathrm{m}\mathrm{i}\mathrm{n}}}}(\mathrm{i}=1,2)$$

(9)

where $J_{\mathrm{i},\mathrm{k}}^{\mathrm{\prime }}$ is the normalized value of the ith objective for the kth solution on the Pareto front. J_i,min and J_i,max are the minimum and maximum values of that objective across the entire front.

Subsequently, the theoretical method was defined at the coordinate (0, 0) in the normalized space, representing an ideal but unattainable state where both objectives were simultaneously minimized. The optimal solution was determined by calculating the Euclidean distance Dk from each Pareto solution to the Utopia point, assuming equal weighting w₁ = w₂ = 0.5 for both objectives:

$$D_k=\sqrt{w_1{\left(J_{1,k}^{\mathrm{\prime }}\right)}^2+w_2{\left(J_{2,k}^{\mathrm{\prime }}\right)}^2}$$

(10)

The solution that yields the minimum distance minD_k was selected as the final optimal design, offering the most statistically balanced performance.

However, equal weighting is not universally applicable in practical engineering. The final selection is highly sensitive to varying weight allocations, which reflect different stakeholder priorities. For instance, assigning a higher weight to carbon emissions heavily penalizes energy inefficiency, shifting the selected optimal point along the Pareto front toward the extreme energy-saving boundary. Conversely, assigning a higher weight to thermal comfort w₂ > w₁ shifts the selection toward the health priority region, which is often necessary for premium office buildings where occupant productivity outweighs energy costs. Ultimately, dynamically adjusting these weights allows decision-makers to tailor the method to specific building functions, economic budgets, and local climate policies.

3. Optimization Results Analysis

3.1. Sensitivity of Key Parameters and the Distribution of Optimal Solutions

3.1.1. Sensitivity Analysis of Decision Variables

To investigate the mechanisms by which different control strategies influence the system’s dual objectives, a global sensitivity analysis of the four variables was conducted using the standardized regression coefficient method, as shown in Figure 6. The results indicate that among the four key parameters examined, the set point for the night ventilation strategy activation temperature difference, p₀, exerted a significant influence on both building carbon emissions and the percentage of hours of discomfort. The SRC values for p₀ in the two objective functions reached 0.7456 and 0.5325, respectively, far exceeding those of the other three parameters. p₀ directly determined the duration of the time window during which natural cooling sources were utilized. Night ventilation turn-on temperature difference dominated the other operational parameters. This dominance stemmed from the building’s energy scale and geometric constraints. In Jinan, hot summers and transitional seasons create high sensible cooling loads. These loads drive the baseline energy consumption. Night ventilation utilizes the building envelope’s thermal mass. Therefore, adjusting its temperature threshold directly reduces the operating hours of the main chiller plant.

In contrast, the SRC for roof photovoltaic coverage ratio p₃ was −0.0015; its low sensitivity may be attributed to the geometric characteristics of the study subject. Meanwhile, the coefficients for p₁ and p₂ were almost zero, indicating that within the set range of variation, adjusting the pumps and CO₂ concentration had virtually no impact on the results. As pump and CO₂ concentration control constituted fine-tuning, and the HVAC system’s terminal equipment consisted of fan coil units, the energy-saving effect of the pumps was minimal, whilst the influence of pumps in radiant floor systems could be greater. Sensitivity analysis demonstrated that building decarbonization and indoor thermal comfort are governed by different combinations of parameters and that a distinct trade-off exists between them. Whilst increasing natural ventilation can reduce carbon emissions, it may also compromise thermal comfort.

In conclusion, optimizing passive structural precooling is highly effective. It improves carbon reduction and thermal comfort significantly more than adjusting active auxiliary equipment.

As can be seen from the time distribution characteristics of the PLR in Figure 7, the system operated under partial load for the majority of the time. It is worth noting that a significant frequency peak appeared at PLR = 0.2. This indicates that due to the combined effect of the night ventilation strategy and the building’s actual low load, the chiller unit was constrained by the minimum load protection mechanism for a significant portion of the time, resulting in it being passively locked at the lowest operating frequency.

The scatter plot reveals the key control trajectories of the variable flow system. When PLR > 0.2, the pump power exhibited a conventional non-linear increase as the cooling load rose. However, at the extreme low-load boundary, the data points formed a dense vertical band. This indicated a successful decoupling of the pump and chiller. Even when the chiller could not further reduce its frequency due to protective mechanisms, the variable-frequency water pump continued to down-regulate independently based solely on the actual minimal terminal cooling demand. This deep decoupling reduced the pump’s core operating power to an ultra-low range throughout the year. It effectively eliminated a common distribution redundancy issue in conventional systems, where low chiller loads often force unnecessarily high pump flows. This validates the carbon reduction benefits of the optimization strategy at the equipment operational level.

3.1.2. Parametric Characteristics of the Set of Pareto-Optimal Solutions

Having established the sensitivity of the variables, Figure 8 illustrates the parallel coordinate plots of the non-dominant solution set identified by the NSGA-II algorithm in the four-dimensional decision space and the two-dimensional objective space. Through the visual analysis of high-dimensional data, this reveals the quantitative mechanisms by which different control strategies influence building performance. The distribution of the Pareto-optimal solution set across the decision variables exhibits non-uniform clustering, reflecting the varying sensitivities of different parameters to system performance.

The optimal solutions for the upper limit of CO₂ concentration control were clearly divided into two clusters, concentrated primarily in the 800 ppm and 1000–1100 ppm ranges, with sparse data points in the intervening areas, indicating that the overall benefits of these intermediate ranges were poor; the temperature differential for night ventilation strategies was mainly distributed within the low-differential range of 2~3 °C. The data shows that when the set temperature differential was small, the corresponding percentage of hours of discomfort was around 9%. The line corresponding to the lowest carbon emissions, when traced to the far right of the graph, points to the highest level of discomfort, at approximately 11%. Conversely, a strategy that minimized the percentage of uncomfortable hours to 8% resulted in carbon emissions rising to 479 tCO₂.

3.2. Multi-Objective Pareto Fronts and Solution Selection

3.2.1. Evolution of the Pareto Front and the Boundary of Cooperative Optimization

Figure 9 compares the local single-objective solutions with the co-optimized solutions. The Pareto-optimal set yielded three representative schemes. Across these schemes, carbon emissions ranged from 476.7 to 477.5 tCO₂, and the discomfort duration ranged from 8.8% to 9.9%. Notably, the co-optimized solutions, such as the pentagonal region, strictly dominated the local solutions focused solely on comfort, such as the triangular region. By employing global cooperative optimization, the system achieved a stringent emission limit of 477.5 tCO₂ while simultaneously reducing the discomfort hours to 8.8%. This demonstrates that the proposed multi-objective framework effectively overcomes the limitations of traditional single-objective strategies and discovers a superior operational boundary.

3.2.2. Robustness Assessment of Typical Control Schemes

To evaluate system performance stability, we selected three representative operating points from the Pareto front using the multi-criteria decision-making method proposed by Zahra et al. [32]. These include an energy-priority mode, a health-priority mode, and a balanced mode determined via the LINMAP method. The balanced mode represented the Pareto solution closest to the theoretical ideal point. The balanced mode achieved an optimal compromise between energy extremes and environmental variance. Mathematically, it represented the Pareto solution with the shortest Euclidean distance to the theoretical ideal point of minimum emissions and dissatisfaction. Physically, this prevented severe operational biases and ensured that neither energy efficiency nor occupant comfort was heavily compromised under dynamic loads.

Figure 10 uses box plots to compare the statistical distribution characteristics of carbon emissions under the three strategies: energy-saving priority, balanced mode, and health priority mode. The energy-saving priority mode exhibited the lowest carbon emissions, with a median of 477.8 tCO₂, concentrated between 477.5 and 478.2 tCO₂, indicating the most robust performance in terms of low carbon emissions. In contrast, the median carbon emissions for the health priority mode were 478.1 tCO₂, with the overall distribution range expanding to 477.1~479.4 tCO₂, resulting in higher overall carbon emissions than the energy-saving mode.

3.3. Dynamic Thermodynamic Response of Core Passive Strategies

3.3.1. Dynamic Characteristics of Indoor Thermal Environment in a Typical Week

The night ventilation temperature difference determined the timing and extent of natural cooling. As shown in Figure 11, the two target parameters exhibited distinct non-linear fluctuations in response to changes in the temperature difference. When the night ventilation temperature difference was increased from 2.0 °C to 2.5 °C, the building’s carbon emissions dropped significantly to their lowest point; if the temperature difference was raised further, carbon emissions generally showed an upward trend. This phenomenon indicated that an excessively small night ventilation temperature difference was likely to result in ineffective ventilation and increased fan energy consumption, but setting the temperature difference too high severely reduced the duration of passive pre-cooling, forcing the frequent activation of mechanical cooling equipment during the day.

At the same time, fluctuations in indoor thermal comfort indicators were more pronounced. Comparing the two curves, although the system provided the optimal indoor thermal environment at 4.0 °C, this came at the cost of higher carbon emissions. At the 2.5 °C threshold, however, the system not only maximized carbon reduction benefits but also reliably kept the duration of discomfort within a low range. Therefore, setting the night ventilation strategy activation temperature difference to 2.5 °C aligned with the thermal response characteristics of this building and represented a reasonable control threshold that balanced energy consumption with indoor environmental quality.

Figure 12 explains the physical mechanism behind this optimal threshold using typical summer conditions. At 2.5 °C, natural ventilation triggers earlier and lasts longer than at 3.0 °C. This extended cycle sufficiently precooled the building envelope. Consequently, daytime peak room temperatures dropped by 0.5 to 0.8 °C compared to the 3.0 °C scenario. This passive thermal storage directly reduced the daytime peak cooling load, explaining why the 2.5 °C threshold successfully synergized thermal comfort and decarbonization.

3.3.2. Statistical Validation of Operational Performance

Figure 13 statistically validates the system’s resistance to thermal disturbances. As shown on the left, discomfort duration generally widened and shifted upward as the temperature differential increased. At 2.5 °C, the median discomfort duration hit a local minimum with the narrowest variance. This indicates that the indoor thermal environment was highly stable and least affected by meteorological fluctuations at this threshold. The right-hand graph shows that carbon emissions strictly increased with higher temperature differentials. Maintaining the threshold between 2.0 °C and 3.0 °C kept emissions below 478.5 tCO₂ under most conditions. Continuously raising this threshold weakened the passive cooling effect, increasing reliance on mechanical chillers and deteriorating the thermal environment. Statistically, 2.5 °C was confirmed as the definitive threshold for ensuring robust, low-carbon operation.

4. Discussion

This study utilized the jEPlus+EnergyPlus platform and the NSGA-II algorithm to identify optimal, multi-system control strategies for office buildings. By transitioning from isolated subsystem operations to a globally coordinated framework, this approach successfully balances indoor environmental quality with low-carbon operational targets.

A notable limitation of this study is the absence of empirical validation. Because the framework evaluated thousands of high-dimensional parametric combinations, physical measurement of all scenarios was unfeasible. However, the simulation reliability is grounded in the rigorously validated EnergyPlus engine and strict compliance with national energy standards. Most importantly, this study focused on evaluating the relative performance variations among the selected control variables, rather than predicting absolute real-world energy consumption. This comparative approach ensures the engineering validity of the Pareto-optimal solutions. Future research should calibrate this digital model using dynamic actual building loads and measured microclimate data to fully bridge the gap between simulation and operation.

Furthermore, static temperature set points for night ventilation are highly sensitive to stochastic weather conditions. Recent studies confirm that rigid controls frequently induce overcooling and thermal discomfort [33]. Therefore, future building operations must transition toward dynamic adaptive regulation, integrating model predictive control with real-time weather forecasting.

Regarding system capacity, this study optimized PV coverage without incorporating energy storage. Future research must integrate battery energy storage systems (BESSs) and economic metrics, such as life cycle costs. As demonstrated by Wan et al., applying multi-criteria decision-making methods is essential to prevent equipment oversizing and ensure a sustainable balance between technical efficiency and economic feasibility [34].

Finally, the current findings are constrained by Jinan’s climate and the specific thermal mass of the modeled three-story building. Applying these Pareto solutions to tropical climates or high-rise buildings requires rigorous recalibration. However, integrating these operational controls with physical envelope retrofits offers massive potential. As Zhou et al. demonstrated, combining active ventilated facades with structural optimization significantly mitigates building heat gain. Our future research will couple the thermal dynamics of advanced active facades with detailed HVAC parameters for higher-dimensional global optimization [35].

5. Conclusions

A significant restrictive relationship exists between the decarbonization goals during the building operational phase and the indoor thermal comfort of occupants, which corroborates the core contradiction proposed in the title of this study. Utilizing the NSGA-II algorithm and a dynamic building energy consumption simulation engine, a rigorous two-objective optimization was conducted for operational carbon emissions J1 and the proportion of hours of indoor environmental discomfort J2. The main conclusions are as follows:

(1)

Night ventilation activation temperature is the most dominant variable, yielding SRC values of 0.7456 for carbon emissions and 0.5325 for discomfort hours. Conversely, chilled water pump pressure settings show minimal sensitivity but allow the pump to operate independently at ultra-low power (0–1.3 kW) during extreme low-load conditions (PLR = 0.2).

(2)

A clear trade-off exists between operational carbon emissions and indoor thermal comfort. When the proportion of time spent in discomfort was reduced from 11% to 8%, carbon emissions rose to 479 tCO₂. Compared with a single conventional control strategy, the globally optimal equilibrium solution on the Pareto front obtained through multi-objective cooperative optimization limited carbon emissions to 477.5 tCO₂ whilst maintaining the proportion of time spent in discomfort at 8.8%.

(3)

Decision-making modes exhibit distinct statistical robustness. The energy-priority mode offers the most stable carbon emission control against disturbances. The health priority mode is highly sensitive to external fluctuations, whereas the balanced mode provides the optimal compromise between energy extremes and environmental variance.

(4)

Setting the night ventilation threshold to 2.5 °C is the optimal synergistic control point. It allows outdoor air to sufficiently pre-cool the envelope, ultimately reducing the following day’s peak indoor temperatures by 0.5–0.8 °C. This fundamentally reduces the demand for mechanical cooling.