Multi-Objective Optimization of Ultra-Low Energy Consumption Buildings in Severely Cold Regions Considering Life Cycle Performance

Abstract

Net-zero energy buildings (NZEB) have received widespread attention for their excellent energy and carbon reduction potential in various countries. However, relatively little research has been conducted on the life performance of its primary form: the ultra-low energy building (ULEB). This paper proposes an optimization method combining meta-models to investigate the carbon reduction potential of ultra-low energy buildings in severely cold regions of China. The XGBoost algorithm is used to construct a meta-model of building performance, and the grid search method is used to obtain a high-precision meta-model with an R² of 0.967. Secondly, NSGA-II is used to find passive technical solutions based on the meta-model that minimize the global warming potential (GWP), global cost (GC), and operation energy consumption (OE). Finally, the variables affecting the life-cycle performance of buildings were ranked by sensitivity analysis. The results show that GWP, GC, and OE are reduced by 12.7%, 6.7%, and 7.4% compared with the original building through the optimization process proposed. Sensitivity analysis showed that for GWP, the top four sensitivities are window type (TW) > WWR of south wall (WWRS) > roof insulation thickness (IR) > WWR of north wall (WWRN). For GC, the top four sensitivities are: TW > WWRS > IR > WWR of west wall (WWRW); for OE, the top four sensitivities are: TW > IR > WWRS > WWRN. This paper’s optimization framework and research results can effectively guide the design of the ULEB in severely cold regions.

1. Introduction

Climate deterioration and energy shortage have become significant issues facing human beings. In that context, net-zero energy buildings (NZEB) have emerged as the times require, and many countries and regions have formulated relevant policies. For instance, in Europe, a net zero energy target is set from 2020 for all new buildings. To vigorously promote ultra-low and net-zero energy consumption buildings, the Ministry of Housing and Urban-Rural Development (MOHURD) of China issued the “Passive Ultra-low Energy Consumption Technical Guidelines” in 2015 [1], which led to the trend of the development of ULEBs. In 2019, the MOHURD released the “Technical standard for nearly zero energy buildings”, which laid a theoretical foundation for further standardizing and developing the ULEB and the net-zero energy consumption building [2]. Vigorously promoting NZEBs reduces the construction industry’s share of carbon emissions and excessive energy consumption. Currently, the pressure of carbon reduction in China’s construction industry is still severe. According to the “2021 China Building Energy Consumption and Carbon Emission Research Report” statistics [3], in 2019, the total energy consumption of the country’s construction in the whole process was 2.23 billion tons of standard coal. The carbon emission is 49.97 billion tons, accounting for 49.97% of national carbon emissions. In the future, energy consumption and carbon emissions will continue to increase [4]. Therefore, vigorously promoting the NZEB and the ULEB is of great and far-reaching significance for reducing carbon emissions in the building industry and achieving China’s dual carbon goals.

Currently, the research on NZEBs mainly focuses on the feasibility study of net zero energy consumption [5,6,7], energy and cost-benefit [8,9,10], and comfort exploration. The first is the feasibility study: scholars from various countries have extensively researched the local feasibility and applicability of the NZEB. To explore the feasibility of a house design suitable for net zero energy consumption, Jin et al. took a house in Shanxi Province, China, as the experimental object and first predicted the optimal configuration that can achieve net zero energy consumption through hardware and software simulation. Additionally, through energy monitoring after the construction is completed, the method’s accuracy is verified, which provides a reference for the building design of net zero energy consumption [11]. Ni et al. evaluated a new type of wood-frame house in a severely cold region of China. They explored how to achieve net-zero energy consumption through appropriate methods in the design phase. Experimental simulations demonstrated that the wood-frame building had good thermal performance and that the photovoltaic modules provided much more electricity than the energy used [12]. Pallis et al. studied the cost-effectiveness and energy savings potential of two new residential buildings (single-family and multi-family houses) in two climatic zones in Greece using various energy measures. They found that natural gas boilers, and low-temperature heat pumps, were the most cost-optimized heating systems for single-story buildings. In contrast, the opposite was true for multi-story buildings [13]. Secondly, the energy and cost of NZEBs are topics of much attention. Rezaee et al. developed a parametric model based on the city of Shiraz, Iran, where the optimization variables were divided into geometric and non-geometric and used a design-of-experiment methodology to save a lot of computational time and cost while ensuring the experimental results. The study’s final results demonstrate that the NZEB can be fully achieved in this region by improving the envelope design [14]. Abdou et al. implemented a multi-objective optimization of houses in six climate zones in Morocco, hoping to find a solution that compromises building life-cycle costs, energy consumption, and thermal comfort. The results proved that NZEBs could be achieved in all climatic zones of Morocco through an effective combination of passive design and renewable energy [15]. Li et al. proposed a risk-benefit optimization design approach to address the challenge of NZEBs not achieving the goal of zero energy consumption after use and tested the proposed approach based on the net-zero energy standard in China. The results show that the proposed method in the article can effectively contribute to the expected energy savings after using NZEBs [16]. The concern is gradually being raised about the possible thermal discomfort of the NZEB, as the NZEB often has well-insulated envelopes and excellent airtightness. While these measures prevent indoor heat loss in winter, they pose the same risk of overheating in summer. Rajat Gupta et al. applied advanced power generation measures to two residential buildings in the UK through simulations. They found that achieving net-zero energy under current conditions is not difficult. However, climate change is expected to elevate the risk of summer overheating for the NZEB by 2050 [17].

Research on the NZEB in terms of typical climate zones in China has also been gradually carried out in recent years. Feng et al. reviewed the development of the NZEB in hot and humid regions of China, surveyed 34 NZEB projects in similar climatic zones around the world, and summarized the energy efficiency measures taken by these buildings in response to the local climate. It was found that even with high energy intensity, NZEBs can achieve zero energy consumption with good use of renewable energy [18]. At the same time, the life-cycle carbon performance of NZEBs is gaining attention. Dong et al. paid attention to a NZEB located in China’s hot summer and cold winter climate zone, investigated the carbon emissions of its building life-cycle, and provided a PV system evaluation method to achieve zero carbon emissions throughout the life-cycle. Most studies are based on hot regions and regions with hot summers and cold winters in China. However, there are relatively few studies on the life performance of net-zero energy buildings in severely cold regions, especially public buildings. According to the China 2020 Building Energy Consumption Study Development Report [19], public buildings account for about 36% and 40% of overall energy consumption and carbon emissions, so enhancing the full life-cycle carbon reduction potential of public net-zero energy buildings is an innovative and advanced solution to achieve the dual carbon goal.

In China’s development of NZEBs, there is a transitional form close to zero energy buildings, ULEBs, which are the primary manifestations of NZEBs. The indoor comfort standard of the ULEB is the same as that of the NZEB, with slightly lower energy efficiency indicators than NZEBs. GB/T51350-2019 stipulates that the energy consumption standard of the NZEB is 65~75% lower than the energy standard of GB50189-2015. In comparison, the energy consumption standard of the ULEB is 50% lower than GB50189-2015 [2,20]. This ULEB is suitable for constructing ultra-low energy consumption and ultra-low-cost through passive technologies when implementing renewable energy is inconvenient and is more feasible in underdeveloped areas.

However, from the above literature review, there are relatively few studies on the performance of NZEBs in severely cold regions throughout their life cycle, and almost no studies related to the ULEB. Public buildings are much more challenging to achieve ultra-low energy due to the extreme and long-lasting winters, which are far more demanding in terms of heating than other climate zones [21]. Therefore, more design space must be explored further to improve the life-cycle performance of ULEBs. Multi-objective approaches are proven to be a means to seek high-performance building solutions [22,23,24]. So, this paper uses them as the research method to evaluate the improvement of building life-cycle performance by different combinations of passive measures, including reduction of global warming potential (GWP), global cost (GC), and operation energy consumption (OE). In order to improve the efficiency of accelerated optimization simulations, the use of meta-models as alternative models for optimization, so in this paper, XGBoost is used to construct a meta-model alternative to physical models for optimization. The details of the construction process are presented in Section 2.3. Finally, a sensitivity analysis was applied to explore the importance of the life-cycle impact of single passive measures on the building.

The originality of this research is mainly the following three points:

• The potential of XGBoost for building performance optimization design is explored by constructing a high-precision meta-model for optimization.
• A new building performance optimization strategy is proposed from the life cycle perspective to reduce the GWP, GC, and OE of the ULEB in severely cold regions.
• Sensitivity analysis is used to obtain variables with a significant impact on the target so that architects or engineers can pay more attention to them in the design phase.

2. Materials and Methods

2.1. Definition of Optimization Variables

This paper selects 16 variables that may affect building GWP, GC, and OE. It contains three construction variables and 13 geometric variables, of which exterior wall insulation type (TE), roof insulation type (TR), and window type (TW) are considered construction variables. Among them are four types of external insulation: rock wool board (RW), expanded polystyrene (EPS), extruded polystyrene (XPS), and polyurethane board (PU). According to local practices and climate adaptability, this study only considers external thermal insulation types. In terms of window considerations, this paper selects five kinds of glass, including double glass, triple glass, Low_e, and non-Low_e, as well as air interlayer and argon interlayer, all with conventional plastic-steel window frames. The material’s thermal performance is referred to as the code of “Code for thermal design of civil building” [25], which is shown in

Table 1. The insulation type and glazing type for optimization.

Categories	Names	Transmittance [W/(m2·K)]	Specific Heat Capacity [J/kg·K]
Insulation	RW EPS XPS PU	0.044 0.037 0.030 0.024	1220 1380 1380 1380
Categories	Names	Conductivity [W/(m·K)]	SHGC
Window	Double pane Double pane, low-e Double pane, low-e, argon Triple pane, low-e Triple pane, low-e, argon	2.3 1.8 1.5 1.0 0.8	0.7 0.62 0.62 0.57 0.57

.

Geometry variables include four types, window wall area ratio (WWRN\S\E\W), roof insulation thickness (IR), exterior wall insulation thickness (IN\S\E\W), and shading system (DN\S\E\W). Compared to other conventional studies that consider the thickness of the thermal insulation layer as the same thickness, this study considers the different heat transfer processes in the exterior wall of varying orientations. It distinguishes the thickness of insulation of walls in four directions to obtain more refined low-carbon design results.

At the same time, the horizontal overhang and the vertical fin are used for the north-south and east-west directions, respectively. The range of shading depth was defined to explore the impact of the external shading system on the buildings in severely cold areas.

Table 2. The distribution of variables.

No.	Categories	Symbol	Unit	Range
1	Roof insulation type	TR	/	[0–3]
2	Exterior Wall insulation type	TE	/	[0–5]
3	Window type	TW	/	[0–5]
4	Roof insulation thickness	IR	m	[0:0.33]
5	Insulation thickness of the north wall	IN	m	[0:0.33]
6	Insulation thickness of the east wall	IE	m	[0:0.33]
7	Insulation thickness of the south wall	IS	m	[0:0.33]
8	Insulation thickness of the west wall	IW	m	[0:0.33]
9	Window-to-wall radio of north wall	WWRN	/	[0:0.9]
10	Window-to-wall radio of east wall	WWRE	/	[0:0.9]
11	Window-to-wall radio of south wall	WWRS	/	[0:0.9]
12	Window-to-wall radio of west wall	WWRW	/	[0:0.9]
13	Horizontal overhang depth in north	DN	m	[0:1.5]
14	Vertical fin depth in the east	DE	m	[0:1.5]
15	Horizontal overhang depth in the south	DS	m	[0:1.5]
16	Vertical fin depth in the west	DW	m	[0:1.5]

specific the range of the variable.

2.2. Definition of Optimization Objectives

2.2.1. OE

The annual building energy consumption is the sum of cooling, heating, and lighting energy consumption. The remaining hot water and electrical equipment energy requirements are not included in the calculation range. The energy consumption simulation by Ladybug Tools can obtain these energy consumption values. The sum of the annual building energy requirements is expressed as follows:

$$OE=\frac{\left[E_H+E_c+E_l\right]}{A_{floor}}$$

(1)

where $E_H$—heating energy consumption per year, (kWh/a); $E_c$—cooling energy consumption per year (kWh/a));$E_L$—lighting energy consumption per year (kWh/a); $A_{floor}$—the total floor area (m²).

2.2.2. GC

In this paper, global cost is selected as the economic evaluation index of the life cycle of buildings. It includes the incremental cost of energy-saving technologies (excluding the cost of basic components such as building structures) and the cost of operating energy consumption and converts them to present value. This indicator is widely used in relevant research on the economic cost of the building life cycle. It can judge the economy of the building life cycle in the early stage of building design. The prices of the materials are shown in

Table 3. The prices and CEF of building materials.

Categories	Names	No.	Cost	CEF
Insulation	RW EPS XPS PU	0	48.6 USD/m3	2.4 kgCO2/kg
		1	60 USD/m3	5.7 kgCO2/kg
		2	74.3 USD/m3	20.1 kgCO2/kg
		3	150 USD/m3	5.1 kgCO2/kg
Window	Double pane Double pane, Low-E Double pane, Low-E, argon Triple pane, Low-E Triple pane, Low-E, argon	0	48.6 USD/m2	92 kgCO2/m2
		1	61.4 USD/m2	101 kgCO2/m2
		2	88.6 USD/m2	130 kgCO2/m2
		3	105.7 USD/m2	141 kgCO2/m2
		4	118.6 USD/m2	150 kgCO2/m2
Overhang	\	\	11.4 USD/m2	110 kgCO2/m2
Categories	Names	No.	Cost	CEF
Energy	Gas	\	0.064 USD/kg	2.62 kg/m3
	Electricity	\	0.069 USD/kWh	0.61 kg/kWh

.

$$C_i=C_{wall}+C_{roof}+C_{win}+C_{shade}$$

(2)

$$C_e=\left(\frac{E_H\cdot F}{{\eta }_H}\right)\cdot P_g+\left(\frac{E_c}{{\eta }_c}+E_L\right)\cdot P_e$$

(3)

$$GC=\frac{C_i+{{\displaystyle \sum }}_{i=1}^{50}\left[C_e\cdot R_d\left(i\right)\right]}{A_{floor}}$$

(4)

$$R_d\left(i\right)=\frac{1-{\left(1+R_r\right)}^{-i}}{R_r}$$

(5)

where $C_i$—the investment cost used in the passive technologies (USD); $C_{wall},C_{roof} ,C_{win},C_{shade}$—cost of roof insulation, exterior wall insulation, windows, shading (USD); $C_e$—The energy cost per year (USD); $GC$—global costs (USD/m²); $R_d$—discount rate when year is $i$; $R_r$— real interest rate; $E_H,E_c,E_L$—the energy consumption of heating, cooling, and lighting (kWh); $F$—conversion factor of heating energy to Gas consumption, 0.09 m³/kWh; ${\eta }_H,{\eta }_c—$ efficiency of heating and cooling equipment, 0.8 and 3.45, respectively; $P_g,P_e$—the price of natural gas, and electricity, specifically in Table 3.

2.2.3. GWP

This paper selects global warming potential as the environmental impact indicator widely used in building performance evaluation [26,27,28]. Its calculation formula is as Equation (6). In this study, the calculation of GWP covers two parts: the carbon emissions in the production stage of building materials and the carbon emissions in the building operation stage. Similar to the calculation method of GC, only the cost and carbon emissions of building energy-saving technologies are considered, without the carbon emissions of components such as building structures. Secondly, the carbon emission factor of electric energy refers to the “2021 China Building Energy Consumption and Carbon Emission Research Report” [3], and the carbon emission factor of the power grid in Northwest China is 0.61. In addition, the heating method of the building is independent heating by natural gas in a gas boiler, so it is necessary to know the carbon emission factor of natural gas, which is obtained from “Building carbon emission calculation standard” and” General Principles of Comprehensive Energy Consumption Calculation” [29,30]. The final calculation formula of GWP is as follows:

$$GWP={{\displaystyle \sum }}_{t=1}^T{Q}_t\times f_t+n\times \left[\frac{E_H\cdot F}{{\eta }_H}\cdot f_g+\left(\frac{E_c}{{\eta }_c}+E_L\right)\cdot f_e\right]$$

(6)

where t—the type of material; $Q_t$—quantity of the specific material; $f_t$—the CEF of a specific material (kg/material); $f_g$—the CEF of gas; $f_e$—the CEF of electricity energy.

2.3. Optimization Procedure

The framework of this paper is roughly divided into four parts: building simulation, meta-model construction, objective optimization, and sensitivity analysis, and Figure 1 shows the process of this study.

Building simulation: Firstly, we select 16 passive variables covering construction and geometric variables, sample them with the Latin hypercube method, form different combinations of passive measures, and input them into Ladybug Tools to obtain a large amount of training data for meta-model construction.

Meta-model construction: The data obtained in the first step are input into the Scikit-learn framework of Python, and XGBoost is used to construct a high-precision meta-model.

Objective optimization: To minimize GWP, GC, and OE, the NSGA-II algorithm is selected to optimize the model obtained in the second step to get the optimal combination of passive measures.

Sensitivity analysis: Finally, the data obtained in the first step are input to the Salib software for processing to get the sensitivity analysis results and rank the importance of the variables.

2.3.1. Building Simulation

The building energy model in this study was built in Grasshopper, Rhino. Ladybug Tools is selected as the building performance simulation tool in this study, applied as plugins in Grasshopper to integrate EnergyPlus, OpenStudio, and Radiance. Its accuracy has been widely verified [31,32,33]. In addition, this study uses Gpython to post-process the results to obtain the GWP, GC, and OE and record targets and variables combination by Lunchbox (a plugin in Grasshopper). All steps of building performance simulation are described in detail below.

Step 1 Parameter modeling: The thermophysical properties and carbon emission coefficients of relevant materials are written into Gpython to provide the building envelope properties. At the same time, building blocks are created in Rhino to provide the geometric properties of the building. Both are then assembled into Ladybug tools.

Step 2 Parameter selection. We sample variables using the Latin hypercube sampling method, a sampling method commonly used in machine learning model training, which avoids the problem of sample aggregation in previous sampling [34]. This paper selects 2000 samples, which are sufficient for meta-model training according to the recommendations of related literature [35].

Step 3 Simulate modeling and simulation settings. According to the functional division of the building, set the building’s occupancy rate, lighting power density, and equipment power density by the GB 55015-2021 [36].

Step 4 Result post-processing. According to the calculation rules of the objectives in 2.2, write the corresponding calculation components in Gpython and perform statistical processing on the calculation results of the three optimization objectives (GWP, GC, OE) to get the training dataset for XGBoost.

2.3.2. Meta-Model Construction

XGBoost is used as a meta-model for target optimization. XGBoost, as a combination of multiple weak evaluators with strong evaluators, was proposed in 2016 [37]. Its performance is more stable and accurate than traditional machine learning methods, so it is widely used in supervised learning problems [38,39]. To obtain the optimal model structure, the hyperparameters of the XGBoost algorithm must first be integrated search. The hyperparameters involved in this study include the maximum tree depth (Max_depth), the learning rate (Learning_rate), and the number of trees (N_estimators). The grid search method was used to optimize the above hyperparameters. In addition, 10-fold cross-validation was to improve the meta-model’s generalization ability and model accuracy and divide 80% of the data into training data sets and 20% into test data sets [40]. The evaluation indicators of the model are the root mean squared error (RMSE) and the coefficient of determination (R²). The calculation method is shown in Equations (7) and (8). Finally, the prediction accuracy of the optimal model is obtained, as shown in Figure 2, and the hyperparameters of the meta-model are specific in

Table 4. The tuned hyperparameters and evaluation metrics of XGBoost model.

N_Estimators	Max_Depth	Learning_Rate	Gamma	Subsample	R2
700	3	0.05	0.0	0.7	0.967

. After traversing the search, the R² of GWP, GC, and OE reached 0.97, 0.97, and 0.96, respectively, and the RMSE reached 0.51, 70.5, and 8.7, respectively, proving that the meta-model achieves high accuracy.

$$RMSE=\sqrt{\frac{{{\displaystyle \sum }}_{i=1}^n{\left(y_{pre,i}-y_{sim,i}\right)}^2}{n}}$$

(7)

$$R^2=\frac{{{\displaystyle \sum }}_{i=1}^n{\left(y_{pre,i}-{\overline{y}}_{sim}\right)}^2}{{{\displaystyle \sum }}_{i=1}^n{\left(y_{sim,i}-{\overline{y}}_{sim}\right)}^2}$$

(8)

where $y_{pre,i}$ is the predicted value by the XGBoost; $y_{sim,i}$ is the simulated objective by the XGBoost; ${\overline{y}}_{sim}$ is the average value of the simulated values.

2.3.3. Objective Optimization

The NSGA-II algorithm, a fast non-dominated multi-objective optimization algorithm with an elite retention strategy, is a multi-objective optimization algorithm based on the Pareto optimal solution [41]. Its superior performance has been widely used in building performance simulation optimization [42,43,44]. In this paper, the maximum generation number and the population size are set to 100 and 50. The crossover rate and mutation probability are set to 0.91 and 0.06.

The optimization direction is to minimize GWP, GC, and OE by adjusting the design parameters, expressed as: $\mathrm{Min}\left\{{\mathrm{f}}_1\left(\mathrm{x}\right)=\mathrm{GWP},{\mathrm{f}}_2\left(\mathrm{x}\right)=\mathrm{GC},{\mathrm{f}}_3\left(\mathrm{x}\right)=\mathrm{OE}\right\}$, $\mathrm{x}=\left[{\mathrm{x}}_{1,}{\mathrm{x}}_2,{\mathrm{x}}_3,\dots \dots .{\mathrm{x}}_{\mathrm{n}}\right]$.

2.3.4. Sensitivity Analysis

Sensitivity analysis can effectively help designers clarify the key points in the low-carbon design and give them more attention to achieve better life-cycle performance. This study selects the sensitivity analysis methods based on regression methods, which include SRC, PCC, SRRC, and PRCC [35]. Among them, SRC and PCC can only be used for linear models, while SRRC and PRCC can be used for nonlinear models. Due to the highly complex relationship between building performance and variables, nonlinear models are more suitable for the analysis in this paper, and SRRC can indicate the linear between variables and objectives, while PRCC can effectively exclude the influence of related factors and show the single effect of variables on each optimization objective. The absolute value of PRCC represents the degree of influence, and positive and negative signs represent the direction of impact. In the study of this paper, SRRC and PRCC were selected as sensitivity analysis methods.

3. The Base-Case Building Model

3.1. Weather Conditions

As the capital of Xinjiang Province, Urumqi is located at the northern foot of the Tianshan Mountains, the farthest city from the ocean in the world. It belongs to the temperate continental arid climate. According to the weather file (CSWD), the typical annual meteorological parameters show that the heating degree days in Urumqi are 4530.9 (°C•d) [45], and the cooling degree days are 32.6 (°C•d). Urumqi is classified as a severely cold area, with an average temperature below 0 °C for more than 180 days throughout the year. The lowest temperature in the year can reach −25°C, and the highest can get 38°C in summer. In addition, Urumqi is relatively rich in solar energy resources. The average intensity of sky radiation is 101 W/m² on the horizontal. The dry bulb temperature and radiation of 24 h per month are presented in Figure 3.

3.2. Building Modeling

The case building selected in this paper is the first ULEB in Northwest China, which was completed in 2015, and the German and Chinese governments co-financed the project. Due to its use of various advanced active and passive technologies, it has also achieved good energy-saving results in the operation stage. As a result, it has been widely recognized within the domestic construction industry and is accredited by the Passive House Institute (PHI). Therefore, the building is chosen as a typical case study to investigate its potential for life performance optimization.

The structural system is a frame shear wall structure. The design life of the building is 50 years. Among them, the underground construction area is 3163.50 m², the above-ground construction area is 4627.53 m², and the building base area is 939.59 m². The building has six floors above ground and two floors underground. The second underground floor is an underground garage, the first underground floor is a commercial building, the third floor is a dining hall, and the rest of the floors are office buildings. The exterior of the building is shown in Figure 4a.

Input the relevant parameters of the building envelope structure level into the Ladybug Tools, which are presented in

Table 5. Practices for exterior walls and roofs.

Component	Construction Practices	Conductivity $(\mathbf{W}/\left({\mathbf{m}}^2·\mathbf{K}\right)$
Roof	100 mm Reinforced concrete + 300 mm XPS board	0.1
Exterior wall	200 mm Aerated concrete blocks + 300 mm EPS board	0.15
Interior wall	200 mm Aerated concrete blocks	1.056
Exterior floor	100 mm Reinforced concrete + 300 mm EPS board	0.1
Interior wall	100 mm Reinforced concrete + 250 mm glass wool heat insulating material	0.1
Window	5 + 6A + 5+7A + 5low-E	0.8

. At the same time, the indoor thermal disturbance settings, such as personnel occupancy rate, lighting power density, equipment power density, etc., which are specified in

Table 6. Energy modelling assumptions during all simulations.

Item	Value
Location	Chongqing, China (106.44 °E, 29.49 °N)
Orientation	South
Solar distribution	Full interior and exterior (with reflections)
Occupancy density	0.125 m2/person
Lighting power density	9 W/m2
Electric equipment power density	15 W/m2
Heating setpoint	18 °C
Cooling setpoint	26 °C
Infiltration	0.0006 m3/s per m2 façade
Minimum ventilation	0.0083 m3/h· person
Natural ventilation

, refer to the standard GB 55015-2021 [36]. The air-conditioning system chooses the “Ideal Loads Air System”, which is suitable for early design in EnergyPlus. The cooling and heating setpoints are 18 °C and 26 °C, respectively, and the simulation model as shown Figure 4b.

4. Results

4.1. Analysis of Optimization Results

After 100 generations of traversal optimization, 5000 samples were obtained during optimization. 50 Pareto solutions and the distribution of three objectives were obtained, shown in Figure 5. The analysis of the optimization solution shows that the upper and lower limits of the distribution of the three objectives of GWP, GC, and OE are: 748.6 kg/m²~853.3 kg/m², 1925.5 USD/m²~1323.4 USD/m², 52.1 kWh/m²~52.8 kWh/m². The three indicators of the original scenario are 857.6 kg/m², 1388.5 USD/m², and 56.3 kWh/m². After the optimization, the optimal indicators are improved by 12.7%, 6.7%, and 7.4% compared to the original scenario.

GB/T51350-2019 stipulates that the energy consumption of ULEB should be lower than 50% of the energy consumption standard specified in GB50189-2015 [2,20]. In order to verify whether the completed optimized scheme meets the ULEB standard, the building energy models for Cases A, B, and C were developed, representing the maximum energy consumption scenario in Pareto solutions, the original building scenario, and the scenario based on GB50189-2015 (baseline solution), respectively. Figure 6 shows that Case B is slightly larger than 50% of Case C (dotted line in the figure), while Case A meets the energy consumption requirement of the ULEB. Therefore, it can be determined that all the optimized solutions meet the ULEB standard. The most significant reduction in the optimization process is in the heating energy consumption, while the cooling and lighting energy consumption decrease less.

The distribution of each variable can be seen from the parallel coordinate diagram in Figure 7. Regarding the selection of thermal insulation layer types, the IR are concentrated on XPS and PU, and the IE include EPS, XPS, and PU. The results of the optimal distribution variables show that the external wall has a broader selection range of thermal insulation layer types compared with the roof. The possible reason is that the thermal insulation area of the external wall is larger than that of the roof. Although the thermal performance of EPS is poorer than that of XPS and PU, under the condition of a large thermal insulation area, the cost advantage of EPS is evident since its low price. In the same way, the optimal solution of IR and IE not appear RW. The possible reasons for this are the price advantage of RW is not reflected under the condition that the area is not large enough. Another deeper cause is that the large heat transfer coefficient of RW makes the thermal insulation performance poor.

Regarding the thickness of the thermal insulation layer, IR is the most concentrated, IN and IS are moderately distributed, and IE and IW are widely distributed. In addition, in terms of average thickness, the IN is the largest, while the thickness of the IS is the smallest, and IE and IW are between the first two. The reason for considering the existence is that in severely cold areas, the most profound impact on indoor comfort is the low-temperature weather in winter. Thus, the solar radiation that the south-facing wall receives significantly improves the indoor thermal environment, and the excessively thick insulation layer may prevent heat entry. Walls in other directions receive far less heat from solar radiation than they dissipate, so increasing the thickness of their insulation can reduce heat loss inside the building.

Type 4 of TW is selected for all schemes, namely Triple pane, Low-E, and argon, which has the smallest heat transfer coefficient and can effectively prevent heat loss in winter. At the same time, the higher SHGC can increase the absorption rate of solar radiation in winter. Therefore, the TW that performs well is significant for the life cycle impact of commercial buildings in severely cold regions. The distribution of WWR is more concentrated, in which WWRE and WWRW both choose the minimum value. The WWRS is concentrated at the upper limit of 0.9.

In the depth of shades of four orientations, DN and DS generally occupy the lower limit of the variable value, and DS has a value of 0.2 m depth, which is because, compared to DN, DS can block a certain amount of solar radiation from entering the room in summer, to reduce cooling energy consumption. Secondly, the distribution of DE and DW is relatively more expansive due to the high latitude of Urumqi, the summertime is longer in the west, so the westward shading can effectively improve indoor comfort in summer.

4.2. Comparison between the Optimal Solutions

The three performance indicators of the Pareto front are ranked separately to obtain the design solutions with minimum GWP, GC, and OE, which are named GWP-OPTIMAL, GC-OPTIMAL, and OE-OPTIMAL, respectively, specific in

Table 7. Comparison between the optimal solutions.

Categories	UP-OPTIMAL	GWP-OPTIMAL	GC-OPTIMAL	OE-OPTIMAL
TR	PU	PU	XPS	PU
TE	EPS	EPS	EPS	PU
TW	Triple pane, Low-E, argon	Triple pane, Low-E, argon	Triple pane, Low-E, argon	Triple pane, Low-E, argon
DN	0.1	0.1	0.1	0.1
DE	0.4	0.2	0.4	0.3
DS	0.1	0.1	0.1	0.1
DW	0.4	0.3	0.4	0.4
IN	0.32	0.33	0.33	0.33
IE	0.3	0.24	0.3	0.28
IS	0.24	0.25	0.25	0.25
IW	0.27	0.27	0.27	0.27
IR	0.33	0.33	0.33	0.33
WWRN	0.27	0.25	0.34	0.18
WWRE	0.01	0.01	0.01	0.01
WWRS	0.9	0.9	0.9	0.9
WWRW	0.01	0.01	0.01	0.01
OE	52.6	52.7	52.8	52.1
GC	1298.9	1300.0	1295.4	1323.4
GWP	749.5	748.6	776.5	752.4

. In addition, the integrated optimal value is obtained as UP-OPTIMAL by the Utopian method, which is usually applied to explore the best point from the Pareto front [46].

By comparing the four optimal solutions, it is found that most of the variables take relatively consistent values, while there are differences mainly in the choice of TE and TR.

For TE, only OE-OPTIMAL chose PU, while all other optimal solutions chose EPS as the insulation type in facade. Although the high-performance TE reduces operational energy consumption, its high initial investment and carbon emission coefficient cannot effectively improve the GWP and LCC due to the too-large exterior wall area. For TR, all chose PU as insulation, except GC-OPTIMA, which chose XPS. High-performance insulation type on the roof is more likely to yield more significant benefits than the facade because of its smaller area. However, due to the high price of PU, XPS should be used as the insulation material if the priority is to reduce the LCC.

In terms of WWR, only WWRN is different. The minimum WWRN of OE-OPTIMAL is 0.18, and the maximum WWRN of GC-OPTIMAL is 0.34. Although the reduction of WWRN is beneficial to reduce the energy consumption of building heating, the energy consumption benefits obtained by reducing WWRN by GC-OPTIMAL are less evident than that of OE-OPTIMAL (because GC-OPTIMAL uses an ordinary insulation layer). Therefore, the increased lighting load caused by an excessive reduction of WWRN cannot be compensated by its reduced heating load for GC-OPTIMAL. In the scheme design process, more trade-offs should be made to WWRN according to design preferences instead of blindly reducing WWRN.

5. Discussion

5.1. Analysis the Result of Sensitivity Analysis

The individual solutions of the Pareto frontier can only provide a specific range of solution sets and cannot distinguish the importance ranking of the variables. Therefore, this study chooses a sensitivity analysis method to determine the contribution of variables to the target performance. If we understand the relative importance of the design parameters, we can enhance building performance more effectively by carefully selecting and prioritizing the important design parameters under limited conditions.

This paper selected two sensitivity analysis methods, standardized rank regression coefficient (SRRC) and partial correlation coefficient (PRCC), where positive values indicate a positive correlation between the variable and the target, and negative values indicate a positive correlation between the variable and the target a negative correlation [47]. It can be seen that the absolute value of PRCC is significantly larger than that of SRRC, indicating the influence of parameters on the nonlinearity of the objective function. The sensitivity analysis results are shown in Figure 8.

Firstly, the sensitivity of each passive measure is analyzed from the orientation. It can be seen that in WWR, insulation thickness, and shading depth, the order of the three sensitive targets is WWRS > WWRN > WWRW > WWRE; IR > IN > IS > IW > IE; DS > DN > DW > DE. It can be seen that the roof direction, south and north directions are the focus of the optimization design in life cycle performance because the building from these directions has the most significant heat gain and heat loss.

For the trend of variables, first of all, increasing the shading depth in all orientations is not conducive to improving building performance (the smaller the performance index in this study, the better). On the contrary, increasing the insulation thickness in all orientations enhances the building’s performance. In terms of WWR, only SWWR increase is more favorable for building performance improvement, while the rise in WWR in other directions is not conducive to improving building performance. Since heating energy accounts for the most significant proportion in this area, only south-facing windows gain more heat than lose heat, which helps reduce building energy consumption and costs during operation. The higher the performance of TW and TR, the more favorable the three indicators. Changing TE has contradictory effects on performance indicators. Better TE is beneficial for reducing OE, but if we want to minimize GC, we must choose poorer TE, and for GWP, the choice of TE has only little effect.

For the three objectives, the top four sensitivities for OE are TW > IR > WWRS > WWRN; the top four sensitivities for GWP are TW > WWRS > IR > WWRN. For GC, the top four sensitivities are TW > WWRS > IR > WWRW. TW, IR, and WWRS occupy the top three positions in three performance objectives. These three variables need to be prioritized in the optimization process.

5.2. The Limit and Future Research

Although this article comprehensively considers the impact of various passive variables on the GWP, GC, and OE of buildings in cold regions, passive techniques may require more refinement considerations in future research. For example, window frames have a significant impact on the thermal insulation and airtightness of a building [48]. Although the performance changes caused by changes in their thermal properties can be obtained through performance simulations, the impact on the airtightness of a building caused by them still needs to be supported by more experimental data. It should be given more attention in future research. For the ULEB or NZEBs, there is still a vast potential to improve building performance in active technology implementation. The HVAC system parameters are also important for buildings to achieve ultra-low energy targets [49,50]. It can effectively enhance building performance from the supply side compared to passive technologies that solve the problem from the demand side. This part of the parameters should be extensively covered in future research.

In addition, the passive design of the building is also affected by more personalized factors, such as public aesthetics and interest needs. For example, in the first optimization stage, the building form is used as an optimization variable for multi-objective optimization, then filtered out the architectural forms that meet the public aesthetics. Afterward, the excellent form buildings were combined with other variables for further optimization to obtain the optimal scenario of building considering the life-cycle objectives.

6. Conclusions

This paper presents a research framework for exploring the carbon reduction potential of public ultra-low energy buildings in severely cold regions. Firstly, the energy consumption model of the ULEB is established in Ladybug Tools. GWP, GC, and OE are chosen as life-cycle indicators of the building, and the impact of 16 passive measures on them is explored. A meta-model based on XGBoost was constructed, and optimal hyperparameters were obtained by the grid search method. Furthermore, the NSGA-II algorithm was used to optimize the meta-model to achieve a passive technology solution that maximizes building performance. Ultimately, a sensitivity analysis is performed to derive the importance ranking of each variable, which can help architects and engineers to focus on the consideration. Under the great goal of achieving carbon emission reduction, it is of great practical significance to evaluate the life cycle performance of the ULEB and explore the performance improvement potential of passive technologies for it. This study provides sufficient technical support for the retrofit design of the ULEB with the goal of the NZEB or even zero buildings. Additionally, it presents an effective application to assist design decisions in a seamless workflow.

The detailed findings of this paper are as follows:

(1)

The application of this framework still enhances the carbon reduction potential of ULEBs. The three optimization targets, GWP, GC, and OE, have a maximum increase of 12.7%, 6.7%, and 7.4%. Each optimal indicators generated during optimization are 748.6 kg/m², 1295.5 USD/m², and 4.2 kWh/m², respectively.

(2)

After optimization, the most suitable types of roof insulation in severely cold areas are XPS and PU, and the types of external wall insulation are EPS, XPS, and PU. Insulation thickness on roofs and north walls should be adjusted to the maximum. The window type selection of Triple pane, Low-E, and argon is most beneficial to the life-cycle targets.

(3)

This study provides the optimal solution set and the range of values of geometric variables can provide a reference for designers. Firstly, the length of shading members: DN ≤ 0.1, 0.2 ≤ DE ≤ 0.6, 0.1 ≤ DS ≤ 0.2, 0.3 ≤ DW ≤ 0.4. For insulation thickness: 0.25 ≤ IN ≤ 0.33, 0.21 ≤ IE ≤ 0.30, 0.2 ≤ IS ≤ 0.27, 0.19 ≤ IW ≤ 0.32, 0.32 ≤ IR ≤ 0.33. For window wall area ratio: 0.18 ≤ WWRN ≤ 0.4, 0 ≤ WWRE ≤ 0.02, WWRS = 0.9, 0 ≤ WWRW ≤ 0.01.

(4)

Through sensitivity analysis, it can be seen that: for GWP, the top four sensitivities are: TW > WWRS > IR > WWRN; for GC, the top four sensitivities are: TW > WWRS > IR > WWRW; for OE, the top four sensitivities are: TW > IR > WWRS > WWRN.