Analysis of the Energy Optimization Method of Apartment Buildings by Using Generative Design in Terms of Integrated Design

Abstract

Improving building energy efficiency has become essential for reducing global greenhouse gas emissions. (1) Background: We aim to strengthen early-stage collaboration among stakeholders based on integrated design principles, rather than relying solely on individual designers’ subjective decisions. The goal is to propose an objective method for optimizing apartment building layouts. (2) Methods: Accordingly, key design elements for energy optimization were identified, and corresponding energy usage prediction data were collected to build a database. Generative Design (GD) techniques were applied to generate and evaluate alternative layout configurations. (3) Results: The conventional apartment block layout process, which heavily depends on the expertise and intuition of experienced designers, was automated using Revit-Dynamo. An energy optimization method from the integrated design perspective was subsequently proposed. (4) Conclusions: GD enabled the identification of comprehensively optimized layout alternatives. We demonstrate the applicability of Revit-Dynamo-GBS in apartment complex design from an integrated design perspective and suggest improvements to existing certification systems and procedures in light of domestic policy considerations.

1. Introduction

1.1. Background and Objectives

The building sector accounts for ~30–40% of global energy consumption and is responsible for over 30% of greenhouse gas emissions [1,2], making it a critical domain for achieving a low-carbon society. Consequently, improving energy efficiency in buildings emerged as an urgent priority. While building energy performance is largely determined during the early design stages, traditional architectural design processes often defer energy performance evaluations until after decisions on form and function have already been made. This delay frequently results in a significant energy performance gap between the predicted and actual energy usage, often by a factor of two to five [3]. One of the primary causes of this discrepancy is the lack of integration between architectural design and energy modeling during the early stages, leading to inefficient collaboration. Hashempour et al. [3] analyzed two collaborative architectural projects aimed at high energy performance and found that energy simulation experts were involved too late and in isolation from the design process. Furthermore, energy modeling was primarily used for regulatory compliance rather than as a tool for informed design decision-making. Their findings emphasized the need for strengthened early-stage collaboration among stakeholders to achieve energy performance goals [3].

A seamless integration of architectural design and simulation is essential to optimize energy performance during early design. One emerging approach in this regard is generative design (GD), which leverages algorithmic processes to automatically generate multiple design alternatives. This method reduces repetitive tasks and enables designers to explore creative solutions. GD shows potential in enhancing building energy performance by allowing extensive design space exploration and concurrent performance evaluations [4]. For example, Lee proposed a GD methodology that automatically generates and evaluates various conceptual layout options, effectively improving design productivity and reducing repetitive efforts [5]. Similarly, Suphavarophas et al. conducted a systematic review of GD applications in building energy efficiency and reported an increasing number of successful case studies demonstrating its effectiveness [6].

However, integrating robust energy analysis tools is essential to ensure that alternatives generated through GD are energy-efficient. Although prior research attempted to link design tools such as DesignBuilder and EnergyPlus for evaluating multiple alternatives, these approaches face limitations due to computational complexity and modeling overhead [7]. Accordingly, studies began incorporating machine learning and evolutionary algorithms into GD workflows. For instance, Kallioras and Lagaros [8] introduced MLGen, a framework combining machine learning with topology optimization for structural design. Barbieri and Muzzupappa [9] conducted a comparative study of performance-driven GD and conventional topology optimization tools.

Furthermore, Urquhart et al. applied generative algorithms to human-centered product development [10], expanding the utility of GD beyond architecture to various engineering domains. Moreover, physics-informed neural networks (PINNs) have recently been widely applied across numerous engineering fields, demonstrating significant advancements in solving complex problems [11,12,13]. Building upon this context, the current study explores an integrated approach combining GD with energy performance analysis to support energy optimization in the early stages of architectural design. The objective is to automate the generation and evaluation of diverse design alternatives, ultimately identifying optimal solutions and supporting energy-efficient design decision-making.

1.2. Research Methods

This study identifies key factors for early-stage energy optimization—building form, window-to-wall ratio (WWR), and orientation—through a review of prior research. Based on these factors, 154 apartment building forms were randomly collected from residential complexes in South Korea to generate layout alternatives using Generative Design (GD). The predicted energy consumption of these alternatives was then verified and evaluated to derive a comprehensive, optimized solution. Furthermore, this paper proposes a method for leveraging GD as a strategy for energy-optimized design from an integrated design perspective.

The specific methodology of this paper is as follows: Section 1 introduces the background, research objectives, and methodological approach. Section 2 presents the literature review, organized into three thematic areas:

(1)

energy optimization from the perspective of integrated design,

(2)

energy performance analysis of apartment buildings, and

(3)

GD approaches for energy optimization.

Section 3 describes the simulation-based analysis of GD energy performance.

(1)

Selecting 154 buildings from apartment complexes nationwide using digital maps from the National Geographic Information Institute and converting them into mass models in Revit based on key design parameters such as building type, WWR, and orientation.

(2)

Simulating the generated models in Green Building Studio (GBS) to establish an energy database according to the specified parameters. In this process, default GBS values for HVAC, occupancy schedules, and window performance were used, with some adjustments made in reference to the Korean Zero Energy Building technical guidelines.

(3)

Performing a multiple regression analysis on the database using SPSS 29, with building type, WWR, and orientation as independent variables and annual Energy Use Intensity (EUI) as the dependent variable. Statistical validity was confirmed by examining R², confidence intervals, and multicollinearity (VIF).

(4)

Utilizing a Dynamo-based generative design (GD) algorithm to generate 30 design alternatives and calculate the EUI for each.

(5)

Deriving an optimal design proposal by comprehensively evaluating energy efficiency, regulatory compliance, and economic feasibility (floor area ratio, number of units).

Finally, Section 4 summarizes the key findings and outlines directions for future research.

2. State of the Art

2.1. Theories and Case Studies on Energy Optimization from the Perspective of Integrated Design

Integrated Design refers to a collaborative design approach in which experts from various disciplines—such as architects, engineers, and contractors—work together from the early stages of the design process to improve the overall building performance and quality [14,15]. Traditionally, architectural design and engineering (e.g., structural, mechanical systems) were performed sequentially. However, the growing demand for high-performance buildings has increased efforts to achieve performance goals—such as energy efficiency—through collaborative work beginning in the early design phases [16]. To optimize performance beyond aesthetic outcomes, integrated design considers multiple factors simultaneously, including building form, layout, envelope design, and system planning [17]. Particularly in terms of energy efficiency, integrated design enables critical early-stage decisions through an embedded cycle of design, analysis, and feedback [18].

Earlier studies identified early collaboration, goal alignment, and tool integration as key elements of integrated design. According to Hashempour et al., projects adopting an integrated design process exhibited close communication between architects and energy modelers, allowing major energy-related decisions to be made during the design stage, significantly reducing the risk of design errors or underperformance [3]. Contrastingly, in projects lacking integration, energy modeling was often conducted independently of architectural design—frequently at later stages when modifications were no longer feasible—leading to limited improvements in performance [3,19]. These findings emphasize the importance of integrated design for enhancing energy performance and highlight the need for policy frameworks offering guidelines and incentives for its implementation [20].

One technical enabler of integration between design and energy analysis is the linkage between Building Information Modeling (BIM) and Building Energy Modeling (BEM). BIM allows for centralized management of design information, which can be synchronized with BEM tools to enhance the efficiency of integrated design [21]. Nevertheless, challenges persist in BIM–BEM interoperability [22]. Bastos Porsani et al. [23] conducted tests by converting BIM models into gbXML and IFC formats for use with EnergyPlus-based tools and reported issues such as missing walls, area discrepancies, and significant variations in simulation results—up to 6 to 900 times between tools—indicating critical interoperability limitations. These problems were especially pronounced in complex, large-scale buildings, suggesting that current semi-automated BIM–BEM workflows do not fully support integrated design [23]. Improving data interoperability and standardization between BIM software (Revit 2023 for Students) and simulation engines is a prerequisite for advancing integrated workflows [24].

Additionally, efforts were made to develop unified design-simulation platforms. For instance, Elbeltagi et al. [25] introduced a hybrid optimization approach combining parametric analysis, artificial neural networks (ANN), and genetic algorithms (GA) to quickly predict energy consumption in early-stage design. This approach significantly reduced the time required for iterative simulations while effectively generating energy-efficient residential building designs [25]. Such simulation–optimization integration supports the implementation of integrated design by enabling multi-objective performance improvements. Research in this domain is expanding to include evolutionary algorithms (e.g., NSGA-II) and machine learning techniques [26].

Improving energy efficiency in buildings also requires an integrated strategy combining passive design elements (e.g., high-performance insulation, optimized building envelopes) with efficient active systems. This approach targets a 30–70% reduction in energy load through envelope design, based on a whole-life-cycle energy perspective [27]. One practical application is the work of Lee and Park [28], who developed an integrated planning methodology for achieving Zero Energy Buildings (ZEBs) in office buildings under Korean climate conditions. Their simulations demonstrated that the optimal combination of passive (e.g., thermal transmittance, surface-to-volume ratio) and active (e.g., solar PV systems) elements varies depending on regional climate characteristics. For example, strengthening the envelope may be more critical in some regions, while prioritizing renewable energy generation may be more effective in others [28].

In summary, realizing energy optimization through integrated design requires early collaboration with shared goals, effective use of digital tools such as BIM–BEM, and incorporation of simulation-based optimization techniques. Integrated design is also highly compatible with GD. While GD diversifies design alternatives, and integrated design processes help coordinate and evaluate these alternatives to identify solutions balancing energy performance and design quality. The following section will examine the development of energy performance analysis techniques and their application in the multifamily housing sector, exploring the practical implementation of such integrated approaches.

2.2. Studies on Energy Optimization of Multifamily Housing Using Energy Performance Analysis Techniques

Energy performance analysis techniques for buildings, including multifamily housing, rapidly advanced by integrating simulation tools, big data, and machine learning [29,30]. Building Energy Performance Simulation (BEPS) emerged as a key method for optimizing energy efficiency by quantitatively evaluating various design alternatives during the design phase [31,32,33,34]. However, some studies highlighted that the early-stage application of BEPS in real-world practice remains limited [35,36]. Mahmoud et al. [1], in a survey of 418 architects in the UK, found that while most architectural firms recognized the importance of BEPS, they often did not actively use it in the early design stages. Besides projects pursuing green certifications like Passive House or BREEAM, early-stage simulations were rarely applied beyond basic code compliance calculations. The main barriers included the complexity of tools, time constraints, and a lack of trained personnel. Consequently, recent research suggests shifting focus from developing new simulation tools to methodologies for their practical integration into design workflows, highlighting the need for approaches naturally embedding simulation into the design process [1].

Despite these challenges, simulation-based optimization research continues to grow and significantly improves energy efficiency in multifamily housing [37]. Simulations enable quantitative assessment of the effect of various design parameters—such as insulation thickness, window-to-wall ratio, glazing performance, building shape, and orientation—on energy consumption [38]. For instance, Abbas et al. [39] analyzed the influence of building materials and fenestration characteristics on energy usage across residential, office, and warehouse buildings. They reported that increasing the window-to-wall ratio from 15–95% led to an ~80% increase in annual energy consumption per unit area in residential buildings. The increase was ~17% for office buildings under similar conditions.

Additionally, when evaluating energy consumption by rotating the building orientation 360°, asymmetrical residential floor plans showed a 25 MJ variation in annual energy use per unit area depending on orientation. Additionally, symmetrical office buildings showed a smaller 7 MJ variation, and rectangular warehouses showed only 0.7 MJ [39]. Such simulation-based sensitivity analyses help identify the most critical design variables for energy savings, tailored to local climates and building typologies [40]. This knowledge can be used to adjust design strategies or to inform multi-objective optimization algorithms seeking efficient design alternatives [41].

Evolutionary optimization algorithms are particularly prevalent in building energy optimization research [42]. For example, studies using the Non-dominated Sorting Genetic Algorithm II (NSGA-II) successfully improved energy consumption and thermal comfort [43]. Zhang et al. (2024) applied NSGA-II to find optimal trade-offs between energy use and indoor comfort, suggesting ideal combinations of insulation thickness and glazing performance [26]. Similarly, Zhang, H. et al. implemented a data-driven optimization approach using an artificial neural network (ANN) for energy prediction and genetic algorithms for optimization, effectively minimizing heating and cooling loads in residential buildings [44]. By combining machine learning techniques (e.g., surrogate models) with evolutionary algorithms, these approaches can efficiently explore optimal design solutions without requiring high-resolution simulations at every iteration, improving computational efficiency while still identifying global optima [44].

Adopting digital twin technology is another emerging trend in optimizing energy performance in multifamily housing. Rutkowski et al. [45] demonstrated a method for performing dynamic energy performance assessments from the conceptual design phase by synchronizing real buildings with virtual models in real time. Their study on four multifamily buildings in Poland revealed an average prediction error of less than 3.5% between simulated and measured energy performance, validating digital twins’ high accuracy and potential for early-stage energy prediction and minimizing downstream design changes [45]. Digital twins link the entire building life-cycle—from design and construction to operation—using data, enabling continuous feedback during design, and real-time calibration during operation, thereby enhancing the reliability of energy predictions [46].

Parallelly, studies addressed energy retrofitting and policy-related objectives for multifamily housing. Under the Energy Performance of Buildings Directive (EPBD) of the EU, large-scale energy performance improvements for existing buildings are targeted by 2050, with new constructions required to meet nearly zero-energy building (NZEB) standards by 2030 [47]. Alsabry et al. (2024) assessed the energy, economic, and environmental impacts of façade insulation upgrades and improved ventilation systems in multifamily housing in Poland [2]. Their results showed that the most favorable combination—cost and carbon reduction—involved enhancing envelope insulation and applying mechanical ventilation systems with heat recovery. These findings provide practical recommendations for achieving NZEB performance under local conditions and serve as valuable guidance for policymakers and designers [2].

A wide array of techniques—including simulation, optimization algorithms, digital twins, and machine learning—is being applied to improve energy performance in multifamily housing. These technologies enable scientific, data-driven decision-making from the earliest stages of design. Combined with generative and integrated design processes discussed earlier, they can generate synergistic effects. Practically, this would involve generating a wide range of design alternatives through GD, fostering interdisciplinary collaboration via integrated design, and using simulation-optimization tools to select the best solution. Such a unified framework could generate transformative improvements in building energy efficiency, particularly for multifamily housing; this has significant implications for the future of sustainable architecture and warrants further research and real-world application to verify its effectiveness.

2.3. Energy Optimization Through GD

In recent years, GD has gained significant traction in the architecture domain as a method for improving the energy performance of buildings. By automatically generating diverse design alternatives in the early stages and evaluating their performance, GD enables exploring solutions that may be overlooked in conventional design processes and supports identifying optimal options. Suphavarophas et al. [6] conducted a systematic review of 34 studies on GD and architectural energy efficiency, reporting that thermal performance improved by an average of 28%. In contrast, energy consumption was reduced by up to 23.3% through GD implementation [6]. These findings suggest that GD can offer substantial energy-saving potential compared with traditional design approaches. Many GD studies incorporate evolutionary algorithms and emphasize early-stage parametric exploration to derive optimal designs.

Lee et al. [5] proposed a design automation approach that generates building layout alternatives and performs partial performance evaluations using GD. This study employed a grid-based cellular algorithm to rapidly arrange multiple buildings on a site. Alternatives were filtered based on constraints such as building coverage and floor area ratios. Among the hundreds or thousands of generated alternatives, some closely resembled initial expert-generated designs, demonstrating the feasibility of practical solutions. The study confirmed that GD can significantly improve design productivity by enabling the rapid exploration of multiple viable design options with immediate feedback in the conceptual design phase [5].

Gradišar et al. [48] applied GD to optimize shading design for a building with a large glass curtain wall. Using a BIM-based model in Revit integrated with Dynamo and Project Refinery (a GD tool), the study generated and evaluated horizontal and vertical louvers configurations. The optimization objectives included maximizing shading effectiveness to reduce cooling loads, minimizing material use, and enhancing aesthetics. By exploring hundreds of alternatives using evolutionary algorithms, the researchers identified a shading solution achieving better cooling load reduction and material efficiency than those of manually designed counterparts. This study demonstrated the potential of the Revit–Dynamo–Refinery BIM-based GD workflow to enhance building energy performance without compromising aesthetics [48].

Banti et al. [49] applied GD in industrial building retrofitting, focusing on the optimal configuration of building-integrated photovoltaics (BIPV) and window shading. Using a Revit BIM model of an aging factory, the researchers integrated energy analysis tools via the Grasshopper plugin. Solar radiation simulations generated and evaluated alternatives for PV panel placement on different façade orientations to determine the highest annual electricity output. GD was also applied to explore window shading options and assess the impact of visual comfort. The study found that installing horizontal shades at 0.2–0.4 m intervals most effectively reduced indoor glare discomfort indices and lowered indoor temperatures by 0.46 °C. The BIM–GD integration demonstrated the quick generation of retrofit solutions by enhancing existing industrial buildings’ energy and environmental performance, while enabling economic and energy-informed decision-making [49].

Wan et al. [50] incorporated deep learning into GD to generate residential building floor plans. Focusing on the space allocation problem (SAP), they utilized a generative adversarial network (GAN) to automatically produce energy-efficient floor layouts based on user requirements. The proposed SD-GAN model referenced labeled data characterizing energy performance metrics (e.g., energy use relative to spatial composition), and the resulting layouts improved spatial thermal performance over baseline designs. For instance, GAN-generated layouts featured adjusted window areas and orientations to maximize daylight and natural ventilation and optimized living room openings to enhance passive energy-saving characteristics. The study showed that deep learning–based GD can be applied to architectural layout planning to automatically generate energy-efficient spatial configurations [50].

Storcz et al. [51] conducted a shape optimization study comparing the heating energy performance of various modular housing forms. Using six rectangular blocks, they generated all 167 possible combinations and simulated annual heating energy demand for each configuration. Results showed that annual heating demand varied significantly between 13.163 and 18.461 MWh depending on the form, with some forms consuming more than 40% additional energy compared with that of the optimal shape. The most efficient forms were compact, with minimal surface-to-volume ratios. In contrast, complex shapes with protrusions or increased height were less efficient due to a greater surface area for heat loss. This study highlighted the substantial impact of building geometry on energy performance. It demonstrated the value of integrating shape generation and performance evaluation within a GD framework to identify optimal configurations.

Mukkavaara et al. [4] proposed a GD-based framework for residential block development projects. Aimed at addressing complex problems across architectural design and various engineering disciplines, the study established an iterative design exploration framework using GD and applied it to a real-world case. The method was tested in a new residential block development in Kiruna, Sweden, where solar access duration was used as the performance indicator in the early conceptual design phase. The results demonstrated the effectiveness of GD in generating and evaluating diverse alternatives with similar building volumes by calculating and comparing the duration of interior sunlight exposure for each scheme. The study revealed that GD enabled the exploration of a broad spectrum of configurations—from low-rise, dispersed layouts to high-rise, dense arrangements—thereby uncovering potential optimal solutions that may otherwise be overlooked when relying solely on manual, intuitive design approaches [4].

Fang and Cho [52] optimized building floor plan geometry and window-to-wall ratio in a small office building to simultaneously improve daylighting and energy consumption. They built a parametric model in the Rhino-Grasshopper environment, integrated it with EnergyPlus and Radiance simulators, and performed optimization using a genetic algorithm. The study found that skylights’ horizontal and vertical dimensions were the most influential parameters, and the optimal window-to-wall ratio and building aspect ratios varied depending on the climate. This research demonstrated that multi-objective GD can quantitatively resolve the trade-off between daylighting and energy performance at the early design stage by deriving climate-appropriate building forms and opening configurations [52].

Du et al. [53] extended GD to the urban planning scale by generating and evaluating alternatives for high-rise residential developments in the high-density urban context of Guangzhou. They used parameters such as street networks, block forms, building types, and green space layouts to automatically generate 7500 urban form alternatives. They applied a multi-objective evolutionary algorithm to evaluate them. Eight environmental, social, and economic performance indicators were simulated for each alternative: urban density, green space area, daylight potential, heat stress, accessibility, view quality, shading impact, and construction cost. The study identified alternatives with outstanding overall performance through a two-stage Pareto optimization process. The results revealed trade-offs across development scenarios, from low-rise green-focused layouts to high-rise dense configurations. Some design alternatives achieved balanced performance in daylight, thermal environment, and view quality without increased construction cost. This study demonstrated the applicability of GD in urban-scale planning and highlighted its contribution to multi-objective optimization, including cost metrics, for high-rise development scenarios [53].

However, while these applications demonstrate the growing capabilities of GD, a closer review of the literature reveals several key limitations that this study aims to address. Most existing studies focused on individual buildings or isolated components (e.g., window systems, façade materials, or floor plans), aiming primarily to enhance energy or environmental performance. While a few studies considered economic viability or urban planning density, it is rare to find research incorporating legal regulations (e.g., FAR, solar access rights) and economic feasibility into the early-stage design process simultaneously. This study aims to overcome these limitations. Its core novelty lies in applying GD to the residential complex scale—not individual buildings—to optimize the arrangement and massing of multiple buildings within an integrated Revit–Dynamo–GBS environment. A key distinction of this research is the incorporation of (1) regulatory constraints (e.g., floor area ratio, building coverage ratio, inter-building distance, solar access rights), (2) economic indicators, and (3) energy efficiency into the performance function from the earliest design stages. This generates design alternatives that simultaneously satisfy these three critical requirements. This approach differentiates the study from prior work and represents a novel attempt to expand the scope of GD by accommodating real-world constraints in pursuit of multi-objective optimal design.

3. Energy Optimization Simulation

In Section 3, elements for energy optimization at the early design stage were categorized based on existing research into building typology, window-to-wall ratio (WWR), and building orientation. Based on these criteria, a database (DB) of predicted energy use was constructed using 154 randomly selected apartment types from domestic multifamily housing complexes. Regression analysis was then conducted to examine the energy performance of each indicator.

Using GD, various layout alternatives were generated for the selected 154 apartment types, building on prior research [54]. These alternatives were then evaluated and analyzed using the predicted energy use data, deriving an integrated design alternative to achieve optimal energy performance. Finally, the current state of building energy performance assessments was reviewed, and policy implications and strategies were proposed to enhance energy efficiency through GD-based design within an integrated design framework.

3.1. Defining Variables for Energy Performance Analysis

Based on the theoretical review, previous studies, and relevant building regulations concerning the layout of multifamily housing, key factors and scale conditions influencing energy performance in the early design stage were identified. These elements were then categorized as input variables applicable to the GBS energy simulation tool and are summarized in

Table 1. Input Variables and Constraints for Generative Design.

Category	Indicator Composition	Detailed Components
Regulatory-Based Input	Building setback boundary	Permissible area for placing multifamily housing
	Building coverage ratio, floor area ratio	Building scale (building area, total floor area, height)
	Sunlight, daylight	Required separation distances (adjacent lot boundary, inter-building spacing)
Energy Optimization Input	Building type	Slab-type, mixed-type (L-shape, V-shape, bent), tower-type (closed loop, plus-shape, V-tower)
	Window-to-wall ratio	0% (minimum), 35% (2020 Zero Energy Building Certification Standard), 95% (maximum)
	Orientation	Eight directions (South, Southeast, East, Northeast, North, Northwest, West, Southwest)

.

Inputs were broadly classified into regulation-based values and energy optimization inputs. These were further reorganized into detailed components for energy performance simulation purposes.

The location of the multifamily housing was set as a default parameter, and weather data were collected from the nearest meteorological station to the site. As an energy simulation tool, GBS (Green Building Studio) allows performance analysis during the early design phase by inputting fundamental project information such as the design phase, building type, and site location. More detailed analysis is also possible by entering parameters such as building height, orientation, HVAC systems, room/space configuration, and window specifications (WWR, type, shading depth, etc.). Accordingly, simulation parameters were classified into fixed and adjustable inputs, as organized in

Table 2. Default parameters of this study.

Category	Default Parameters	GBS Energy Settings: Detailed Parameters	Model Energy Settings	Description and Notes
Fixed	Building type: Multifamily housing (choose option)	Occupancy schedule	-	Residential 24 h
		Lighting/equipment schedule	-	Office lighting 6 AM to 11 PM
		Person/100 m2	-	25
		Activity level	-	Standing, light tasks, walking
		Sensible heat gain (w/person)	-	73
		Latent heat gain (w/person)	-	59
		Sensible heat gain (Btu/person)	-	250
		Latent heat gain (Btu/person)	-	200
		Lighting load density (W/square feet)	-	0.7
		Equipment load density	-	1
		Electric equipment radiant fraction	-	0.5
		Carpet	-	Y
		Condition type	-	Heating and cooling
		0 A L/S person	-	75
		O A flow per area (m3/time/m2)	-	37
		Unoccupied cooling setpoint	-	85
	Building operation schedule		-	Default
	HVAC system		Residential 17 SEER/96 HSFP split HP < 55 tons	Literature review
	Outdoor air information		Default	8.00 L/s per person
	Export category	Space/room data	Space
	Roof		-	Default
	Exterior wall		-	Default
	Window	Window type (choose option)	Low-E triple glazed SC = 0.65	According to the 2020 Zero Energy Building Certification Technical Guide, a glass type with SHGC of 0.4 or higher and the lowest possible U-value close to 12 was selected.
		Target window sill height	-	750
		Shading depth	-	457.2
		Target ceiling height ratio	-	0%
	Location		Seoul, Korea
	Mode		Use of conceptual mass and building elements
	Project phase		Demo
Adjustable	Orientation
	Number of floors (height)			Floor-to-floor height 3 m If fixed: 20 floors
	Gross floor area
	Building area
	Window	Window-to-wall ratio		30–35% (2020 Zero Energy Building Certification Technical Guide)

.

When parameters are undefined in the model, GBS uses default values to generate the minimum necessary energy model for simulation. These default values are primarily based on ASHRAE 90.1 [55], ASHRAE 90.2 [56], ASHRAE 62.1 [57], and CBECS [58] data, and may vary depending on building type, location, size, and number of stories [59]. As this study focuses on the early-stage design of multifamily housing, the building type was fixed as multifamily housing. According to the Zero Energy Building Certification Technical Guide and previous research, HVAC systems and window types were also defined based on domestic conditions.

Such simplifications may lead to discrepancies with actual conditions, thereby limiting the absolute accuracy of the results. Accordingly, some input values were adjusted by referencing the domestic Zero Energy Building technical guidelines, and this limitation should be taken into account when interpreting the findings.

Adjustable elements included orientation, WWR, building shape, and scale, all identified earlier as key factors in energy optimization design.

3.2. Building-Level Energy Performance Analysis

As illustrated in the Figure 1, this study sought to identify an optimal energy alternative by leveraging data generated across the Revit-Dynamo-GBS workflow.

First, we acquired data for 154 apartment building forms from digital maps in DWG format provided by the National Geographic Information Institute. These forms were then modeled in Revit, as illustrated in the figure, and energy simulations were performed using GBS. During this stage, modeling was conducted separately for each of the pre-defined adjustable factors, and the predicted energy consumption data collected from these simulations was compiled into an Excel file.

The compiled data was utilized to conduct a multiple regression analysis for each energy performance indicator, thereby identifying the most energy-efficient conditions for each metric.

Subsequently, an experiment was conducted on a target site to optimize the energy performance of an apartment complex based on this data. As illustrated in the Figure 2, the experimental sequence was as follows: First, data on legal regulations, the site, and building forms were organized. Second, this organized data was prepared for input into Dynamo. Third, Generative Design (GD) was used to automate the layout process and generate alternatives, which were then reviewed for regulatory compliance and energy performance to derive an energy-optimized layout plan. Finally, the derived layout was exported into various formats, such as 3D models and CAD files.

Afterwards, alternatives that failed to meet the constraints of building regulations (e.g., floor area ratio, building coverage ratio, and solar rights) were excluded. The remaining alternatives were then ranked based on energy efficiency (EUI) and economic feasibility (floor area ratio, number of households) to determine the optimal design proposal.

3.2.1. Establishment of the Energy Usage Database

To build an energy database based on predefined performance indicators, 200 residential buildings within apartment complexes across South Korea were collected using numerical maps in DWG format provided by the National Geographic Information Institute. Of these, 45 buildings causing modeling errors during energy model generation were excluded, resulting in a final dataset of 154 buildings. These were categorized into three main types: slab-, tower-, and mixed-type. The mixed- and tower-type categories were further subdivided into L-, V-, and bent-mixed and tower, plus-, and V-tower, as detailed in

Table 3. Domestic Study on Generative Design.

Classification	Classification
Slab-type
Mixed-type
	L-mixed-type	Bent-mixed-type	V-mixed-type
Tower-type
	Tower-type	+Tower-type	V-tower-type

.

The classification resulted in 54 slab-type, 61 L-mixed, 17 V-mixed, nine bent-mixed, six tower, three plus-tower, and 5 V-tower buildings. The classified building shapes were converted from DWG to Revit files to generate mass models, and floors were applied to each mass. The building location and weather data were set based on the GBS weather station in Jung-gu, Seoul. Parameters such as building type, HVAC system, and window-to-wall ratio were then adjusted to reflect the typical conditions of South Korean multifamily housing. To proceed with energy simulation, each model was converted into an energy model via the analysis tab, and simulations were run to transmit energy data to Green Building Studio. The resulting data included energy use intensity (EUI), annual electricity consumption, and annual fuel consumption based on weather data. Energy usage was compiled for each building according to relevant optimization indicators to construct a comprehensive energy use database. As this study does not consider other mechanical systems, only the EUI data were extracted and organized for analysis.

3.2.2. Analysis of the Impact of Energy Performance Indicators

Earlier studies on optimal energy models often lacked analytical tools, limiting their ability to account for the interactions among influencing factors.

To address this gap, we employed analytical tools to evaluate the effect of various energy performance indicators on energy usage. We compared the results with the existing literature to derive the most optimal building forms. SPSS was used for the statistical analysis, and multiple regression analysis was conducted to assess the influence of each indicator on energy consumption. Regression analysis is a statistical method used to estimate a mathematical model describing the relationship between variables based on empirical data, helping to explain the effect of independent variables on a dependent variable.

The dataset of 154 multifamily buildings identified optimal solutions at the building scale. Performance indicators were identified based on earlier studies and the data requirements of energy simulation tools used in early-stage design. An energy use database was built for each indicator, and its effects on energy performance were assessed. Initially, SPSS was used to explore correlations, followed by an average EUI and load analysis for each performance indicator, with particular attention to the window-to-wall ratio. This approach introduced a level of objectivity lacking in earlier studies by complementing empirical observations with correlation analysis using statistical tools.

(1) Analysis of Impact by Window-to-Wall Ratio (WWR)

First, based on the assumption from prior research that the Window-to-Wall Ratio (WWR) influences energy performance, we analyzed its effect on Energy Use Intensity (EUI). For the energy simulations, the WWR was set to the GBS minimum of 0%, the maximum of 95%, and 35% to align with the 30–35% certification standard from the Zero Energy Building guide. Other parameters were held constant: orientation was set to due south, and the number of stories was set to 35, corresponding to a high-rise building.

Based on this, the influence of WWR on energy consumption was statistically analyzed using SPSS. Initially, a multiple regression analysis was conducted with WWR and building area as independent variables. However, the building area was found to be statistically insignificant (p > 0.1) and was therefore excluded. Consequently, a simple linear regression analysis was performed with WWR as the sole independent variable to determine its impact on EUI, using the “Enter” method. The regression model can be represented as follows in Equation (1).

$$Y={\beta }_0+{\beta }_{1}X+ϵ$$

(1)

The results of the analysis are presented in

Table 4. The impact of the window-to-wall ratio (WWR) on energy consumption.

Variable	Unstandardized Coefficient		Standardized Coefficient	t(p)	F(p)	R2
	B	SE	β
Constant	804.919	13.880		57.991
window-to-wall ratio (WWR)	1.048	0.237	0.330	4.415	19.489 ***	0.109

*** p < 0.001.

. The regression model was found to be statistically significant (F = 19.489, p < 0.001), with an R² value of 0.109, indicating that the model explains 10.9% of the variance. The WWR showed a significant positive effect on EUI (β = 0.330, p < 0.001), meaning that as the WWR increases, the EUI also tends to increase. This confirms the findings of previous studies, demonstrating that window area impacts energy performance and that a smaller WWR is more efficient from an energy performance perspective. This underscores the importance of appropriately sizing window areas.

This characteristic ultimately indicates that WWR affects energy performance. Therefore, using this as a baseline, we proceeded to analyze the relationship between energy performance and building form and orientation.

(2) Analysis of Impact by Building Orientation

Next, we sought to analyze the effect of building orientation on EUI. The orientation was based on eight cardinal directions (S, SE, E, NE, N, NW, W, SW). We first statistically analyzed the impact of orientation on energy consumption using regression analysis in SPSS and then analyzed the average energy load for each orientation.

To facilitate the analysis, building orientation was converted into a continuous variable with 45-degree increments (0, 45, 90, 135, 180, 225, 270, 315). To improve simulation efficiency, given the large amount of data across all building types, the models were set to a single story for this part of the analysis.

The results are shown in

Table 5. The impact of the Orientation (Angle) on energy consumption.

Variable	Unstandardized Coefficient		Standardized Coefficient	t(p)	F(p)	R2
	B	SE	β
Constant	30.192	0.457		65.998
window-to-wall ratio (WWR)	0.094	0.036	0.043	2.592	6.719 *	0.002

* p < 0.05.

. The regression model was found to be appropriate (F = 6.719, p < 0.05), and building orientation had a significant effect on EUI (β = 0.043, p < 0.001). The positive β coefficient indicates that as the orientation angle increases (moving away from south), the EUI also tends to increase. This statistically suggests that a due-south orientation is the most efficient, followed by southeast, east, and northeast, in order. However, the R² value for this analysis was only 0.002, indicating an explanatory power of just 0.2%. This low value suggested that a direct examination of average energy use and loads was necessary.

Therefore, we analyzed the EUI and energy loads by orientation, using WWR as a baseline factor. This analysis was conducted on all building samples that included information on form, orientation, and WWR.

First, an examination of the average EUI by orientation, as shown in Figure 3, Figure 4 and Figure 5, confirmed the regression analysis finding: as the building rotates away from south, the EUI increases. Furthermore, EUI increased sharply when the building faced north before decreasing again, a phenomenon predicted to be caused by facade changes corresponding to the core’s location. It was also observed that the rate of EUI increase was greater for smaller WWRs.

Next, when the energy loads for each orientation were classified into Heating and Cooling based on WWR (as shown in Figure 4 and Figure 5), it was found that energy loads for both increased as the WWR and orientation angle increased, though the rate of increase varied.

(3) Analysis of Impact by Building Form

Finally, to analyze how building form affects energy performance, we analyzed the EUI by building type, again using WWR as a baseline influencing factor.

This study aimed to move beyond prior research—which often only observed energy consumption by type without analytical tools—by using statistical methods to determine which forms have the most significant impact on energy, thereby establishing greater objectivity. The building forms were based on a collection and classification of existing apartment types in South Korea to ensure a practical analysis.

To isolate the impact of form, other variables were held constant: WWR was set to 35% (in line with Zero Energy Building standards), the number of stories was set to 35 (high-rise), and orientation was set to south. The analysis focused on EUI (MJ/m²/year). Because building type is a nominal scale variable rather than an interval or ratio scale, a multiple linear regression analysis using dummy variables was performed with the “Enter” method. The model was deemed appropriate, with an F-statistic of 146.756 (p < 0.001). The adjusted R² was 0.870, indicating that the model explains 87% of the variance. The results are shown in

Table 6. The impact of the main building form on energy consumption.

Variable	Unstandardized Coefficient		Standardized Coefficient	t(p)	VIF
	B	SE	β
Constant	816.867	11.175		73.100 ***
L-Mixed	−238.369	16.258	−0.460	−14.662 **	1.157
V-Mixed	1374.012	53.978	0.752	25.455	1.024
Bent-Mixed	82.682	32.473	0.077	2.546	1.068
Tower	−111.126	38.978	−0.085	−2.851	1.047
+Tower	−167.905	53.978	−0.092	−3.111	1.024
V-Tower	−182.969	42.405	−0.128	−4.315	1.040
F(p)	146.756 **
adj. R2	0.870
Durbin- Watson	0.873

** p < 0.01, *** p < 0.001.

. All variables were found to be significant (p < 0.05), influencing EUI. When substituted into the multiple regression equation, it can be seen that the V-mixed and bent-mixed types have a relatively higher impact on EUI than the slab-type (the reference category). The regression model can be represented as follows in Equation (2).

$$Y={\beta }_0+{\beta }_1{X}_2+\cdots +{\beta }_i{X}_i+ϵ$$

(2)

Next, we analyzed the energy performance according to building form, using WWR as a baseline. The performance was analyzed based on a south orientation, comparing the changes in EUI according to WWR. To account for variations in building size (area) by type, the analysis used the energy consumption per unit of average area for that type.

The EUI by building form is shown in Figure 3, Figure 4 and Figure 5. At a WWR of 35%, the L-mixed-type had the lowest EUI, followed in order by the tower-type, +tower-type, V-tower-type, V-mixed-type, slab-type, and bent-mixed-type. A lower value indicates a more efficient form. However, efficiency varied depending on the WWR. For instance, the V-mixed and tower-types were more efficient at lower WWRs, whereas the slab-type showed higher efficiency at higher WWRs. This suggests that selecting an appropriate window area is critical and depends on the building form.

The findings showed that smaller window areas were associated with greater energy efficiency, and buildings with a due south orientation were more energy-efficient. These results align with trends identified in prior research regarding window-to-wall ratio and building orientation. However, in terms of building form, we differed from an earlier study [60] that analyzed nine cases across four building types. A more granular classification of seven building types was applied to 154 samples. As shown in Figure 6, the L-mixed type emerged as the most energy-efficient form, while the V-mixed type was the least efficient.

This simulation-based approach to deriving optimal energy performance at the building level is significant because it enables the construction of a predictive energy usage database for each form type. Considering these characteristics during the early design stage—before specific architectural elements are introduced—more efficient multifamily housing complexes can be achieved.

3.3. Energy Performance Review at the Complex Level

The energy performance review at the apartment complex level was conducted by applying the predicted EUI of the building types used in each site plan, based on the results of the building-level analysis. The goal was to identify and analyze the characteristics of the most energy-efficient layout alternatives. Based on prior research, the layout alternatives were derived using GD within the boundaries of legal regulations, resulting in 30 design scenarios [54].

Five alternatives were selected based on their energy efficiency, specifically those with the lowest EUI. The information for each of these five alternatives, ordered by energy performance, is summarized in

Table 7. Information on the Top 5 Energy-Performing Alternatives Derived via GD.

Rank	Site Information		Building Information
			Type	Gross Floor Area (m2)	Building Area (m2)	Building Type	Orientation	Average Number of Floors	Number of Buildings
1st	Number of Households (units)	4187	A	20,475.8	276.7	Slab-type	South	74	3
	Floor Area Ratio (%)	297.4	B	51,955.4	702.1	Slab-type	South	74	1
	Building Coverage Ratio (%)	4	C	42,194.8	570.2	L-mixed-type	South	74	2
	Number of Buildings (blocks)	14	D	57,172.4	772.6	L-mixed-type	South	74	5
	Average Number of Floors	74	E	54,042.2	730.3	V-mixed-type	South	74	3
	Average EUI (MJ/m2/year)		684.9
2nd	Number of Households (units)	3537	A	20,476.8	276.7	Slab-type	South	74	2
	Floor Area Ratio (%)	267.4
			B	51,955.4	702.1	Slab-type	South	74	3
	Building Coverage Ratio (%)	3.6
			C	42,194.8	570.2	L-mixed-type	South	74	1
	Number of Buildings (blocks)	12
			D	57,172.4	772.6	L-mixed-type	South	74	6
	Average Number of Floors	74
	Average EUI (MJ/m2/year)		699.5
3rd	Number of Households (units)	3948	B	37,211.3	702.1	Slab-type	South	53	4
	Floor Area Ratio (%)	259.8
			C	30,220.6	570.2	L-mixed-type	South	53	10
	Building Coverage Ratio (%)	4.9
			D	40,947.8	772.6	L-mixed-type	South	53	2
	Number of Buildings (blocks)	17
			E	38,705.9	730.3	V-mixed-type	South	53	1
	Average Number of Floors	53
	Average EUI (MJ/m2/year)		703.7
4th	Number of Households (units)	4073	A	15,218.5	276.7	Slab-type	South	55	2
	Floor Area Ratio (%)	286.5	B	38,615.5	702.1	Slab-type	South	55	3
	Building Coverage Ratio (%)	5	C	31,361.0	570.2	L-mixed-type	South	55	4
	Number of Buildings (blocks)	17	D	42,493.0	772.6	L-mixed-type	South	55	5
	Average Number of Floors	55	E	40,166.5	730.3	V-mixed-type	South	55	3
	Average EUI (MJ/m2/year)		707.1
5th	Number of Households (units)	4198	A	11,898.1	276.7	Slab-type	South	43	2
	Floor Area Ratio (%)	273.4	B	30,190.3	702.1	Slab-type	South	43	5
	Building Coverage Ratio (%)	6.3	C	24,518.6	570.2	L-mixed-type	South	43	5
	Number of Buildings (blocks)	21	D	33,221.8	772.6	L-mixed-type	South	43	6
	Average Number of Floors	43	E	31,402.9	730.3	V-mixed-type	South	43	3
	Average EUI (MJ/m2/year)		720.1

.

Subsequently, these high-performing alternatives were compared with those of the least energy-efficient scenario among the 30 GD-generated alternatives. As compliance with regulations, economic feasibility, and optimal residential environments are essential in multifamily housing [61], these factors were also considered to derive implications.

As shown in

Table 8. Energy-Optimized Layout Alternative Generated by GD.

View				Information
				Number of Households (units)		4187
				Floor Area Ratio (%)		297.4
				Building Coverage Ratio (%)		4
				Number of Buildings (blocks)		14
				Average Number of Floors		74
				Average EUI (MJ/m2/year)		684.9
Building Information
Type	Gross Floor Area (m2)	Building Area (m2)	Building Type	Orientation	Number of Floors	Number of Buildings
A	20,475.8	276.7	Slab-type	South	74	3
B	51,955.4	702.1	Slab-type	South	74	1
C	42,194.8	570.2	L-mixed-type	South	74	2
D	57,172.4	772.6	L-mixed-type	South	74	5
E	54,042.2	730.3	V-mixed-type	South	74	3

, the most energy-efficient alternative also had the highest floor area ratio (FAR) and average building height. Among all 30 alternatives, it recorded the highest FAR, with a building coverage ratio (BCR) of 4%, and included 14 residential blocks.

Contrastingly, the 5th ranked alternative had an EUI of 720.1 MJ/m²/year, and its FAR was 273.4%, ranking eighth-lowest among all alternatives. This alternative featured 21 blocks, an average height of 43 floors, and a BCR of 6.3%. Compared with the most efficient design, it had a lower FAR and average height but a greater number of blocks and a higher BCR.

As shown in

Table 9. Least Energy-Efficient Layout Alternative Generated by GD.

View				Information
				Number of Households (units)		4927
				Floor Area Ratio (%)		250.3
				Building Coverage Ratio (%)		10.7
				Number of Buildings (blocks)		40
				Average Number of Floors		Approx. 22
				Average EUI (MJ/m2/year)		1071.6
Building Information
Type	Gross Floor Area (m2)	Building Area (m2)	Building Type	Orientation	Number of Floors	Number of Buildings
A-1	8301.0	276.7	Slab-type	South	30	1
A-2	8301.0	276.7	Slab-type	West	30	8
B-1	14,744.1	702.1	Slab-type	South	21	3
B-2	20,389.9	703.1	Slab-type	West	29	2
C	13,684.8	570.2	L-mixed-type	West	24	4
E-1	16,796.9	730.3	V-mixed-type	South	23	2
E-2	15,336.3	730.3	V-mixed-type	Southwest	21	20

, the single least efficient alternative had an EUI of 1071.6 MJ/m²/year and the lowest FAR among all alternatives at 250.3%. The buildings in this scenario were primarily slab-type, with a BCR of 10.7%, 40 blocks, and an average height of 22 floors.

The most energy-efficient alternative featured the highest area ratio of L-mixed buildings at 57.3%, followed by slab-type (17.6%), and V-mixed (25.1%). Similarly, the relatively less efficient alternative also had a high proportion of L-mixed buildings (54.5%), with slab-type at 29.6% and V-mixed at 15.9%. Contrastingly, the least efficient alternative had only 9.9% of L-mixed buildings. In comparison, V-mixed accounted for 61.3% and slab-type for 28.8%, indicating that a higher proportion of energy-efficient building types, such as L-mixed, significantly contributed to the overall energy performance of the complex.

However, as previously discussed, traditional site planning methods are often driven by the subjective judgment of the designer and lack objectivity. Furthermore, energy performance evaluations typically require detailed design information, making mid- or late-stage design revisions time-consuming when performance targets are unmet.

To overcome these limitations, we applied GD to streamline the decision-making process and reduce the time required for manual calculations. By integrating energy performance analysis at the early design stage, we quickly identified reliable high-performance alternatives. Specifically, using GD to generate layout scenarios enables the designer to increase the proportion of energy-efficient buildings while also deriving objectively optimized layouts that comply with regulations and are economically viable. Furthermore, this approach allows for a phased energy optimization process: first, various layout alternatives are generated via GD, followed by energy performance evaluations. Next, regulatory and economic reviews are conducted based on the configuration of different building types. This process facilitates the derivation of design alternatives that can be easily understood across project stakeholders—including clients, designers, and contractors—and supports efficient integrated design for optimal outcomes.

4. Discussion

Earlier studies exhibited three key limitations. First, a lack of research exists utilizing GD for energy optimization at the multifamily housing complex scale. Second, few studies comprehensively considered automated site layout and energy optimization within an integrated design framework from the early design phase. Third, existing workflows often suffered from compatibility issues between modeling tools and simulation platforms, leading to inefficiencies in energy analysis.

Therefore, we defined energy performance evaluation indicators through a review of prior research and proposed an integrated design approach for optimizing energy performance in the early design stage. Specifically, three key design parameters—Window-to-Wall Ratio (WWR), building orientation, and building type—were identified and used to construct a building-level energy use prediction database through GBS simulations. This database was then applied to 30 legally compliant building layout alternatives generated via GD, and site-level energy performance and economic feasibility were assessed to identify optimal configurations.

This study yielded several novel insights in multifamily housing design optimization research. First, by leveraging the high interoperability of Revit-Dynamo-GBS, the workflow eliminated the need for separate file conversions, enabling the simultaneous automation of site layout and energy performance evaluation. Unlike conventional tools, GBS allows for early-stage performance analysis, making it possible to determine optimal site layouts during the conceptual phase; this demonstrates the practical applicability of Revit-Dynamo-GBS for integrated design in multifamily housing projects.

Second, by ranking the generated layout alternatives by energy performance and analyzing their characteristics, we found that a higher proportion of energy-efficient building types corresponded with improved overall energy performance and increased FAR; this highlights the importance of maximizing the share of energy-efficient building forms. Often driven by subjective judgment, traditional design methods struggle to quickly and objectively derive optimal alternatives. Contrastingly, we used GD scripts based on zoning regulations to generate objective layouts and conducted performance and massing evaluations to identify energy- and cost-efficient configurations. Consequently, the method enables objective and regulation-compliant optimization of multifamily housing layout design using GD.

Furthermore, from a policy perspective, we examined the certification system and criteria for multifamily housing energy performance evaluation and proposed strategies for applying GD to energy optimization from the early design phase. It recommends classifying certification items and criteria by the model development (LOD) level, enabling performance requirements to be tailored to design maturity.

We offer three primary contributions. First, it presents a novel application of GD for energy optimization in multifamily housing complexes from an integrated design perspective. While earlier research explored indoor spatial layouts or energy performance at the building scale, few studies addressed energy optimization during the early design stage—when performance gains can be maximized—within an integrated design framework. By proposing a GD-based approach to optimizing energy performance during this critical phase, the study contributes to maximizing energy efficiency in housing complexes and reducing carbon emissions on a broader scale.

Second, we demonstrate objectifying a traditionally subjective, experience-driven design process using a data-driven GD approach. We generate layout alternatives grounded in real-world constraints by applying numerical regulatory parameters and leveraging energy consumption prediction data for 154 buildings across seven typologies. Prior research was reviewed to identify energy-influencing factors at the early design stage, and corresponding energy use prediction data were constructed for each building typology, enhancing the objectivity and realism of the resulting design solutions.

Third, we integrate the automated generation of layout alternatives with energy performance analysis to derive the optimal energy-efficient layout for multifamily housing complexes. Whereas earlier research often struggled with incompatibility between modeling and energy simulation tools—requiring manual conversion and extensive review—the proposed method uses Revit–Dynamo–GBS to automate layout generation and performance evaluation. This integrated and efficient workflow not only improves design productivity but also offers a viable strategy to address the construction industry’s low productivity and environmental challenges.

From a policy perspective, we emphasize the need to improve existing regulations to support energy performance optimization from the early design stage. While many countries implement legal mandates or certification systems to reduce building energy consumption, most focus on post-occupancy performance after construction. Future policy directions could require mandatory energy simulation during the design phase, with buildings meeting a certain performance level as a prerequisite for permit approval [62]. Such initiatives would encourage designers to explore GD-based alternatives early on, steering their designs toward energy targets and ultimately contributing to national energy savings and environmental sustainability goals. Furthermore, if government-mandated BIM design policies were expanded to include guidelines linking energy simulation and GD techniques, the widespread adoption of integrated design approaches in practice could be accelerated.

Lastly, we suggest several directions for future research. First, GD was applied based solely on mass models, without detailing internal spatial layouts or unit typologies. Future studies should incorporate residential zoning and modular unit configurations to comprehensively address housing quality. Second, although we utilized pre-simulated GBS results in a static database, future research could integrate energy analysis algorithms directly into the Dynamo script to enable real-time performance feedback during GD iterations. This enhancement would enable more efficient solution discovery, and developing and validating such a prototype remains a task for future work. Furthermore, to build upon the predictive models used in this study, future research could leverage advanced machine learning methods, such as the physics-informed neural networks (PINNs) introduced earlier, to potentially achieve even higher accuracy and computational efficiency in energy performance prediction. Third, practical housing complex design must consider building spacing, pedestrian circulation, wind corridors, and outdoor community spaces. Furthermore, expanding the optimization objectives to include other crucial performance criteria, such as daylight availability, thermal comfort, and embodied carbon, would enrich the analysis and lead to more holistic and realistic design outcomes. Therefore, future studies should expand the scope of optimization by including these urban and environmental parameters as constraints or objectives. For example, linking GD with GIS-based contextual data or microclimate analysis could improve the contextual suitability of layout alternatives [63]. Furthermore, integrating BIM models with GIS data—such as site topography, surrounding buildings, and solar/wind environments—could facilitate larger-scale energy optimization. Finally, this study focused on passive design strategies determined by architectural form. Future research should integrate the analysis of active systems by modeling various HVAC configurations to more accurately predict and minimize long-term operating costs, thus bridging the gap between architectural optimization and mechanical engineering. Advancing automated design techniques in this direction would not only overcome the limitations of uniform housing complex layouts but also significantly contribute to improving the productivity and sustainability of the architecture industry.

5. Conclusions

We propose an integrated and automated approach to multifamily housing site planning and energy performance evaluation by linking a BIM-based design tool (Revit), visual programming (Dynamo), and a cloud-based energy simulation tool (Green Building Studio, GBS). Unlike conventional design methods that rely heavily on the experience of professional designers, this approach enables objective and systematic decision-making in the early design phase.

Building on this workflow, the study confirms that maximizing the share of energy-efficient building forms and automating layout generation and energy evaluation can simultaneously enhance energy performance, FAR, and design productivity. The demonstrated interoperability of Revit–Dynamo–GBS proves its practicality for early-stage, regulation-compliant, and data-driven decision-making in multifamily housing projects.

The simulation of the 30 alternatives generated via the GD algorithm took a total of approximately 6 h, with an average processing time of about 12 min per scenario. The analysis was conducted on a system with an Intel i7 CPU and 32GB of RAM, and the processing speed was enhanced by leveraging the cloud computing capabilities of Green Building Studio (GBS).

It is estimated that expanding the number of scenarios to 100 would require approximately 20 h. This represents a significant time saving compared to manual design and individual simulation processes. This finding indicates that the GD-based workflow achieves a level of computational efficiency sufficient for practical application in professional settings.

Policy analysis underscores that early-phase mandates—such as compulsory energy simulation before permit approval and LOD-based certification criteria—could institutionalize GD-driven optimization and accelerate the diffusion of integrated design practices. Such initiatives align with broader national goals for carbon reduction and industry-wide productivity gains.

While this study presents a robust workflow, future work should address its current limitations by following a clearer roadmap to evolve its capabilities. The design scope should be enriched by integrating modular unit typologies and internal spatial layouts, allowing the optimization to consider not only energy performance but also housing quality. Furthermore, the framework can be enhanced by incorporating urban environmental parameters, such as pedestrian circulation and wind corridor simulations, to create more contextually responsive site plans. On a larger scale, coupling the BIM-GD process with GIS data for site topography and surrounding contexts, alongside microclimate modeling, would enable more sophisticated and resilient optimization. Advancing these directions will move the field beyond uniform massing practices toward truly adaptive, sustainable, and high-performing housing complexes.