Multi-Objective Optimization of Façade and Roof Opening Configurations for Sustainable Industrial Heritage Retrofit: Enhancing Daylight Availability, Non-Visual Potential, and Energy Performance

Abstract

During the adaptive reuse of industrial heritage buildings, existing opening systems and envelope performance often pose major constraints. These restrictions make it difficult for the building to meet the requirements of the updated indoor environment, resulting in insufficient daylight and increased energy consumption. Therefore, optimizing lighting and energy performance has become the primary goal of the retrofit design. However, with limited interventions, the retrofit of heritage buildings to achieve significant overall performance improvement is still a challenge. From a sustainability perspective, improving daylight utilization and reducing energy demand are essential strategies for achieving low-carbon and resource-efficient building retrofit. This study proposes a grid-based parametric multi-objective optimization approach to optimize the window openings of the building envelope. The approach defines the position, size and material properties of the roof and facade openings as design variables. Implemented via the Honeybee and Octopus platforms, it integrates a genetic algorithm with EnergyPlus and Radiance simulations to co-optimize daylight performance, circadian frequency, and energy use intensity. Taking a single-story typical industrial heritage building in China’s cold climate zone as a case study, it is shown that coordinated multi-objective constraints significantly improve the overall performance across various evaluation metrics. The optimization results also provide interpretable window configuration strategies and recommended parameter ranges, which fully consider the climate adaptability of the surrounding environment. These findings offer useful guidance for sustainable retrofit design decision-making in similar single-story industrial heritage buildings.

1. Introduction

As urban development in China transitions from incremental expansion to stock-based optimization, industrial heritage buildings have become an important part of urban renewal [1,2]. Their adaptive reuse not only retains historical value but also introduces new spatial functions and social significance [3]. In the context of global green and low-carbon development, the sustainable retrofitting of such buildings is widely considered a key strategy for reducing resource consumption and environmental impacts in the building sector. Compared with newly built buildings, industrial heritage buildings were originally designed to accommodate large-scale machinery and assembly line production. This design background creates its unique architectural and technical features [4], including large-span spaces, heavy building shells and relatively simple window layout. Although its spacious space scale and flexible layout provide favorable conditions for functional adaptation, the overall architectural form still reflects the original mechanical-oriented design logic. With the growth of the cultural and creative industries, the adaptive reuse of existing buildings into office spaces or artistic spaces has become increasingly common [5]. This transformation raises performance requirements for health, low carbon and energy efficiency [6], and also reflects the necessity of bridging the gap between existing environmental performance and contemporary functional demands as well as sustainability goals during the retrofitting process [7]. However, large-scale alterations to building form are typically constrained by the dual requirements of heritage conservation and structural safety [8]. To maintain the authenticity of industrial heritage buildings, interventions should be minimized. Against this background, daylight-oriented space can be strategically introduced to improve the comfort of occupants while avoiding the loss of the character-defining features of the building [9]. The effect of daylighting interventions largely depends on the configuration of building openings. As the main interface between indoor and outdoor environments, openings have a significant impact on energy consumption and indoor environmental quality, making their configuration a critical focus in energy retrofit strategies [10,11]. Appropriate opening design can integrate daylighting strategies, provide uniform and comfortable daylight distribution, reduce electric lighting demand, and modulate solar heat gains to balance heating and cooling loads, thereby improving overall building performance [12,13]. In addition, the non-visual effects of daylight also affect the circadian rhythm, psychological state and work efficiency of the occupants [13,14]. Despite these advantages, daylight intensity, direction, and spectral characteristics vary dynamically over time and across climatic conditions, increasing the complexity of daylight utilization and control [15,16,17]. Therefore, how to achieve a balance between daylight availability, non-visual potential and energy performance through high-precision optimization of window configuration has become a key challenge for sustainable renewal of industrial heritage buildings.

Many studies have explored the energy-saving retrofit and environmental performance improvement of historical buildings from the perspective of opening design. Marzouk et al. [11] used the Rhino and Grasshopper platforms to simulate daylight and energy consumption in a historical palace in Egypt, optimizing the geometry and material properties of the skylight. Subsequent research [9] further evaluated the optimal configuration of vertical and horizontal skylight mullions in the palace, considering size and glazing material as key parameters. Al-Sallal et al. [18] evaluated the daylight performance of a traditional courtyard house in the United Arab Emirates transformed into a history museum by optimizing the space layout, lighting design and sunshade system. Piraei et al. [19] studied daylight optimization strategy for transforming historical buildings into office spaces, examining the use of skylights and light courts to meet the needs of lighting and vision. Sönmez et al. [20] focused on roof daylight enhancement strategies for historic buildings, employing simulation and optimization methods to improve indoor lighting performance by adjusting skylight configuration. Nocera et al. [21] assessed daylight performance in a school located within a historic building and proposed multiple window retrofit solutions to enhance classroom visual comfort while preserving heritage value. Wu et al. [22] explored daylight optimization strategies for traditional dwellings in western Hunan that improve daylighting performance at low cost while maintaining historical integrity. Shirzadnia et al. [23] optimized the skylight design parameters of a historic boiler room in Iran and provided design suggestions for lighting improvement. These studies show that daylight design based on opening optimization has been widely used in the retrofit projects of buildings such as exhibition halls, residences and office buildings, showing its potential in improving visual comfort and energy efficiency. Although the demand for circadian-related lighting performance in long-term occupied environments such as offices is increasing [24], existing building optimization studies have paid relatively limited attention to the non-visual effects of the daylight environment. In recent years, static metrics for single time points, including equivalent melanopic lux (EML), melanopic equivalent daylight illuminance (m-EDI), and circadian stimulus (CS), as well as the dynamic metric Circadian Frequency (CF), which is mainly used for annual evaluation of building spaces, have been increasingly applied to assess the non-visual potential of indoor environments. For example, Anaraki et al. [25] used the EML metric to evaluate the effects of office partition layout, height, and optical properties on circadian potential. Potočnik et al. [26] combined the CS and EML metrics to investigate the influence of different glazing types and interior wall colors on the non-visual effects of daylight in cellular offices. Jin et al. [27] controlled daylight availability by adjusting shading devices under different artificial lighting conditions and used the CS metric to evaluate subjects’ circadian stimulus levels. Konis [24,28] proposed the Circadian Frequency (CF) metric and applied it to assess the non-visual daylighting performance of a multi-story commercial office building in downtown Los Angeles under different floor plate configurations, and the results showed that CF and related metrics can effectively reflect the non-visual daylighting performance of spaces on different floors. Xu et al. [29] conducted annual simulations and used the CF metric together with other daylighting metrics to evaluate the overall daylighting performance of a classroom in Suzhou, China. Hao et al. [30] used the CF metric to analyze the annual variation characteristics of vertical m-EDI in specific areas of a classroom in Suzhou. However, although these studies have introduced non-visual metrics into the evaluation of building daylight environments, most of them remain at the level of assessing non-visual potential using a single metric and rarely incorporate them into a multi-objective optimization framework for systematic analysis. In particular, the CF metric integrates multidimensional information such as time and space into a unified whole and demonstrates strong comprehensive expressiveness, making it highly promising for multi-objective optimization assessment. However, the synergistic relationships between the CF metric and other building environmental performance metrics, as well as their specific implications for building retrofit design strategies, remain unclear and require further investigation. On the other hand, from an overall perspective, existing retrofit optimization studies tend to focus on parameter adjustments within a limited scope, such as specific skylight types, shading devices, or opening areas. The optimization process is typically based on predefined opening rules, which to some extent constrains the expansion of the design space and makes it difficult to explore more flexible, irregular, and diverse opening configuration strategies. At the same time, such optimizations are often conducted for a single orientation, with limited consideration of the synergistic effects between skylights and side windows, and there is also a lack of systematic analysis of combined applications of different glazing types. These limitations prevent the full potential of opening design in regulating light and thermal environments and improving energy performance from being fully realized. In contrast, conducting optimization within an integrated framework that considers multiple orientations, diverse opening forms, and combinations of materials can help enhance overall building performance under constrained intervention conditions. Meanwhile, some studies have begun to explore the use of grid-based subdivision of building envelopes to coordinate opening configurations and material properties within the optimization process. For example, Fathy et al. [31] proposed a facade-based natural lighting optimization framework for museum exhibition space. By pixelating the façade into multiple micro-openings and employing a genetic algorithm to identify optimal combinations of opening positions and dimensions, the study demonstrated the flexibility and adaptive potential of free-form façade daylighting design. Ma et al. [16] conducted multi-parameter optimization analyses of opening configurations and material selections for various skylight typologies, revealing that diversified opening patterns generated through irregular layouts and multi-material combinations could achieve more balanced environmental comfort and energy performance across different climatic conditions. However, current studies are mostly conducted on a small spatial scale, where computational complexity remains relatively low. In contrast, building retrofit projects often involve complex envelope surfaces and large-scale spaces. Directly applying existing methods in such contexts can lead to a significant increase in computational demand, making optimization difficult to implement. Therefore, free-form grid-based opening optimization methods for complex scenarios, as well as their practical application performance, still require further investigation.

To address these limitations, this study proposes a multi-objective optimization approach for retrofitting openings in industrial heritage buildings. Based on a grid-based subdivision of the building envelope, opening variables are defined to systematically explore configuration combinations. An existing industrial heritage building serves as the case study under a conversion scenario to office use, where the optimization pursues coordinated improvement of daylight performance, non-visual potential, and energy consumption. The main goal is to enhance overall building performance through daylight-oriented design strategies under limited intervention, thereby supporting the sustainable transformation of industrial heritage buildings in terms of energy efficiency and indoor environmental quality, while also promoting the sustainability of urban development.

2. Methodology

Figure 1 shows the proposed multi-objective optimization approach, which includes four continuous stages: (1) field investigation and measurement; (2) building performance simulation; (3) multi-objective optimization exploration; (4) optimization result analysis and design strategy extraction. The field investigation provides a comprehensive understanding of the types and preservation status of industrial heritage buildings in the study area, followed by selection of a representative case. Geometric and spatial data are used to build baseline models, while indoor environmental measurements are used for subsequent simulation validation. Define appropriate opening areas and design variables based on the building’s structural system and spatial layout. After determining the performance metrics, the Honeybee program in the Rhino-Grasshopper platform was used to simulate daylight and energy consumption. Then, the genetic algorithm in Octopus is used for multi-objective optimization to generate Pareto optimal solutions for different opening parameter combinations. This workflow is applied to a single-story industrial heritage building in Zhengzhou, China, and produces a practical opening and retrofit design plan.

2.1. Case Study Description

The Zhengzhou Oil and Fat Chemical Plant, established in 1952 in Zhengzhou, Henan Province, China (34.78° N, 113.65° E), was a key industrial project during the First Five-Year Plan. The plant stopped production in 2006, and then retained several representative buildings, forming a relatively complete early industrial complex. The workshops are arranged in parallel, and the adjacent buildings share longitudinal load-bearing walls, forming an overall structural unit. This layout was a typical feature of industrial buildings at that time, which not only reduced material consumption but also improved structural integrity. Figure 2 shows the internal and external situation of the selected single-span workshop. The building is a rectangular plane running east–west, measuring 39.2 m in length and 12 m in width. The roof adopts a double-slope roof, with a slope of 35°, a ridge height of 8.3 m, and an eaves height of 4.1 m. The high windows of the north facade introduce diffuse daylight, while the surrounding buildings of similar size partially block the lateral sunlight, which highlights the importance of skylight design in improving indoor lighting performance.

Zhengzhou has a northern temperate continental monsoon climate with four distinct seasons. The average annual temperature is 15.4 °C, the annual precipitation is about 631.3 mm, and the annual sunshine hours are about 1870 h. The horizontal illumination is 35 to 40 klux, indicating that there are sufficient sunlight resources. According to GB 50176-2016 [32], Zhengzhou belongs to a cold climate zone. The average monthly temperature of the coldest month is between −10 °C and 0 °C, and 90–145 days per year record a daily mean temperature below 5 °C. Winters are cold and dry, while summers are hot and rainy, resulting in a large annual temperature difference. Although favorable daylight conditions support natural lighting and passive solar heat gains in retrofit design, such conditions also increase the risk of heat loss through openings during winter.

2.2. Evaluation Metrics

The optimization objectives include annual dynamic daylight metrics, circadian-related metrics, and annual energy consumption of air conditioning and lighting. The selection of metrics refers to the literature review and is adjusted according to the specific needs of office space.

In terms of daylight metrics, traditional static metrics, such as illuminance and the daylight factor, are widely used in indoor daylight evaluation under cloudy conditions. However, these metrics have limitations in capturing climate change, building orientation and annual performance characteristics [33]. In response, climate-based daylight metrics have been developed, including daylight autonomy (DA), spatial daylight autonomy (sDA), annual sunlight exposure (ASE), useful daylight illuminance (UDI) and related metrics [34,35,36]. DA indicates the percentage of hours occupied by indoor illumination exceeding the preset threshold but lacks control of upper illuminance limits. By incorporating it into the spatial coverage, sDA indicates the proportion of floor area that meets the objective illuminance threshold throughout the year, but there are still similar limitations. Hence, sDA is usually used in combination with ASE, which evaluates the risk of overexposure by calculating the proportion of time when the illumination exceeds the upper limit [9,19]. Compared with these metrics, UDI quantifies the percentage of time when a given point of illumination falls within the predefined useful range, thus reflecting daylight adequacy and overexposure control at the same time. In order to simplify the optimization process and maintain the independence of the metric, we use the average UDI of all sensor points as the daylight metric. The UDI range was originally proposed at 100–2000 lx [35] and later expanded to 300–3000 lx [37]. In this range, daylight can usually meet the needs of most visual tasks while reducing dependence on artificial lighting, which has contributed to its wide application in previous studies [38,39]. Accordingly, we choose 300–3000 lx as the UDI range to balance visual comfort and energy efficiency.

Since the discovery of intrinsically photosensitive retinal ganglion cells (ipRGCs), growing attention has been directed toward the non-visual effects of light, especially its role in circadian rhythm synchronization [40,41]. This additional non-visual photosensitive system extends the scope of light environment research from visual performance to broader human health and well-being considerations. It also highlights the limitations of the lux-based evaluation method in evaluating non-visual effects [42,43], thus accelerating the development of metrics related to circadian rhythm in architectural research. EML, m-EDI and CS are the most commonly used static metrics to evaluate the performance of circadian lighting [29,44,45]. EML weights photopic illuminance according to the spectral sensitivity of melanopsin-containing ipRGCs to represent relative circadian stimulus intensity [46]. m-EDI converts the melanopic-weighted response of a given light source into the equivalent illuminance under standard daylight (D65) producing the same melanopic effect [47]. CS derives from the circadian light response model proposed by Rea et al. and quantifies the biological effectiveness of light in suppressing nocturnal melatonin under specified conditions [48]. Based on these static metrics, Konis introduced the CF as a dynamic metric applicable to the annual assessment of the indoor building environment [28]. CF represents the percentage of days in which a given visual vector reaches or exceeds the specified EML threshold in the specified day period during the annual analysis cycle. In view of its computational efficiency and spatial integration representation ability, this study uses CF to evaluate the non-visual potential of space in different visual directions. Specifically, eight visual vectors are calculated at each sensor point (Figure 3), and the average CF values of all visual vectors in space are included in the optimization process. According to the WELL building standard (WELL v2), 275 EML is adopted as the threshold for effective circadian rhythm stimulation. This level is considered to be sufficient to support circadian rhythm regulation and daytime alertness [49]. Meanwhile, the Chinese standard GB/T 46119-2025 [50] specifies that, during daytime lighting periods, the vertical m-EDI at the seated eye level (1.2 m) should not be less than 250 lx to ensure circadian-stable lighting. Considering that, under near-daylight conditions, EML values are typically about 1.103 times higher than the corresponding m-EDI values [51,52], the threshold of EML adopted in this study provides a reasonable reference in the Chinese context. The effective day requires a given view vector to reach at least 275 EML for four consecutive hours between 9:00 and 16:00 on working days. The WELL spectral calculation tool determines EML from the spectral power distribution of light, computing α-opic equivalent illuminances for the five retinal photoreceptors and applying a calibration constant to maintain consistency with photopic lux under equal energy spectrum conditions (CIE standard illuminant E). Melanopic illuminance can be expressed as the product of photopic illuminance L and the conversion coefficient R (EML = L × R), where R reflects spectral characteristics. Regarding the impact of glass spectral transmittance characteristics on EML, Bellia et al. [53] found that clear glass does not significantly alter the spectral distribution of incoming daylight under either sunny or overcast conditions. Escobar et al. [54] also noted that high-transmittance glass can significantly increase indoor illuminance while largely maintaining the spectral characteristics of natural light. In other words, without the application of coatings or material modifications that selectively absorb specific spectral components, the spectral power distribution of light passing through glass typically does not change significantly. Since the clear and diffusing glass materials used in this study only affect the transmission and scattering distribution of light without substantially altering its spectral composition, their impact on the overall EML calculation results is expected to be relatively limited. On-site measurements in office environments in China and Italy show that CIE standard daylight D55 can reliably approximate the spectral power distribution of daylight under different sky conditions, with absolute errors generally below 10% in EML calculations [14,55,56]. Based on this, we adopted the R value of 0.998 under the CIE standard daylight D55, where L corresponds to the vertical illuminance at the level of the human eye.

For energy performance evaluation, energy use intensity (EUI) provides a normalized metric of overall building energy consumption. In building energy research, it is usually used to describe the annual operating needs of heating, cooling and artificial lighting, expressed in terms of total annual energy consumption divided by the total building area. This metric reflects the comprehensive impact of climatic conditions and internal thermal loads, so it is used as the energy optimization objective of this study.

2.3. Model Parameters and Simulation Settings

Based on field survey and measured data, a three-dimensional geometric model of the existing workshop was first built in Rhinoceros (Figure 4 and Figure 5). Then, we integrated the model into the Honeybee of the Rhinoceros-Grasshopper platform as a baseline model for daylight and energy consumption simulation. Daylight simulation is carried out using Honeybee’s built-in Radiance engine, which has been widely verified and is recognized as a standard tool in the field [34,57]. Energy consumption simulation is also carried out through the Honeybee plug-in based on OpenStudio 3.7.0/EnergyPlus 23.2.0. OpenStudio is a graphical modeling interface of EnergyPlus. EnergyPlus is widely used in the comprehensive energy consumption analysis of buildings, covering heating, cooling, lighting, ventilation and other aspects related to energy consumption.

The simulation parameters of the building envelope are shown in

Table 1. Envelope parameter settings for the existing building.

Attributes	Construction/Materials	U-Value (W/m2·K)	Visible Reflectance
Wall	390 mm brick	1.67	0.60
Floor	200 mm concrete + 50 mm cement mortar	3.10	0.30
Roof	15 mm tile + waterproofing membrane + 30 mm concrete screed + 30 mm timber deck	2.30	0.20
Door	metal surface + insulation layer	0.55	0.50
Window	existing glazing (SHGC: 0.54 Transmittance: 0.5)	5.15	0.20

. Since no door exists on the east façade and the original west door is damaged, a conventional insulated metal door was adopted as a replacement. The ground floor and the wall shared with the adjacent building were defined as adiabatic boundaries to isolate the influence of opening design. According to the recommended values of office space in GB55015-2021 [58], the annual personnel occupancy schedule, HVAC parameters, lighting power density and other relevant operating parameters are set to ensure compliance with regulatory requirements (see

Table 2. Simulation and optimization parameters.

Radiance Parameters	Ambient bounces: 6
	Ambient division: 1500
	Ambient sampling: 100
	Direct thresholding: 0.15
	Direct certainty: 0.75
Optimization Parameters	Elitism: 0.5
	Mutation probability: 0.2
	Mutation rate: 0.9
	Crossover rate: 0.8
	Population size: 300
	Maximum number of Generations: 100
Energy Simulation Parameters	Floor space per capita: 10 m2
	Lighting power density: 8 W/m2
	Equipment power density: 15 W/m2
	Ventilation: 30 m3/(h·person)
	Air-conditioning schedule (weekdays): 07:00–18:00
	Cooling setpoint: 26 °C
	Heating setpoint: 20 °C

). The climate input data adopts Zhengzhou’s typical meteorological year data compatible with EnergyPlus. Figure 4 shows the sensor layout for UDI and CF calculations. Under the same grid configuration, horizontal illuminance at 0.75 m above the floor was used to calculate UDI, while vertical eye-level illuminance at 1.2 m in different orientations was used for CF evaluation. In order to calculate the annual lighting energy consumption, a set of dimmable lighting system controlled by daylight sensors is used in the simulation, and the target illuminance is 300 lx. When the daylight is insufficient, the control system will automatically adjust the artificial lighting output to maintain the target illumination.

The optimization area is defined and parameterized in the baseline model. The north facade and roof are subdivided into regular grid units to systematically explore various opening configurations while avoiding facade columns and timber roof trusses. Each roof slope follows the actual internal roof size, forming two rows of ten grid units, each unit size is 3.5 m × 3.3 m. The two roof slopes contain a total of 40 grid units. The north facade follows the existing structural span to form ten units (3.5 m × 4.8 m). In total, 50 grid units across the roof and façade constituted the optimization area. Four variables control the geometry of each opening. First, generate a rectangle corresponding to the boundary of the grid cell, and use the horizontal and vertical scaling coefficients (Sx, Sy) to resize it. Then, the rectangle moves along the normal direction of the corresponding grid edge, and the displacement is determined by the offset coefficient (Ox, Oy), which is measured from the midpoint of the edge of the rectangle. This parametric method enables the flexible change in the opening geometry (Figure 6). The value range of Sx and Sy is 0 to 1, while the value range of Ox and Oy is −1 to 1, with a step length of 0.01. Different material type variables represent the types of glass that can be selected for skylights and side windows, respectively. There are three options for side windows: single-layer transparent glass, double-layer transparent glass and opaque (no opening). Considering the higher lighting efficiency of the skylight and the light diffusion that may be required, a frosted diffused glass option has also been added.

Table 3. Glazing Material Variables and Parameters.

Glazing	U-Value (W/m2·K)	Visible Transmittance	Solar Heat Gain Coefficient
Transparent Glass (single glazing)	5.15	0.90	0.85
Transparent Glass (double glazing)	2.59	0.81	0.75
Translucent Glass	3.30	0.36	0.39

summarizes the typical physical properties of each type of glass.

Multi-objective optimization relied on the Octopus plugin based on a genetic algorithm, which encoded the defined design variables as genetic information to evaluate UDI, CF, and EUI. Numerous recent studies have demonstrated the strong optimization capability of this plugin [9,38,59]. To explore a broader range of trade-offs among objectives, the fast multi-objective optimization method based on HypE was employed [11]. Multi-objective optimization does not yield a single global optimum. Instead, it generates a diverse set of solutions distributed along the Pareto front. Architects can select appropriate schemes according to their priorities. Extreme points on the front correspond to the best solutions for individual objectives, while compromise solutions arise through evaluation of the fitness function. This study adopted the fitness function proposed by Konis et al. [60] (see Equation (1)), and multiple multi-objective optimization studies have applied this method [38,61].

$$\begin{gathered} Y=\left({UDI}_i-{UDI}_{min}\right)C1+\left({CF}_i-{CF}_{min}\right)C2+\left(-1\right)\left({EUI}_i-{EUI}_{min}\right)C3 \\ C1={\displaystyle \frac{100}{{UDI}_{max}-{UDI}_{min}}} \\ C2={\displaystyle \frac{100}{{CF}_{max}-{CF}_{min}}} \\ C3={\displaystyle \frac{100}{{EUI}_{max}-{EUI}_{min}}} \end{gathered}$$

(1)

where i represents the result of a specific solution. “Min” and “max” denote the minimum and maximum values attainable within the solution set, while Y indicates the value of the fitness function. To prevent any single metric from dominating the overall evaluation during result aggregation, the UDI, CF, and EUI metrics undergo range normalization prior to integration, typically scaling their values to a unified interval from 0 to 100 [60]. Because EUI is defined as a metric to be minimized, its normalized value is multiplied by −1. The normalized metric values are then combined to derive the final optimization evaluation outcome.

2.4. On-Site Measurement and Validation

To verify the simulation model and evaluate the daylight performance of the workshop, on-site measurement was carried out. The indoor illuminance measurement took place on 17 February 2023, when the weather was cloudy to sunny. In order to reduce the impact of climate fluctuations, measurements were scheduled during relatively stable daylight periods at 11:30 and 12:30, and the repeated readings of each point are averaged to improve the reliability of the data. According to the requirements of GB/T5699-2017 [62], 30 reference points are evenly distributed on the horizontal surface (Figure 4). During the measurement, no artificial lighting was turned on in the workshop, windows and east entrance remained open. The illumination measurement uses TES-1339R digital illuminance meter manufactured by TES Electrical Electronic Corp., with an accuracy of ±3% ± 5 digits and a measurement range of 0.01–999900 lx. Figure 7 shows the comparison between the measured illuminance value and the simulated illuminance value. Except for P20, the two data sets show an overall consistent distribution pattern at the measurement point. The peak deviation at P20 is due to its proximity to the opening on the east side and exposure to fluctuating direct sunlight, which is the expected source of change in daylight simulation. It should be noted that the results still carry some uncertainty due to the limited number of measurement periods and weather conditions. First, measurements were taken only during a limited period on a single day, without covering the full daily cycle or different weather conditions. Second, the weather on the measurement day was partly cloudy, with sky conditions changing rapidly over short periods, especially near openings, where fluctuations in direct sunlight caused instantaneous deviations in local illuminance, thereby amplifying differences between measured and simulated values. However, because the original openings in the factory are small and the north-facing windows are exposed only to diffuse skylight, most areas (except near windows and doors) remain stable despite changes in direct sunlight. Accordingly, the influence of varying sky conditions, solar altitude, and incidence angle on most measurement points is relatively limited.

The model accuracy is evaluated by the relative mean bias error (MBEr) and relative root mean square error (RMSEr), which are often used in daylight simulation validation [63,64,65]. MBEr reflects the overall deviation of the model, while RMSEr shows the deviation between each measurement point more clearly. Equations (2) and (3) define the calculation methods of MBEr and RMSEr.

$${MBE}_r={\displaystyle \frac{1}{N}}{\sum }_{i=1}^N{\displaystyle \frac{x_{sim,i}-x_{mea,i}}{x_{mea,i}}}$$

(2)

$${RMSE}_r=\sqrt{{\displaystyle \frac{1}{N}}{\sum }_{i=1}^N{\left({\displaystyle \frac{x_{sim,i}-x_{mea,i}}{x_{mea,i}}}\right)}^2}$$

(3)

where ${\mathrm{x}}_{\mathrm{m}\mathrm{e}\mathrm{a},\mathrm{i}}$ and ${\mathrm{x}}_{\mathrm{s}\mathrm{i}\mathrm{m},\mathrm{i}}$ represent the measured illuminance and the simulated illuminance of the i-th measurement point, respectively, and N represents the total number of measurement points. In the two measurement periods, MBEr was −18.4% and −12.8%, while RMSEr was 29% and 31.6%, respectively. In the field of daylight simulation, there is still a lack of a clear model accuracy acceptability benchmark, which makes it difficult to establish a consistent error tolerance range. Nevertheless, previous studies generally believe that the relative error of 20%~35% is reasonable [38,65]. In this study, both MBEr and RMSEr are within the recommended range, indicating that the model has reliable overall accuracy. Further analysis found that most of the simulation results tend to be slightly overestimated. However, this deviation is still within the reasonable range and is conducive to more conservative design choices, such as reducing the window-wall ratio, window-to-ground ratio or using low-transmittance glass, which help reduce glare and unnecessary solar heat gain [38]. Since UDI and CF metrics derive from illuminance data, their reliability depends on the accuracy of illuminance calculation. Once the daylight simulation demonstrates sufficient reliability, the corresponding metrics can also provide reliable data about the overall trend, amplitude and spatial distribution. Regarding energy efficiency verification, the workshop does not operate in a closed air-conditioned environment but is directly connected to the outdoor environment. This situation makes it difficult to obtain the energy consumption and thermal environment data required under the operating conditions of HVAC to verify the simulation results, making it hard to directly verify the energy consumption model. Hence, the energy consumption value reported in this study is not absolute energy consumption, but a relative metric of energy consumption changes under different design solutions, which is used to support comparative evaluation and optimization decision-making.

3. Results and Discussion

3.1. Baseline Model Analysis

To compare performance before and after optimization, we simulated the condition of the existing workshop. Figure 8 shows the spatial distribution of UDI and CF under reference conditions. The results show that the daylight in the whole space is generally insufficient, and the average UDI is only 4.06%. Except for the area near the east entrance, the UDI value of most areas is extremely low. Even in the area near the opening, the improvement is minimal, and the year-round illumination of most measurement points is less than 300 lx. Similarly, the average CF value of all measurement points and orientations is only 5.14%, indicating insufficient circadian rhythm stimulation during occupied hours. The higher CF value only appears near the east entrance, which may be due to the increase in vertical illumination caused by low-angle sunlight in the morning. The main reason for the lack of overall daylight is that the window-to-floor ratio is only 0.04, and the north-facing openings rely primarily on diffuse daylight. Figure 9 shows the monthly energy consumption distribution of cooling, heating and lighting during occupied hours under the baseline model. The demand for cooling is mainly concentrated from May to September and reaches its peak in August, while the demand for heating in winter is still significant. This trend reflects the strong seasonal thermal contrast in Zhengzhou climate, and a balanced thermal strategy needs to be adopted. The energy consumption of lighting is relatively high throughout the year, which is consistent with the lower UDI and CF values, indicating considerable potential for improvement through enhanced daylight utilization. EUI reaches 131.86 kWh/m², of which 61.36 kWh/m² for cooling, 48.07 kWh/m² for heating, and 22.43 kWh/m² for lighting. The high demand for cooling may reflect not only the external climatic conditions, but also the continuous increase in internal heat during the operation of the office. In summer, the heat generated by personnel, equipment and lighting increases the cooling load, while in winter, it partially offsets the heating demand. In general, the baseline model has significant defects in daylight performance and circadian rhythm stimulation and has high energy consumption. Therefore, the focus of the subsequent optimization is to improve the daylight utilization by redesigning the opening system.

3.2. Pareto Front Analysis

After 50 generations of iterative evolution, a clear and continuous Pareto front can be observed from the Octopus interface shown in Figure 10, without evident clustering or loss of diversity. Meanwhile, the gray and black bands below the Pareto front represent the range and variation trends of the objective values during the optimization process, which can be used to assess convergence. The distribution of EUI remains relatively stable, with small fluctuations and an almost unchanged central tendency, indicating that it has approached a stable state. UDI shows some variation in the early stages but gradually stabilizes after approximately the 45th generation, suggesting that the improvement diminishes and convergence is approached over time. Although the gray band of CF exhibits a relatively wide range, indicating extensive exploration, the black band remains stable, implying that the overall trend has stabilized. Considering the stable shape of the Pareto front and the steady variation in the objective value ranges, it indicates that the optimization process tends toward convergence. Figure 11 shows a three-dimensional scatter diagram containing all iterations and illustrates the overall distribution of solutions in the Octopus optimization process. The blue dot represents all the generated solutions, and the red dot represents the non-dominant solution on the Pareto front. Each solution occupies a position in the three-dimensional objective space according to its objective function value, forming a continuous distribution pattern. As the iteration progresses, the Pareto solution gradually gathers in areas with higher UDI and CF and lower EUI, while the dominant solution tends to gather at the outer boundary of the objective space. Figure 12 shows three two-dimensional projections, which represent the relationship between the last generation of Pareto solutions and the objective function. The last generation contains 349 non-dominant solutions and defines the reachable performance range of each metric. The extreme values on the coordinate axis show that there are significant differences in the optimization results of different objective functions. The range of UDI is 35.11% to 87.86%, and the difference between the minimum and maximum values is large, indicating that the change in the opening configuration directly affects the indoor lighting performance. The range of EUI is 106.49 to 117.77 kWh/m², indicating that the thermal effect and lighting effect related to the opening make a significant contribution to the overall energy demand. In contrast, the range of CF is 91.89% to 99.93%, which is relatively small. This narrow range may be due to the lack of a strict upper limit, because increasing the opening area in the optimization process can easily improve the amount of lighting in each direction of sight.

UDI-EUI Pareto front shows that EUI increases with the increase in UDI, and in the same solution, the maximum UDI coincides with the maximum EUI. This pattern shows that improving daylight performance requires higher energy consumption. One explanation lies in the opening strategy adopted to control the illuminance level. The strong direct sunlight from the south makes the use of south-facing skylights more cautious, while the use of north-facing side windows and skylights is more frequent. This layout limits the absorption of solar radiation heat in winter. At the same time, the thermal transmittance of the existing roof is relatively high, close to double-layer transparent glass. Thus, the transparent south-facing skylight can increase the absorption of solar radiation heat gain in winter without significantly increasing the cooling demand in summer. When the design solution tends to be north-facing and reduces the south-facing opening, the UDI value will increase, but the annual energy consumption will also increase due to the reduction in solar radiation heat in winter. Therefore, there is a positive correlation with daylight performance and energy performance.

The EUI-CF Pareto front does not show a stable linear relationship between the two metrics. When the EUI value remains in a low range, such as below 110 kWh/m², the CF value is more concentrated. With the increase in EUI value, the CF value shows a clearer stratification and dispersion pattern within the high value range. The CF value mainly depends on the overall amount of daylight, and increasing the opening area usually leads to a higher CF value. Because the south-facing skylight can provide stronger sunlight penetration, high-CF solutions tend to include them more frequently. These skylights can also enhance solar radiation heat gain in winter, which usually means lower annual energy consumption. Therefore, low-EUI solutions can often achieve relatively high CF at the same time. As EUI increases, the proportion of south-facing skylights decreases, and the design relies more on alternative opening positions and construction combinations. Different configurations can still achieve comparable CF levels at similar energy intensities, thus forming the hierarchical distribution of CF observed along the Pareto front.

UDI-CF Pareto front shows that CF decreases with the increase in UDI, and in the same solution, the maximum UDI corresponds to the minimum CF. This pattern shows that improving daylight performance may limit the potential of non-visual aspects to a certain extent. However, the overall impact is still moderate, and most solutions still maintain a relatively high CF. This may be related to preventing excessive illumination caused by direct sunlight. In high UDI solutions, the design usually limits the area of the south-facing skylight to avoid exceeding the upper limit threshold of illumination. Reducing these openings will reduce the overall daylight exposure and slightly inhibit CF. Nevertheless, the diffuse daylight introduced through the opening in other directions can still ensure relatively uniform illumination, thus keeping the CF at a relatively high level.

3.3. Sensitivity Analysis and Characteristics of Design Variables

Sensitivity analysis is widely used in building performance studies to identify the influence of design variables on different performance metrics and to quantify their relative importance. In this study, the opening system consists of 50 window units, involving a total of 250 geometric and material variables. The numbering and naming convention of these variables are illustrated in Figure 13, which presents the unfolded façade with the indexing of the 50 window units. Each window unit includes five variables, distinguished by the suffixes Sx, Sy, Ox, Oy, and MT (material type), as introduced in Section 2.3. Due to the large number of variables, a global sensitivity analysis based on the Morris random one-at-a-time (OAT) method was adopted. This method is suitable for high-dimensional problems, offering high computational efficiency while enabling a global evaluation of parameter sensitivity. To ensure adequate coverage of the parameter space and improve the statistical robustness of the results, r = 10 trajectories were employed for k = 250 design variables. According to the Morris formulation, the total number of simulations is N = r × (k + 1) = 10 × (250 + 1) = 2510 [66]. Figure 14 presents the μ*–σ results, where μ* represents the importance of each parameter, and σ characterizes the combined effects of nonlinearity and interactions with other variables. Figure 15 shows the top 50 variables ranked by μ* for each performance metric. For EUI, most influential variables are concentrated around the south-facing skylights, where the window size parameters (Sx, Sy) and material types (MT) have a significantly greater impact than positional parameters (Ox, Oy). Moreover, a pronounced gap exists between the top-ranked variables and the remaining ones, indicating that energy performance is dominated by a small number of key parameters. For UDI, in addition to the continued importance of south-facing skylights, the influence of north-facing skylights becomes more prominent. In contrast to EUI, the differences in μ* values between higher- and lower-ranked variables are relatively small, suggesting that changes in most window units have a non-negligible effect on daylight performance. The distribution pattern of CF is similar to that of UDI, indicating that it is influenced by variations across multiple window units. For all performance metrics, the top-ranked variables are associated with relatively high σ values, indicating strong nonlinear and interaction effects, and therefore require coordinated optimization.

Analyzing the distribution of design variables within the Pareto-optimal solution set helps identify reasonable value ranges for high-performance designs, thereby providing guidance for decision-making. However, directly performing statistical analysis on all 250 variables leads to high dimensionality and limited interpretability. The sensitivity analysis results indicate that, compared to positional parameters of openings, window size, orientation, and material types have a greater impact on the simulation outcomes. Accordingly, the variables were aggregated based on window size, orientation, and material type, and group variables were constructed according to area ratios and glazing categories for subsequent analysis. Specifically, the opening area of a single window unit is divided into north windows, north skylights and south skylights, and further grouped according to glass types, including single-layer glass, double-glazing and diffused glass. All values are normalized by the area ratio relative to the area of the indoor floor. The process simplifies the original 250 variables into 11 grouped variables, including the area ratios of single glazing for north-side windows (NW-SG), double glazing for north-side windows (NW-DG), single glazing for north skylights (NS-SG), double glazing for north skylights (NS-DG), diffusive glazing for north skylights (NS-DF), single glazing for south skylights (SS-SG), double glazing for south skylights (SS-DG), diffusive glazing for south skylights (SS-DF), as well as the total area ratios of north-side windows (NW-AR), north skylights (NS-AR), and south skylights (SS-AR). Although these grouped variables cannot distinguish the geometry or position changes in individual windows, the combined distribution of area and glass composition can still reflect the overall design trend of the opening system and provide useful guidance for early parameter value screening.

Based on the last generation of Pareto optimal solutions, the parallel coordinate diagram visualizes the relationship between these group variables and the performance objectives UDI, CF and EUI, revealing the distribution of design variables under different combinations of daylighting and energy consumption performance (Figure 16). The curve color represents the change in EUI value. Red indicates a higher EUI value, and blue indicates a lower EUI value. The parallel coordinate distribution diagram shows that the window-to-floor ratio is not concentrated in the high-value area but generally maintains a low level in different types of openings. The south-facing skylight dominates all openings, and its total area ratio (SS-AR) is mainly between 0.20 and 0.35. Within this range, the area ratio of transparent glass (SS-SG) and double glass (SS-DG) is mostly between 0.05 and 0.15, while the area ratio of diffuse glass (SS-DF) is usually less than 0.05. Optimal solutions tend to enhance daylight efficiency by increasing transparent south-facing openings, while relying on double glazing to control glare, overheating, and energy consumption. In contrast, north-facing skylights remain more restricted. Their north skylights area ratio (NS-AR) is usually less than 0.10, and most glass ratios are less than 0.05. This shows that they mainly act as auxiliary lighting sources, helping to maintain uniform lighting while limiting additional heat loss. The north-side windows area ratio (NW-AR) seems to be slightly higher than the north skylights area ratio (NS-AR), usually between 0.02 and 0.12, and the distribution is more dispersed. Double glazing dominates in this category, which reflects the flexible use of lateral openings according to performance priorities in different solutions.

3.4. Analysis of Optimal Solutions

In view of the large number of solutions and different performance orientations, we have selected representative solutions for analysis, including extreme solutions and optimal compromise solutions, to reveal the design characteristics under different performance priorities. The extreme solution describes the boundary conditions of each objective and helps to identify key design variables and their influence directions. The optimal compromise solution corresponds to the highest fitness value (Y), representing the most balanced overall performance and stronger practicality. Figure 17 shows the open configuration of the extreme solution of each objective, as well as the top five solutions ranked by fitness value.

Table 4. Extreme solutions and selected optimal solutions.

Solution	Y	Parameters											Objectives
		NW-SG	NW-DG	NS-SG	NS-DG	NS-DF	SS-SG	SS-DG	SS-DF	NW-AR	NS-AR	SS-AR	UDI /%	EUI /(kwh/m2)	CF /%
Baseline model	-	0.04	-	-	-	-	-	-	-	0.04	-	-	4.06	131.86	5.14
Max UDI (Max EUI)	32.48	0.10	0.14	0.00	0.02	0.03	0.04	0.03	0.00	0.24	0.06	0.07	87.86	117.77	94.50
Max CF (Min UDI)	92.85	0.00	0.14	0.02	0.03	0.01	0.15	0.16	0.00	0.15	0.07	0.30	35.11	107.30	99.93
Min EUI	113.97	0.00	0.06	0.00	0.02	0.01	0.14	0.16	0.02	0.06	0.03	0.31	43.68	106.49	99.75
Min CF	11.48	0.03	0.08	0.01	0.01	0.02	0.03	0.05	0.00	0.11	0.04	0.08	86.78	116.25	91.89
Op-01	127.39	0.00	0.10	0.00	0.01	0.00	0.02	0.20	0.00	0.10	0.02	0.22	64.72	108.87	99.31
Op-02	126.17	0.00	0.07	0.01	0.00	0.00	0.01	0.21	0.00	0.08	0.01	0.23	63.19	108.64	99.29
Op-03	125.04	0.00	0.07	0.00	0.01	0.00	0.07	0.20	0.02	0.07	0.01	0.30	50.56	106.59	99.66
Op-04	124.89	0.02	0.05	0.00	0.03	0.00	0.01	0.24	0.00	0.07	0.03	0.25	56.67	107.68	99.50
Op-05	124.47	0.00	0.07	0.01	0.02	0.00	0.03	0.22	0.00	0.07	0.03	0.25	54.21	107.55	99.74

summarizes the corresponding Y values, design variables and performance metrics, including UDI, EUI and CF. Compared with the baseline model, the optimization yields significant improvements across all metrics. The UDI value of most optimization solutions is higher than 50%, which means that the indoor space maintains effective daylight conditions for more than half of the annual usage time, thus reducing dependence on artificial lighting. The CF value is usually more than 90%, indicating that sufficient and stable circadian rhythm stimulation can be maintained at most workstations and perspectives. At the same time, the EUI has been significantly reduced, indicating effective coordination between daylight utilization and building energy performance after introducing improved opening configurations. Take the best-performance solution (Op-01) as an example, UDI, CF and EUI increased from 4.06%, 5.14% and 131.86 kWh/m² to 64.72%, 99.31% and 108.87 kWh/m² of the baseline model, respectively.

Judging from the overall performance of the extreme solution, the solution achieving the maximum UDI (which also corresponds to the maximum EUI solution) maintains a UDI value above 75% at all measurement points. Despite this strong daylight performance, the overall energy consumption is still relatively high, which means that lighting energy savings from increased daylight cannot completely offset the new demand for HVAC. From the perspective of daylight uniformity, the ratio of the minimum UDI to the average UDI serves as the evaluation metric. A value of 0.88 indicates relatively consistent daylight levels across most areas, resulting in a well-balanced spatial distribution with few under-lit zones. The average CF value remains high overall but decreases noticeably near the south wall where view directions face the wall surface, with several view vectors falling below 50%. In terms of opening configuration, the solution tends to adopt a large area of transparent north-side windows, supplemented by limited north-side skylights and a small number of south-side skylights, so as to create a stable lighting environment with less fluctuations. Because the north-facing opening mainly introduces diffuse daylight rather than direct solar radiation, the thermal gain of passive solar energy is limited, which leads to an increase in annual energy consumption. North-facing skylights are mainly located at the roof ends and center, and the central zone typically uses diffuse glazing to spread top-incident daylight more evenly over a larger interior area. In contrast, the south-facing skylight forms a continuous strip shape, far from the ridge and close to the eaves. Roof shading from adjacent factory buildings likely reduces direct solar penetration in these areas and helps moderate daylight intensity. Openings near the ridge introduce stronger direct sunlight and therefore appear more cautiously controlled to avoid local over-illumination. The performance characteristics of the minimum CF solution are roughly similar to those of the maximum UDI solution. However, it further reduces the diffuse daylight from the north windows and the north skylight, while slightly increasing the south-facing opening near the ridge. Although the overall sunlight input is still sufficient, the stronger direction of south-facing direct sunlight reduces the vertical illumination of the line of sight facing the south wall, resulting in a significant decline in CF performance. The performance characteristics of the maximum CF solution (which also corresponding to the minimum UDI solution) are similar to those of the minimum EUI solution. The extensive use of large transparent south-facing skylights leads to favorable CF and EUI performance, while overall UDI values remain relatively low. The UDI uniformity values for the two solutions are 0.72 and 0.49, respectively, showing a significant difference. The minimum UDI solution results in consistently low UDI values across the overall space due to excessive daylighting. In contrast, the minimum EUI solution relies primarily on south-facing skylights for daylighting, leading to localized over-illumination and uneven daylight distribution.

Among the optimal trade-off solutions, Op-01 to Op-05 mainly adopt large south-facing double-glazed skylights as the main openings, supplemented by limited north skylights and north-side windows. This configuration achieves balanced performance across daylight, circadian rhythm, and energy objectives. The top-ranked trade-off solutions achieve UDI uniformity above 0.55, indicating relatively balanced annual daylight distribution despite localized over-illumination that calls for functional zoning or targeted shading in subsequent design. Take Op-01 as an example, its opening strategy shows obvious spatial differences. In the central area of the building, the south-facing double-glazed skylight dominates, forming two continuous daylighting bands, located near the ridge and eaves, respectively. The skylight in the central area can affect a wider range of indoor space while introducing more stable solar thermal energy. The openings layout on the east and west sides reflects the different daylight control needs. The east side pays more attention to morning lighting. At this time, daylight can more effectively promote the adjustment of circadian rhythm, and the indoor and outdoor temperature is relatively low, thus effectively controlling the cooling demand. Accordingly, the opening near the ridge increases moderately, while the opening near the eaves decreases, thus controlling the absorption of direct sunlight and solar heat. For areas with insufficient lighting, it is supplemented by limited north-facing skylights and high-level north-side windows. In the afternoon and evening, the needs of daylight and circadian rhythm are basically met, but the accumulated heat increases the risk of overheating and excessive light. Therefore, the proportion of openings near the ridge is correspondingly reduced, and the diffuse daylight from the north window plays a more important role in maintaining indoor lighting and limiting excessive solar heat absorption. In summary, the results show that subdividing façade and roof regions and assigning differentiated opening locations, area ratios, and glazing types can substantially improve the overall environmental performance of industrial heritage buildings.

Based on the above analysis, the ranges for key opening design variables are summarized as follows:

• South skylight area ratio: Recommended to be controlled within 0.20–0.30, as it enables a balanced performance among daylight availability (UDI), circadian potential (CF), and energy use (EUI);
• North skylight area ratio: Recommended to be limited to 0.00–0.10 to introduce stable diffuse daylight while minimizing heat loss;
• North side window area ratio: Suggested to range between 0.07–0.24, mainly serving to supplement daylight and improve illuminance distribution in deeper spaces, while excessive values should be avoided to prevent increased energy use;
• Material type: Double-glazed transparent glass is recommended as the primary option, balancing solar transmission and thermal insulation and supporting high CF and low EUI, while diffuse glass is used as a local supplement.
• Spatial layout strategy: For balanced performance, continuous south-facing skylight bands are recommended in the central zone; ridge openings can be increased on the east side to enhance morning daylight and circadian effects, while reduced on the west side to mitigate overheating and excessive illumination. North-side high windows and limited skylights serve to supplement daylight in deeper areas and stabilize the indoor environment.

In practical applications, the above strategies still need to be comprehensively evaluated in conjunction with architectural conservation principles. The adaptive reuse of industrial heritage typically requires a balance between preserving authenticity and improving environmental performance. To address this conflict, this study controls the intervention from two main aspects: First, opening interventions are restricted within the grid framework of the existing roof and façades, avoiding fundamental alterations to the overall volume, primary structure, and principal spatial configuration. This approach preserves the basic architectural character at the methodological level. Meanwhile, the research object is not a key protected building; instead, it follows the principle of “repair the old as old, and let the new be distinctly new” in the process of adaptive reuse. Second, the optimization results do not advocate unrestricted addition of openings. Rather, through Pareto front analysis and representative solution sets, the study reveals parameter ranges and combinational patterns under different performance objectives. This enables designers to make informed trade-offs between performance improvement and preservation of architectural character. It is worth noting that the more favorable solutions tend to rely on south-oriented skylights and localized differentiated configurations, thereby reducing visual interference with façade authenticity to a certain extent. For components and interfaces requiring strict conservation in similar buildings, the optimization results should still be further verified in relation to protection level, reversibility of intervention, and on-site conditions. In addition, the study by Yu et al. [67] indicates that optimization methods are better suited as tools for scheme selection. Their value lies not only in identifying optimal solutions, but also in revealing trade-offs among multiple objectives through post-evaluation and criteria-based filtering, thereby preserving necessary professional judgment in design decision-making. Therefore, in this study, the potential conflict between authenticity and performance is not resolved by a single algorithmic outcome. Instead, it is addressed by constraining the optimization scope, maintaining solution diversity, and leaving the final decision to be adjusted collaboratively by architects and conservation processes. This reflects the fundamental position that performance improvement should serve conservation goals, rather than replace them.

4. Conclusions

This study has developed a multi-objective parameter optimization approach for the improvement of environmental performance in the adaptive reuse of industrial heritage buildings. The approach divides the facade and roof of the building envelope into grid units, and defines the opening position, size and glass type in each grid unit as optimization variables. It aims to maximize daylight utilization and non-visual potential by introducing daylight elements, while reducing energy consumption. Taking a typical industrial heritage workshop in Zhengzhou City, China as a case study, the effect of the approach was verified.

The baseline model evaluation shows that under the office building conversion scenario, the natural lighting and circadian rhythm stimulation of the existing workshop are insufficient, resulting in its high dependence on artificial lighting and relatively high annual energy consumption. In the solution with the best performance, UDI increased from 4.06% to 64.72%, CF increased from 5.14% to 99.31%, and EUI decreased from 131.86 kWh/m² to 108.87 kWh/m². Analysis of the optimized solution set shows distinct performance trade-offs. Strategies that prioritize improving the UDI tend to expand the north-facing diffuse opening while limiting the south-facing direct sunlight. This improves the proportion of illuminance within the useful range and reduces overexposure risk but weakens winter solar heat gains and increases heat loss, leading to higher annual energy consumption. In contrast, solutions aimed at reducing EUI or improving CF rely more on south-facing transparent skylights, which strengthen circadian stimulus and winter solar radiation gains and partially offset heating demand, although they may introduce greater illuminance fluctuation and glare risk. Further analysis of the extreme and optimal trade-off solutions reveals consistent design patterns. In general, high-performance solutions tend to adopt south-facing skylights as the primary daylight source, while limiting the proportion of north-facing skylights and using north-side windows as supplementary openings to achieve a stable and balanced lighting environment. In terms of material configuration, most openings favor double-glazed transparent glass to balance daylight transmission and thermal performance, whereas diffuse glass and single glazing are mainly used as local supplements rather than primary options. For industrial buildings that rely on north-oriented daylighting, a stable balance among daylight performance, circadian potential, and energy efficiency can be achieved by adopting a south-dominant skylight strategy, complemented by north-facing skylights and side windows, combined with regionally differentiated opening layouts.

It should be noted that there are some limitations. Although incorporating UDI into the optimization process helps to limit excessive illumination, high-performance solutions affected by CF and EUI objectives still tend to have a large area of south-facing transparent openings, which may bring potential glare risk. Future research should integrate glare metrics (such as DGP or DGI) into the optimization framework to enhance the robustness of visual comfort. A daylight-responsive dimming lighting system with sensors was assumed for lighting energy calculations, while its practical implementation in industrial heritage buildings remains to be explored in future research. One limitation of this study is its focus on lighting and HVAC loads, without fully considering the influence of natural ventilation and equipment energy use on overall performance. Assuming adiabatic ground conditions may underestimate its thermal mass effect and overestimate summer room temperatures, thereby affecting the optimization of window size and shading design. Natural ventilation plays an important role during transitional seasons, potentially affecting thermal comfort and reducing cooling demand, but due to its dynamic nature and modeling limitations, it has not been included in the analysis. Meanwhile, variations in equipment energy use under different adaptive reuse functions may also alter the overall energy consumption structure. Future research could incorporate natural ventilation and equipment loads into a more comprehensive evaluation framework to enhance the completeness and applicability of the results. In addition, multi-objective genetic optimization requires a large amount of performance simulation, so the calculation cost is quite high. The proxy model based on machine learning can significantly accelerate the optimization speed and expand the searchable solution space. The circadian rhythm evaluation method adopted in this paper relies on the EML threshold, which is still an engineering approximation of physiological reactions. Future research can improve the evaluation method through on-site measurements and occupant-based variation models. Despite these limitations, the proposed method still helps to understand and select the opening design strategy of industrial heritage buildings. The optimization results are visually clear and easy to understand and can be directly applied to the early design stage. It also shows the potential of promotion and application across similar single-story building types and climatic conditions.