Adaptive daylight-linked lighting control system: Integrating spatial daylight factor and real-time sensing

Abstract

Indoor lighting system is one of the significant contributors of building’s energy consumption. This study aims to improve indoor lighting system efficiency by developing an adaptive lighting control system that integrates the spatial dimensions of DF and real-time outdoor illuminance sensing based on soft sensors in two educational rooms in Yogyakarta, Indonesia. The research developed the system using in situ measurements, developed predictive and control algorithms, integrated them into a building energy management system, and validated its performance. The soft sensor algorithm demonstrated good predictive performance with MAE and RMSE distributions of $8.09 \pm 5.69$ lux and $10.76 \pm 7.70$ lux at the first room, respectively, and $8.62 \pm 6.54$ lux and $11.85 \pm 8.95$ lux in the second room, respectively. The constructed DF-based soft sensor performs accurately in real-world implementations with MAPE values of $11 % \pm 8 %$ and $14 % \pm 6 %$ for the first and second rooms, respectively. The tests also showed data acquisition, communication, and algorithm computation duration of $57.9 \pm 0.4$ ms, $22.7 \pm 0.8$ ms, and $6.1 \pm 1.2$ s, respectively, with no data loss during the transmissions. This study successfully developed and implemented an adaptive lighting control system based on a DF-based soft sensor that is reliable, scalable, and applicable for both existing and new buildings. However, the system should be treated as an operational estimation framework for maintaining illuminance within a certain performance range, rather than as an absolute point-estimation model with high precision at every location.

Practical application

The proposed DF-based soft sensor system offers a practical and scalable solution for adaptive indoor lighting control with a cost-effective yet accurate, reliable, and applicable approach. The system was developed using a calibrated outdoor illuminance sensor, an embedded microcontroller, and integration with a Building Energy Management System (BEMS), and thus allows real-time prediction of indoor illuminance without the need for costly hardware sensor networks. This approach reduces installation complexity and operational costs and is applicable to both existing and new buildings, is flexible, and is scalable for other types of buildings, occupant and activity patterns, and climates and regions.

Keywords

daylight factor illuminance monitoring soft sensor indoor environmental quality building energy management system

Introduction

The International Energy Agency (IEA) report shows that greenhouse gas (GHG) emissions from the energy sector reached a record high of 37.4 GtCO₂e in 2023.^1,2 The building sector, in particular, contributes around 37% of global energy-related GHG emissions, where approximately 27% of this number comes from building operations, such as cooling, heating, and lighting systems.³ Building energy consumption itself has been increasing significantly in recent decades, in line with population growth, the intensification of indoor activities, and rising demands on Indoor Environmental Quality (IEQ). Modern buildings serve functions beyond shelter but also as work, study, and activity spaces, increasing the demand of various supporting systems such as lighting, air conditioning, and electronic equipment.⁴ Among these systems, lighting remains a major contributor to energy consumption, accounting for approximately 15–20% of a building’s total energy use.^5–7 Optimizing the lighting system thus plays a crucial role in improving building energy efficiency and reducing overall energy consumption.

Efficiency in lighting systems has been an important focus in sustainable buildings. In the context of Indonesia, green building certification has adopted lighting system efficiency as a key indicator. The Indonesian Ministry of Public Works (PUPR) has launched an official green building rating system (BGH) through the Ministerial Regulation No. 21: 2021, where daylighting becomes a key parameter in its assessment.⁸ The Green Building Council Indonesia (GBCI) also includes this parameter in the Greenship rating system.⁹

Daylighting strategies typically involve dividing lighting systems into groups. Areas closest to the space fenestration are typically controlled based on readings from a light intensity sensor (luxmeter) installed at a representative point in the area. However, the remaining two-thirds of the area is thus illuminated via artificial lighting, regardless of any fluctuations in natural light that may reach those areas. Mangkuto¹⁰ demonstrated that daylight distribution in Indonesia may cover areas larger than one-third of a room, thus single-sensor approach may pose a reliability issue. Since spatially uniform control strategies do not account for local variations in daylight availability across the space, such strategies may pose inefficiency towards lighting operation. To further improve energy efficiency, adaptive lighting control systems were developed by integrating occupancy information and natural light contributions. These systems have been shown to reduce energy consumption by 61–91% in open office spaces.^11,12

The recent application of Artificial Intelligence (AI) and Internet of Things (IoT) further advances the potential for energy-efficient lighting systems. Seyedolhosseini¹³ proposed an Artificial Neural Network (ANN)-based lighting control system that actively adjusts the dimming level of each luminaire based on the variation of daylight supply at different task planes. This system quickly and efficiently responds to changes in natural lighting and saves up to 37% energy with a relatively lower illuminance deviation (<10%) while reducing the number of sensors required per zone. Machine learning and IoT-based approaches expanded this system by adding real-time connectivity and adaptive decision-making capabilities to balance energy consumption and visual comfort.^14–16 While those systems have significantly improved the temporal efficiency and intelligence of control systems, they lack consideration of the spatial dimension. Current advancements predominantly address temporal adaptation. However, spatial representation remains insufficiently resolved, as these systems are capable of adapting to changes over time (day–night cycle) and to weather conditions, yet they do not account for variations in illuminance between zones due to fenestration position, room geometry, and surface reflectance.

Spatial representation of daylighting profiles requires detailed measurements at multiple points within a room, thus requiring a higher number of sensors.¹⁷ Such an approach brings in complexity, as more sensors mean more calibrations and potential disruptions to occupant activities. Thus, researchers come up with ideas to simplify the system without sacrificing spatial accuracy. Feyzi and Mojallali¹⁸ proposed a sensor position optimization method using Battle Royale Optimization (BRO) algorithm to balance energy efficiency and visual comfort, while Jo and Jang¹⁹ developed a linear regression-based illuminance model that connects two reference sensors to other points within the room. While both approaches improve the efficiency of daylight measurement and reduce the number of sensors, they still rely on specific sensor locations and cannot depict the spatial distribution of daylight in real time. These approaches remain dependent on predefined sensor locations and are unable to comprehensively reconstruct real-time spatial daylight distribution.

On the other hand, Van De Meugheuvel et al.²⁰ and Peruffo et al.²¹ proposed the concept of decentralizing lighting control functions by placing illuminance and occupancy sensors at each luminaire, thus allowing each unit to adjust luminance flux locally. However, this system can only represent a small portion of the area under the luminaire and does not account for spatial variations in daylight across different zones. As a result, the system is unable to assess the differences in illuminance between areas around windows, the center, and the back of the room. Lee et al.²² utilized the Relative Illuminance Influence Level (RIIL) metric to demonstrate significant potential in connecting the illuminance contribution of each luminaire to the actual room conditions. Although effective for localized control, these approaches lack a comprehensive mapping of the spatial distribution of daylight and fail to capture the overall spatial variability of daylight within a space.

Gentile et al.¹⁷ proposed a combination of finite measurements and numerical simulation to obtain a larger daylight introduction while reducing sensor load. Although this approach simplifies installation complexity, it fails to capture real-time daylight fluctuations due to assumptions on static daylight distribution and instantaneous measurements. Acosta et al.²³ proposed a metric called Continuous Overcast Daylight Autonomy (Dao.con) that predicted the spatial representation of daylight using simulated overcast sky and historical meteorological data. This metric does improve daylight predictions, but still has poor daylight predictions due to fluctuations from the weather, environmental reflections, or user interventions.

The literature described above demonstrates that previous research has not successfully combined the spatial and temporal dimensions of daylight in a single, responsive, real-time adaptive control system. They remain limited in representing the real-time spatial variability of daylight under dynamic environmental conditions, since they rely either on limited sensor configurations or on predictive models based on simplified assumptions. Existing systems only provide average estimates or predictive projections, but fail to represent actual daylight distribution in a space at any given time.

This study addresses the current limitations by integrating a Daylight Factor (DF)-based soft sensor with real-time outdoor illuminance measurements to capture both spatial and temporal variations of daylight. This study integrates soft sensors into an adaptive lighting control system to better combine the spatial and temporal dimensions of daylight. The use of soft sensors based on mathematical models allows for real-time prediction of indoor environmental conditions at a lower cost and with higher reliability.²⁴ The soft sensing approach has been widely used in the building context to reconstruct the distribution of indoor environmental parameters, such as airflow, based on a combination of outdoor sensor data and predictive models.²⁵ The principles lie in the comparable phenomena of indoor airflow and daylight distribution that are simultaneously shaped by the building’s external boundary conditions and the spatial characteristics of the indoor spaces.

Indoor lighting conditions are significantly influenced by variations in outdoor illumination, which can be measured continuously using outdoor illuminance sensors. Based on this principle, this study proposes a Daylight Factor (DF)-based adaptive lighting control model combined with outdoor illuminance sensors as real-time input. This approach transforms static daylight distribution into a dynamic parameter capable of estimating daylight illuminance at each point in a space based on changes in actual outdoor illumination conditions. Thus, the daylight serves as a soft sensor model that utilizes outdoor illuminance data to represent the spatial distribution of daylight in a space in real time, enabling the system to construct a daylight distribution map without the need for additional sensors within the space.

As a control function, outputs from the soft sensor are integrated into a Building Energy Management System (BEMS) via an intelligent switch, forming an adaptive daylight-linked lighting control system. The predicted daylight distribution is used to automatically adjust artificial lighting settings according to the actual daylight conditions in the space. This approach enables a lighting control system that adapts to daylight dynamics throughout the day by using information about the spatial distribution of daylight, maximizing the energy-efficiency potential of a building’s lighting system without relying on numerous physical sensors. It also bridges the gap between spatial daylight modelling and real-time adaptive lighting control in practical building applications. While the proposed system is intended to support energy-efficient lighting operation, this study focuses on the development and validation of the control framework and does not directly quantify energy savings.

This study further develops and validates a prototype of an adaptive lighting control system based on daylight-linked control that integrates the spatial dimensions of DF and real-time outdoor illuminance sensing through soft sensors in a BEMS. The focus of this study includes the design of a DF-based soft sensor model, its integration into the lighting control system architecture, validation of its results, and the evaluation of the system’s performance in representing the distribution of natural lighting in a room as a basis for lighting control decision making.

Methodology

Research workflow

This study consists of three main stages, as shown in Figure 1: (1) field observations to obtain data regarding space characteristics and measurements of outdoor and indoor illuminance; (2) development of predictive and control algorithms; and (3) development of an integrated intelligent lighting system. The lighting system developed in this study comprises two main components: a soft-sensor-based natural lighting distribution monitoring system and an intelligent-switch-based adaptive lighting control system. Both systems are designed to operate in real time to support building energy efficiency by optimizing daylight utilization and are integrated into the Building Energy Management System (BEMS) platform as part of a data-driven energy management strategy.

Figure 1.

Research workflow.

Figure 2 shows the schematic diagram of the system developed in this study. The system consists of two parts: the monitoring and control of indoor illuminance. The indoor illuminance at a point $E_{i n} (x, y)$ consists of natural lighting from daylight $E_{n} (x, y)$ and artificial lighting $E_{l} (x, y)$ , which can be expressed as:

E_{i n} (x, y) = E_{n} (x, y) + E_{l} (x, y)

(1)

Figure 2.

Schematic diagram of the system.

Daylight factor ( $D F$ ) is defined as the ratio between the illuminance at a reference point indoors, $E_{P}$ , and the unobstructed outdoor horizontal illuminance, $E_{o}$ , under uniform overcast sky conditions, see equation (2)²⁶ This definition has long been used as the basis for evaluating daylighting in buildings since it provides stable and reproducible sky conditions for comparative analysis across building designs.²⁷

D F (%) = \frac{E_{P}}{E_{o}} x 100 %

(2)

Daylight Factor is a static parameter that describes the potential distribution of daylight within a space for a given building design, which is shaped by the space geometry, the size and position of openings, and the reflectance properties of interior surfaces.^27,28 As a static metric, daylight factor does not account for temporal variations in daylight due to changes in time, season, or weather conditions; therefore, for a given building configuration, its value can be considered relatively constant as long as there are no changes in building geometry or materials.^29,30 Thus, $D F$ represents the intrinsic capacity of a building to introduce natural light from outdoors to indoors and is widely used as an initial indicator in evaluating daylighting performance at the early design stage of buildings.³¹

Soft sensor technology using the DF model is utilized to monitor the distribution of indoor illuminance by making use of the measured outdoor illuminance detected by sensors and the on-off status of the light switches, as expressed:

E_{i n} (x, y) = D F (x, y) . E_{o u t} + \sum_{i = 1}^{n} S_{i} {. E}_{l i} (x, y)

(3)

where,

$E_{o u t}$ : outdoor illuminance of natural lighting.

$D F (x, y)$ : daylight factor at position $(x, y)$ .

$S_{i}$ : on-off status of the lighting switch $i$ , 1 for On and 0 for Off.

$E_{l i} (x, y)$ : indoor artificial lighting illuminance at position $(x, y)$ by lighting group $i$ .

The algorithm for indoor illuminance control is constructed by comparing the real-time illuminance value predicted by the soft sensor against a given set point. The intelligent switch will manipulate the lighting in the space according to the controller output to make sure it remains at the desired condition.

Case study

This study involved two different rooms in a single building, i.e., the Sensor and Telecontrol Systems Laboratory (SSTK) and the TN-7 Classroom (TN-7). Both rooms are located on the second floor of the Department of Nuclear Engineering and Engineering Physics (DTNTF), Faculty of Engineering, Universitas Gadjah Mada (UGM), see Figure 3(a).

Figure 3.

Layout of the 2nd floor at the DTNTF UGM building for (a) overall floor; and details of the (b) SSTK laboratory; and (c) TN-7 classroom showing their layout, elevation, and furniture arrangements.

Figure 3(b) and (c) display interior layouts and elevation of the studied rooms indicating the positions of windows, desks, and luminaires. The SSTK laboratory has an area of 206 m² and is located on the southern side of the building, with natural light entering through glass windows on the south-facing wall. Meanwhile, the TN-7 classroom is located on the eastern side of the building, with an area of 129 m² and is equipped with window openings on the east-facing wall, allowing natural light to enter the room. The control strategy is designed to maximize the use of natural daylight during building operation, and no shading devices (e.g., blinds or curtains) were used. Both spaces are equipped with 20 W LED artificial lighting systems, installed uniformly on the ceiling in a grid configuration adapted to the space’s function and area. The artificial lighting system uses Philips Ecofit LED Tubes E8 luminaires with a power rating of 16 W, a colour temperature of 6500 K, and a luminous flux of 1600 lumens. The luminaire has a wide beam angle distribution that suits general lighting requirements in educational and laboratory spaces.³²

The artificial lighting systems in both spaces were designed using a luminaire grouping scheme into several control groups to support operational flexibility and efficiency. All luminaires in each space were grouped into several lighting zones, each connected to an independent switch. The luminaires in the SSTK were grouped into eight lighting groups, and those in TN-7 into three (see Figure 3). This configuration enabled stepwise lighting control in accordance with activity requirements within the space as well as natural lighting conditions.

Development of DF-based soft sensor algorithm

The soft sensor algorithm used in this study was constructed based on equation (3) using measurement data of $D F (x, y)$ and $E_{l i} (x, y)$ at all measurement points $(x, y)$ . The distribution of indoor illuminance was observed through a series of illuminance measurements at multiple points distributed across the horizontal work plane at 75 cm height above the floor surface.^33,34 The positions of the measurement points in the $(x, y)$ coordinate is shown in Figure 4.

Figure 4.

Measurement points in (a) SSTK; and (b) TN-7.

Natural and artificial lighting simultaneously shape indoor lighting characteristics. Therefore, measurements were taken for both outdoor illuminance $E_{o u t}$ and indoor illuminance $E_{n} (x, y)$ , at every work plane $(x, y)$ as illustrated in Figure 4. Measurements were taken at the same time between indoor and outdoor to retain comparable results. Ideally, DF measurements are performed under uniform overcast sky conditions.^26,35 However, in real-world practical field measurements, variations in the sky conditions could not be avoided and may influence the result fluctuations. To suppress the influence of such variations, measurements were taken 10 times at different times for each measurement point. The DF value was then calculated as the average of the measurement results.^27,36

On the other hand, artificial lighting characteristics are shaped by its luminance distribution. To isolate the artificial lighting profiles without being influenced by natural light, measurements were conducted at night by switching the lighting groups on sequentially, thus obtaining data on the artificial lighting contribution from a lighting group $i$ at measurement points $(x, y)$ , $E_{l i} (x, y)$ .

Furthermore, validation was performed to assess the lighting prediction results generated by the soft sensor algorithm. The validation was conducted by comparing the results with actual indoor illuminance measurement results were evaluated using two parameters: Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). MAE represents the average error physically experienced by the system, while RMSE represents the intensity of fluctuations or the dominant error energy in the system, particularly due to large errors.^37,38

Development of control algorithms

Figure 5(a) demonstrates the decision-making process of the lighting control algorithm to directly respond to the predicted illuminance values generated by the soft sensor. Each lighting group is associated with a specific lighting zone that includes several measurement points beneath it (Figure 5(b)). For each zone, the illuminance value of the zone $i$ , $L_{i}$ , is defined as the average predicted illuminance from $n$ measurement points located within that zone. Based on the comparison between $L_{i}$ and the pre-determined set point, the system adaptively determines the status of the corresponding lighting group (on or off), thereby enabling zonal lighting control that is aligned with the actual spatial distribution of indoor illuminance.

Figure 5.

Flowchart of the control algorithm for (a) every lighting zone; and (b) in SSTK.

System development

The soft sensor and control algorithms developed in the previous stages were embedded into an actual real-time lighting monitoring and control system and were integrated into the existing BEMS. The development includes the design and implementation of an outdoor illuminance monitoring system, the construction of smart switches as lighting actuation devices, and the integration of all system components into the BEMS platform.

The DF-based indoor lighting control system requires real-time outdoor illuminance data as the primary input to the soft sensor algorithm that predicts indoor lighting (see equation (3)). Thus, an Outdoor Illuminance Monitoring System was developed to fetch real-time data consisting of a lux sensor integrated with a Raspberry Pi microcontroller and was installed at a representative, unshaded rooftop location, enabling accurate capture of natural lighting fluctuations. The measurement data were transmitted via a wired Local Area Network (LAN) using the Modbus communication protocol and stored in a server database as part of the BEMS platform.

The smart switch feature was developed as an actuator device for lighting control and monitoring the room’s operational status and was integrated into the BEMS. The operational communication between the smart switch and the BEMS was directed via the MQTT protocol, thus allows bidirectional exchange of data and control commands. Subsequently, all subsystems, including the outdoor illuminance monitoring system with the DF-based soft sensor prediction algorithm and the smart switch with the lighting control algorithm, were integrated into the BEMS platform. This platform functions as a centralized system for monitoring lighting conditions and making zonal lighting control decisions based on real-time data.

System validation

System testing was performed to assess the performance of the developed system, covering both monitoring and control aspects. The tests focused on accuracy, real-time characteristics, and system reliability, and were performed under actual operational conditions. Measurement accuracy was evaluated using Mean Absolute Percentage Error (MAPE), by comparing system measurement results with reference data.³⁹ Real-time characteristics were evaluated by measuring the average system latency, which represents the time delay from the sensor data acquisition process until the data are successfully received and stored in the BEMS database. System reliability was evaluated by comparing the number of data successfully stored in the database with the number of data acquired by the sensors during the testing period.

Testing on the control system focused on evaluating the performance of the smart switch as a lighting actuator within the BEMS. This testing included both functional and operational tests to ensure the reliability of control responses and status monitoring through the BEMS. The functional testing included validation of the ON/OFF control function and load status feedback mechanism, testing of system configuration via WiFi and BLE connectivity, as well as testing of the power supply/battery indicator. The operational testing included evaluation of response delay, smart switch current consumption, and system operating time. The operating time test was conducted by considering relay conditions in both connected and disconnected states to assess their influence on the smart switch’s duration in standby and active modes, thereby enabling a comprehensive evaluation of system reliability for long-term operation.

Result and discussion

Indoor daylight characterisation

Figure 6 shows the mean DF distributions at SSTK and TN-7, revealing a differences in both magnitude and spatial uniformity. The SSTK room is dominated by low DF values at most measurement points, ranging approximately 0.03–0.06%, with only a few points exhibiting higher DF values up to 0.176%. In contrast, the TN-7 room exhibits relatively higher and more evenly distributed DF values, with most points falling within the range of 0.04–0.07% and a similar maximum value of approximately 0.176%. This difference indicates that the TN-7 room has greater potential for utilizing daylight compared to the SSTK room.

Figure 6.

Spatial distribution of average DF at (a) SSTK; and (b) TN-7.

Differences in DF distribution are strongly correlated with the room layout and interior configurations (Figure 3). In SSTK, storage cabinets and other interior elements aligned parallel to the window plane act as visual obstructions that block daylight penetration indoors. In contrast, the more open layout of TN-7, with minimal obstructions around the openings, allows daylight to spread more evenly across the work area. These findings confirm that daylighting performance is not determined solely by window position and size, but is also influenced by interior layout. Therefore, interior configuration should be considered in DF modeling and in the design of adaptive lighting control systems based on building energy management systems (BEMS).

Fluctuations in DF values at each measurement point are caused by variations in the sky conditions during data acquisition, where the sky might be uniform overcast, partly cloudy, and clear conditions.^26,35 These variations lead to changes in the intensity and direction of incoming daylight, thereby affecting daylight illuminance at points that receive direct sunlight from outside the building, typically near windows. Figure 7 displays the phenomena, where wider DF ranges in repeated measurement boxplots reflect higher sensitivity to changes in outdoor lighting conditions.

Figure 7.

Distribution of DF measurement results at each measurement point in (a) SSTK; and (b) TN7.

In SSTK (Figure 7(a)), the highest DF fluctuations are identified at points (1, 3) and (1, 4). These points are located near the windows and are unobstructed by interior elements, making them highly responsive to changes in daylight intensity and incident angle. In contrast, points such as (2, 4), (3, 4), and (4, 4), although relatively close to the windows, exhibit lower fluctuations due to the presence of interior elements that obstruct direct sunlight. Meanwhile, point (5, 4) shows relatively low DF fluctuation compared to other points in the same row. This is due to its position being obstructed by the eastern wall, as shown in the room layout in Figure 3. As a result, the daylight received at this point is dominated by reflected and diffused light, leading to smaller DF variations despite changing sky conditions.

In TN-7, the greatest DF fluctuations occur at points (3, 4), (4, 4), (6, 4), and (7, 4), which are located near the east-facing windows and are not obstructed by interior elements. The east-facing window orientation causes these points to receive direct sunlight that is highly sensitive to changes in sky conditions, resulting in wider DF fluctuation ranges compared to SSTK. Point (5, 4), although close to the window, exhibits lower DF fluctuation due to the presence of surrounding interior walls that obstruct direct sunlight entry and cause the lighting to be dominated by diffused light.

Indoor artificial lighting characterisation

Figure 8 displays the illuminance distribution measured when all artificial lighting groups were switched on. The SSTK room has an average illuminance of 340.35 ± 35.06 lux, with smaller spatial variation across measurement points. In contrast, TN-7 shows a lower average illuminance of 288.97 ± 80.74 lux, with a substantially larger standard deviation, indicated by low uniformity and widely distributed illuminance between 120 lux and 450 lux.

Figure 8.

Distribution of artificial lighting illuminance in (a) SSTK; and (b) TN-7.

The differences in illuminance variation are caused by a combination of geometric factors and operational characteristics of the artificial lighting system. Beside the influence of the distance between measurement points and luminaires and the lighting zoning configuration, differences in luminaire age and performance degradation contribute significantly to nonuniform light output, even when all lighting groups are operated simultaneously.

The distribution of artificial lighting illuminance by each lighting group in the SSTK Lab (Figure 9) and TN-7 (Figure 10) shows zonal contribution patterns, consistent with the lighting group divisions in the room layout. In SSTK, each lighting group achieves the highest illuminance in the area surrounding the luminaires within its respective zone, with a sharp decline beyond zone boundaries, resulting in a relatively controlled inter-group distribution contribution. In contrast, in TN-7, inter-group illuminance contributions show greater variations, caused by differences in distance and zoning configuration, differences in luminaire age, and performance degradation, which lead to variations in luminous flux among groups even under identical operating conditions. This effect is further amplified by the location of several measurement points on the north and south sides, which are relatively farther from the main luminaires, resulting in significantly lower illuminance and increased spatial variation compared to SSTK.

Figure 9.

Illuminance distribution of each artificial lighting group in SSTK.

Figure 10.

Illuminance distribution of each artificial lighting group in TN-7.

Performance of the soft sensor algorithm

Figure 11 demonstrates the MAE and RMSE distribution maps of the DF-based soft sensor algorithm in SSTK and TN-7. The map demonstrates good predictive performance at most measurement points in both rooms. In the SSTK room, the MAE and RMSE distributions are $8.09 \pm 5.69$ lux and $10.76 \pm 7.70$ lux, respectively. These values are relatively homogeneous in the central area of the room, with MAE predominantly in the range of 4–8 lux and RMSE below 15 lux. A significant increase in error appears only locally at the edge of the room, particularly at coordinate (1, 4), where the MAE reaches 30.42 lux and the RMSE reaches 40.84 lux. The clear contrast between the high values at this point and the low values in the surrounding area indicates the presence of large but localized deviations.

Figure 11.

Distribution of MAE and RMSE of the soft sensor algorithm in (a) SSTK; and (b) TN-7.

In TN-7, the MAE and RMSE distributions are $8.62 \pm 6.54$ lux and $11.85 \pm 8.95$ lux, respectively. The error maps for TN-7 show stronger spatial nonuniformity, particularly along row $y = 4$ , where several consecutive points have high MAE values of up to 29.84 lux and a maximum RMSE reaching 40.67 lux, thus indicating a higher prediction sensitivity to specific spatial conditions than SSTK. Nevertheless, other areas in TN-7 continue to show low and homogeneous MAE and RMSE values, confirming that the increase in error is zonal in nature and does not represent a failure of the algorithm as a whole.

Most measurement points in both SSTK and TN-7 have MAE values below 10 lux, indicating that the average prediction errors are relatively small and consistent with the actual measurement results. The error patterns are spatial rather than random and are closely related to the spatial position of the measurement points within the room. These MAE values are far below the working illuminance levels recommended by SNI 6197:2021 and IES guidelines (350–500 lux),^35,40 such that a 10 lux deviation represents only 1–3% and is not significant for visual comfort. These findings are consistent with previous studies emphasizing the dominant role of lighting distribution and uniformity over small absolute illuminance variations.^41–43

The observed spatially structured error can be interpreted in relation to variations in the characteristics of Daylight Factor (DF) distributions discussed in the previous subsection. Variations in sky conditions alter the intensity and direction of incoming daylight, directly causing fluctuations in DF values at specific measurement points. These fluctuations directly contribute to variations in prediction error in the DF-based soft sensor algorithm, particularly at locations with high daylight sensitivity. This effect is particularly evident at measurement points located near window openings, where direct solar radiation increases prediction errors due to its exclusion from the Daylight Factor (DF) formulation. As a static parameter, DF has a known limitation of not explicitly accounting for dynamic sky conditions such as direct sunlight and sky luminance distribution, especially in tropical climates with frequent clear or partly cloudy skies.^44,45

The influence of these variations on prediction accuracy is illustrated in Figure 12 through a comparison between the measured and predicted distributions of natural lighting. In the best-prediction scenario, the DF-based soft sensor algorithm consistently captures the pattern of natural lighting distribution, with spatial gradients and the locations of relatively high and low illuminance values in the prediction maps aligning with the actual measurements. In the worst-case scenario, prediction deviations become larger, particularly at measurement points with high MAE and RMSE values. In the worst-case scenario, larger deviations are observed near window openings under direct solar exposure. Despite these localized errors, the predicted illuminance distribution still preserves the overall spatial pattern, and the resulting performance remains acceptable for practical lighting control applications.

Figure 12.

Comparison of measured and predicted natural lighting distributions under best- and worst-case scenarios for (a) SSTK and (b) TN-7.

System development

The outdoor illuminance monitoring system was successfully implemented to provide real-time $E_{o u t}$ data as the primary input for the DF-based soft sensor algorithm. A WN90LP illuminance sensor was installed with the following technical specifications: a measurement range of 0–200,000 lux, 100 lux resolution, and a manufacturer-stated accuracy of ±25% (Figure 13). The measurement range enables the sensor to capture variations in natural illuminance under both low-intensity overcast sky conditions and maximum solar radiation typical of tropical climates. The sensor was integrated with a microcontroller-based data acquisition unit and mounted at a representative location on the building rooftop, free from shading by structural elements or vegetation.

Figure 13.

The WN90LP sensor installed on the rooftop.

To ensure the sensor’s accuracy across the illuminance range, calibration was performed by comparing the sensor’s reading with a Luxtron LX-1128SD luxmeter, which has higher accuracy ±4.00%. Figure 14 displays the comparison results between the WN90LP and LX-1128SD, where the regression results indicate a very strong linear correlation with the reference luxmeter ( $R^{2} = 0.99$ ), despite the WN90LP’s lower nominal accuracy. A systematic bias was observed (slope 0.892 and offset 2626 lux), leading to overestimation at high illuminance levels. However, the deviation can be corrected through linear calibration.

Figure 14.

Correlation of measurement results between the WN90LP and LX-1128SD.

Photographs of the smart switch circuitry and its installation in the SSTK room are shown in Figure 15. The device was successfully implemented as a microcontroller- and Internet of Things (IoT)-based lighting control and actuation system integrated with the BEMS platform. ON/OFF control of lamp groups was implemented via a relay module driven by a relay driver, providing safe isolation between the low-voltage logic circuit and the AC power circuit. The microcontroller executed control logic based on the lighting set point and predictions from the soft sensor, while current sensors installed on the supply line of each lamp group enabled continuous verification of actual ON/OFF conditions as well as estimation of power and energy consumption. This study limits the control implementation to ON/OFF operation to ensure robustness and simplicity in real-time BEMS operation. However, although the proposed DF-based soft sensor can be expanded to support proportional control strategies, this approach does not fully utilize the dimming capabilities of modern LED systems, which are known to improve energy efficiency and visual comfort.^46,47 The entire system receives a stable DC supply from an internal power module that converts AC mains voltage to the operational level required by the control circuitry.

Figure 15.

Smart switch implemented in the study showing (a) its block diagram; (b) hardware components inside the equipment; and (c) its installation in SSTK room.

WiFi connectivity enabled bidirectional data exchange between the smart switch and BEMS via an MQTT broker. Control commands based on the soft sensor prediction results were transmitted from BEMS to the smart switch to control lamp group status, while actual ON/OFF status and current data were sent back to BEMS for monitoring. Communication operated continuously without significant connection loss during testing. Meanwhile, BLE communication was used during the initial configuration stage for setting network parameters such as set point, SSID, and password.

Figure 16 shows an example of the system output display for indoor lighting condition control. These data represented the integration results of the outdoor illuminance monitoring module, the prediction algorithm for $E_{i n} (x, y)$ distribution, and the zonal lighting actuation system on the smart switch. When the $E_{o u t}$ is 65,000 lux and all lamp groups are ON ( $S_{1} - S_{8} = 0$ ), as displayed in Figure 16(a) and (b), the system produces an $E_{i n} (x, y)$ illuminance distribution across 20 workplane points representing the combined contribution of natural (DF model) and artificial lighting. Prediction results were displayed on the GUI as a heatmap with spatial interpolation to visualize continuous illuminance distribution, and its alignment with manual calculations shows that the algorithm integrated into BEMS operated consistently without computational distortion in real-time.

Figure 16.

Output display of the lighting control system in BEMS displaying (a) GUI of outdoor illuminance monitoring ( $E_{o u t} = 65, 000$ Lux); (b) GUI of ON/OFF status of the lamp groups (all groups are ON); (c) distribution of $E_{i n} (x, y)$ from manual calculation; and (d) distribution of $E_{i n} (x, y)$ from the DF-based soft sensors.

System evaluation

The performance of the monitoring system in the DF-based soft sensor was assessed by evaluating its accuracy, real-time capability, and reliability. System accuracy evaluation was conducted using MAPE under various combinations of active lamp groups and outdoor illuminance variations between 10,000 and 80,000 lux. Test results showed MAPE values of $11 % \pm 8 %$ and $14 % \pm 6$ % for the SSTK Lab and TN-7, respectively. Practically, MAPE values within this range do not cause significant deviation from the standardized illuminance for a work plane (350–500 lux),^41–43 demonstrating the system’s capability in maintaining lighting quality in accordance with recommended standards. However, the stability of MAPE values indicated by relatively small standard deviations may suggest that the system performance was not sensitive to changes in lamp group combinations or fluctuations in $E_{o u t}$ .

The system’s real-time characteristics were assessed by measuring data acquisition, communication, and algorithm computation duration: $57.9 \pm 0.4$ ms, $22.7 \pm 0.8$ ms, and $6.1 \pm 1.2$ s, respectively. Figure 17 shows the data distribution, which indicates that the DF-based soft sensor algorithm computational stage was the dominant component in the overall processing time. With a total processing time of $6.1 \pm 1.2$ s, the system was still suitable for daylight-based adaptive lighting control, considering that natural illuminance dynamics generally do not occur on the millisecond scale but rather on the scale of minutes to hours.^48,49 This finding has important implications for system scalability, since its implementation across multiple rooms or zones will not significantly increase computational loads.

Figure 17.

Distribution of DF-based soft sensor delay test results.

Meanwhile, reliability testing was performed over eight consecutive time intervals with a total of 480 data transmissions from the outdoor illuminance sensors, with all data successfully received and stored in the BEMS database (480/480, success rate = 100%). No data loss was detected (packet loss = 0%), indicating the stability of LAN-based communication and integration with MySQL under real operational conditions. This reliability is crucial in demonstrating that the implemented communication system is capable of supporting continuous operation without performance degradation during the measurement period.

Functional testing of the smart switch, including ON/OFF control, feedback transmission, system configuration (WiFi and BLE), and battery monitoring, demonstrated consistent and reliable operation. All test scenarios achieved a 100% success rate across multiple repetitions, indicating stable bidirectional communication, accurate control execution, and reliable system performance under real-world operating conditions. The 100% success rate confirms that the hardware and communication components are sufficiently robust to support the implementation of the DF-based soft sensor within the BEMS framework.

The operational test results could be indicated by the delay time under relay ON and OFF conditions. The smart switch delay time was calculated as the interval between command issuance and receipt of lamp status feedback on the BEMS dashboard. The test results conducted during working hours indicated that the feedback delay time was consistently higher than the control delay time under both conditions. The command delay for turning the lights ON was 1.8 ± 1.0 s, with a maximum of 4.5 s, while for turning the lights OFF was 1.6 ± 0.7 s, with a maximum of 3.3 s. The feedback delays were 3.4 ± 0.9 s for the ON condition and 3.5 ± 1.3 s for the OFF condition, with maximum values of 6.1 s and 9.0 s, respectively. This variation might be attributed to the bidirectional communication mechanism, which includes device status reading, internal processing, and retransmission via MQTT to the BEMS, whereas control delay time involves only instructions to the actuator. Despite these variations, the response time remains within a few seconds, acceptable for building automation systems that do not require hard real-time control.⁵⁰

The measurement results of smart switch current consumption for the two conditions of relay OFF and relay ON, show that during the OFF condition the average current was 95.22 ± 2.04 mA, while it increased to 164.39 ± 1.91 mA during the ON condition. This increase was due to additional energy required to maintain relay coil activation. During the measurement period, the current in both conditions remained relatively stable, with no significant fluctuations, indicating stable device operation. Finally, the relay condition affected the smart switch operating duration in one battery cycle. When the relay was ON, the system operated for 10.61 h, whereas when the relay was OFF, the duration increased to 27 h due to the continuous energy consumption of the relay coil. The voltage drop of the Lithium 18,650 battery during the final discharge phase occurred more rapidly due to limitations in electrochemical reactions that maintain voltage stability.

Soft sensor applicability

The DF-based soft sensor developed in this study is highly suitable for application in existing buildings since the algorithm was constructed based on measurements of actual building conditions. This approach allows the model to represent the building’s actual spatial characteristics, including geometry, surface reflectance, opening distribution, and environmental conditions. By using empirical data as the training basis, the system can accommodate actual lighting variations, producing illuminance estimates that are representative of the building’s operating conditions.

As for new buildings, the approach can be implemented at the design stage using daylighting simulation data, enabling the soft sensor algorithm to be developed prior to construction. A practical workflow would be as follows: (1) develop a building model and perform daylighting simulations; (2) derive the DF matrix from simulated illuminance data; (3) integrate real-time outdoor illuminance as model input; and (4) deploy the prediction and control algorithm within the BEMS. This approach enables adaptation to different geometries and layouts without extensive in-situ measurements. However, the use of simulation data may produce certain uncertainties, especially in areas near windows that receive direct sunlight. The issue may occur given that daylighting simulations generally calculate DF based on overcast sky conditions, thus the light distribution results are more uniform and do not fully represent clear-sky conditions with direct sunlight components. Consequently, differences between the estimated and actual conditions may occur in perimeter zones of the building. These localized discrepancies characterize the inherent limitation of the DF framework under non-overcast sky conditions, particularly in tropical environments with significant direct solar components. Needless to say, these deviations can be effectively mitigated through targeted calibration at selected perimeter points, allowing the DF-based model to better represent local lighting conditions without modifying the underlying modeling framework.

The performance of the proposed system is influenced by building-specific characteristics such as room geometry, window orientation, and façade configuration, which affect daylight distribution patterns and consequently the accuracy of the DF-based model. Despite these limitations, the DF-based soft sensor remains applicable for practical implementation, as the overcast sky assumption does not necessarily reduce the applicability of the model, but rather emphasizes the importance of integrating architectural design strategies and control systems to limit direct sunlight penetration. Optimization can be achieved through façade design and shading devices such as horizontal or vertical shading, light shelves, spectrally selective glazing, and adjustment of the window-to-wall ratio to reduce direct solar gains without eliminating diffuse light. In addition, spatial zoning arrangements that place primary work areas in the central (core) zone and use perimeter areas for secondary functions may increase illuminance stability in visually critical areas. Integration with adaptive control systems, such as automated blinds and dimmable LED lighting, may enable real-time compensation for light fluctuations, thus deviations remain within acceptable operational limits even under clear-sky conditions. Future studies shall investigate the system by incorporating these optimization strategies.

From a systems perspective, the DF-based soft sensor should be seen as an operational estimation framework for maintaining illuminance within a certain performance range, rather than as an absolute point-estimation model with high precision at every location. In adaptive lighting applications, maintaining illuminance within a target range is functionally more relevant than minimizing lux differences at highly dynamic local points. The study results demonstrate that deviations occurring in a small portion of perimeter areas do not significantly affect overall system reliability, especially when most occupancy zones exhibit good prediction agreement. Thus, the DF-based soft sensor framework has high flexibility for application in both existing and new buildings, is computationally efficient for real-time control, and remains reliable within standard illuminance ranges when supported by integrated design strategies. Future studies shall demonstrate these flexibility, efficiency, and reliability in different types of buildings, different occupants and activity patterns, as well as different climates and regions to assess the system better. The evaluation of actual energy savings is beyond the scope of this study and will be addressed in future work.

Conclusion

This study aims to develop and validate an adaptive lighting control system based on daylight-linked control that integrates the spatial dimensions of DF and real-time outdoor illuminance sensing through soft sensors in a BEMS. The research successfully integrates outdoor illuminance measurements, indoor DF-based illuminance predictions, and a smart switch to enable adaptive lighting controls. The investigated rooms exhibit different spatial characteristics and demonstrate the capability of the proposed methods to predict accurate spatial illuminance distributions.

The constructed DF-based soft sensor algorithm demonstrates good predictive performance at most measurement points in both studied rooms. The algorithm’s MAE and RMSE distributions in SSTK are $8.09 \pm 5.69$ lux and $10.76 \pm 7.70$ lux, respectively, displaying. These values are relatively homogeneous in the central area of the room, despite a significant increase in error appearing only locally at the edge of the room. RMSE distributions in TN-7 are $8.62 \pm 6.54$ lux and $11.85 \pm 8.95$ lux, respectively. The error maps for TN-7 show a higher prediction sensitivity to specific spatial conditions than SSTK, although this increase in error is zonal and does not represent a failure of the algorithm as a whole.

The constructed DF-based soft sensor performs satisfactorily in real-world implementations. The system shows high accuracy with MAPE values of $11 % \pm 8 %$ and $14 % \pm 6$ % for the SSTK Lab and TN-7, respectively, with data acquisition, communication, and algorithm computation duration of $57.9 \pm 0.4$ ms, $22.7 \pm 0.8$ ms, and $6.1 \pm 1.2$ s, respectively. Moreover, the system demonstrates high reliability by successfully delivering all 480 data transmissions over eight consecutive time intervals into the BEMS database.

This study demonstrates a step-by-step approach to develop and implement an adaptive lighting control system based on a DF-based soft sensor that is reliable, scalable in its methodological framework, and applicable for both existing and new buildings. Optimizations via comprehensive façade design, shading installation, and spatial zoning arrangements may improve the indoor daylighting performance. The system should be treated as an operational estimation framework for maintaining illuminance within a certain performance range, rather than as an absolute point-estimation model with high precision at every location. Future studies shall incorporate various optimization strategies into the system, as well as apply the system in different building types, occupant and activity patterns, as well as climates and regions.

Footnotes

Acknowledgements

The authors would like to acknowledge the Department of Nuclear Engineering and Engineering Physics (DTNTF), Faculty of Engineering, Universitas Gadjah Mada (UGM), for their support throughout the research process.

ORCID iDs

Faridah Faridah

Dinta Wijaya

Athallah Naufal Hadi

Mohammad Itqon Alexander

Thomas Oka Pratama

Yakub Fahim Luckyarno

Rachmawan Budiarto

Sunarno Sunarno

Muhammad Khalis Farhan Azvi

Dimas Fredy Prakasa

Funding

This research was funded by the Faculty of Engineering, Universitas Gadjah Mada, through Internal Research Funding No. 1610402/UN1.FTK/SK/HK/2025.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

References

IEA . CO2 emissions in 2023. IEA, 2023. https://www.iea.org/

IEA . World energy outlook 2024. IEA, 2024. https://www.iea.org/terms

UNEP . Global status report for buildings and construction. UNEP, 2023. https://doi.org/10.59117/20.500.11822/45095

Cao

Dai

Liu

. Building energy-consumption status worldwide and the state-of-the-art technologies for zero-energy buildings during the past decade. Energy Build 2016; 128: 198–213. https://doi.org/10.1016/j.enbuild.2016.06.089

Azmi

Mansor

Mokhlis

, et al. Lighting system optimization, components, and simulation tools for building installations: a review. IEEE Access 2025; 13: 160508–160535. https://doi.org/10.1109/access.2025.3578983

Pompei

Blaso

Fumagalli

, et al. The impact of key parameters on the energy requirements for artificial lighting in Italian buildings based on standard EN 15193-1:2017. Energy Build 2022; 263: 112025. https://doi.org/10.1016/j.enbuild.2022.112025

Wijaya

DDA

Utami

Mangkuto

. Multi-objective optimisation of skylight design parameters for a low-rise building in the tropics. International Journal of Technology 2024; 15: 1012–1025. https://doi.org/10.14716/ijtech.v15i4.5484

PUPR . Peraturan menteri pekerjaan umum dan perumahan rakyat nomor 21 tahun 2021. PUPR, 2021. https://peraturan.bpk.go.id/Details/217002/permen-pupr-no-21-tahun-2021 (accessed 2 February 2026).

GBCI . GREENSHIP for new building v1.2. GBCI, 2013. https://www.gbcindonesia.org/greens/new (accessed 15 April 2023).

10.

Mangkuto

Asri

Rohmah

, et al. Prediction of daylight areas with the use of students assessment and computer simulation in the tropics: case study in Bandung, Indonesia. In: 14th conference of international building performance simulation association, Hyderabad, India, 7–9 December 2015. https://doi.org/10.26868/25222708.2015.2208

11.

Lowcay

Gunay

O’Brien

. Simulating energy savings potential with high-resolution daylight and occupancy sensing in open-plan offices. J Build Perform Simul 2020; 13: 606–619. https://doi.org/10.1080/19401493.2020.1807604

12.

Plörer

Hammes

Hauer

, et al. Control strategies for daylight and artificial lighting in office buildings—a bibliometrically assisted review. Energies 2021; 14: 3852. https://doi.org/10.3390/en14133852

13.

Seyedolhosseini

Masoumi

Modarressi

, et al. Daylight adaptive smart indoor lighting control method using artificial neural networks. J Build Eng 2020; 29: 101141. https://doi.org/10.1016/j.jobe.2019.101141

14.

Chiesa

Di Vita

Ghadirzadeh

, et al. A fuzzy-logic IoT lighting and shading control system for smart buildings. Autom ConStruct 2020; 120: 103397. https://doi.org/10.1016/j.autcon.2020.103397

15.

Gao

Zhong

Zou

. Human-centric IoT control: a framework for quantifying the impact of occupant behaviour on energy efficiency in shared offices. J Build Eng 2025; 108: 112784. https://doi.org/10.1016/j.jobe.2025.112784

16.

Colaco

Varghese

Kurian

, et al. A state-of-the-art artificial intelligent techniques in daylighting controller: models and performance. Sci and Techn for Energy Trans (STET) 2023; 78: 35. https://doi.org/10.2516/stet/2023035

17.

Gentile

Lee

Osterhaus

, et al. Evaluation of integrated daylighting and electric lighting design projects: lessons learned from international case studies. Energy Build 2022; 268: 112191. https://doi.org/10.1016/j.enbuild.2022.112191

18.

Feyzi

Mojallali

. Optimal placement of light sensor for improving energy efficiency and visual comfort in smart buildings. E-Prime - Adv in Electr Eng, Elect and Ener 2024; 9: 100681. https://doi.org/10.1016/j.prime.2024.100681

19.

H-G

Choi

S-H

Park

. Linear regression based indoor daylight illuminance estimation with simple measurements for daylight-linked lighting control. J Build Perform Simul 2023; 16: 574–587. https://doi.org/10.1080/19401493.2023.2185684

20.

Van De Meugheuvel

Pandharipande

Caicedo

, et al. Distributed lighting control with daylight and occupancy adaptation. Energy Build 2014; 75: 321–329. https://doi.org/10.1016/j.enbuild.2014.02.016

21.

Peruffo

Pandharipande

Caicedo

, et al. Lighting control with distributed wireless sensing and actuation for daylight and occupancy adaptation. Energy Build 2015; 97: 13–20. https://doi.org/10.1016/j.enbuild.2015.03.049

22.

Lee

Lim

. Lighting control method based on RIIL to reduce building energy consumption. Energy Rep 2024; 11: 2090–2098. https://doi.org/10.1016/j.egyr.2024.01.075

23.

Acosta

Campano

MÁ

Domínguez-Amarillo

, et al. Continuous overcast daylight autonomy (DAo.con): a new dynamic metric for sensor-less lighting smart controls. Leukos 2023; 19: 343–367. https://doi.org/10.1080/15502724.2022.2135528

24.

Perera

Ratnaweera

DAAC

Dasanayaka

, et al. The role of artificial intelligence-driven soft sensors in advanced sustainable process industries: a critical review. Eng Appl Artif Intell 2023; 121: 105988. https://doi.org/10.1016/j.engappai.2023.105988

25.

Faridah

Utami

Wijaya

DDA

, et al. An indoor airflow distribution predictor using machine learning for a real-time healthy building monitoring system in the tropics. Build Serv Eng Res Tecnol 2024; 45: 293–315. https://doi.org/10.1177/01436244241231354

26.

DHW

Lou

Ghaffarianhoseini

, et al. A review of calculating procedures on daylight factor based metrics under various CIE standard skies and obstructed environments. Build Environ 2017; 112: 29–44. https://doi.org/10.1016/j.buildenv.2016.11.019

27.

Chinazzo

Wienold

Andersen

. Influence of indoor temperature and daylight illuminance on visual perception. Light Res Technol 2020; 52: 350–370. https://doi.org/10.1177/1477153519859609

28.

Acosta

Campano

MÁ

Domínguez

, et al. Minimum daylight autonomy: a new concept to link daylight dynamic metrics with daylight factors. Leukos 2019; 15: 251–269. https://doi.org/10.1080/15502724.2018.1564673

29.

Hraška

Čurpek

. The practical implications of the EN 17037 minimum target daylight factor for building design and urban daylight in several European countries. Heliyon 2024; 10: e23297. https://doi.org/10.1016/j.heliyon.2023.e23297

30.

Jung

Yoon

. Study on the prediction and improvement of indoor natural light and outdoor comfort in apartment complexes using daylight factor and physiologically equivalent temperature indices. Energies 2018; 11: 187. https://doi.org/10.3390/en1107187

31.

İnan

. An investigation on daylighting performance in educational institutions. Struct Surv 2013; 31: 121–138. https://doi.org/10.1108/02630801311317536

32.

Signify Holding BV . Ecofit LED tubes T8: 16W (36W) G13 cool daylight linear tube – datasheet. Signify Holding BV, 2025.

33.

Kim

Lee

Kim

. Evaluation of illuminance measurement data through integrated automated blinds and LED dimming controls in a full-scale mock-up. Energies 2020; 13: 3238. https://doi.org/10.3390/en13123238

34.

Lin

Chen

. Strategies on uniformity lighting in office space under energy-saving environment. Buildings 2023; 13: 1797. https://doi.org/10.3390/buildings13071797

35.

DiLaura

Houser

Mistrick

, et al. The IES lighting handbook. 10th ed. Illuminating Engineering Society of North America, 2011.

36.

Susan

Prihatmanti

. Daylight characterisation of classrooms in heritage school buildings. Plan Malaysia J 2017; 15: 236. https://doi.org/10.21837/pmjournal.v15.i6.236

37.

Hodson

. Root-mean-square error (RMSE) or mean absolute error (MAE): when to use them or not. Geosci Model Dev 2022; 15: 5481–5487. https://doi.org/10.5194/gmd-15-5481-2022

38.

Robeson

Willmott

. Decomposition of the mean absolute error (MAE) into systematic and unsystematic components. PLoS One 2023; 18: e0279774. https://doi.org/10.1371/journal.pone.0279774. https://api.semanticscholar.org/CorpusID:256939695

39.

Chao

Twycross

Garibaldi

. A new accuracy measure based on bounded relative error for time series forecasting. PLoS One 2017; 12: 1–23.

40.

BSN . SNI 6197-2020 konservasi energi pada sistem pencahayaan. BSN, 2020.

41.

Lee

J-H

Moon

Kim

. Analysis of occupants’ visual perception to refine indoor lighting environment for office tasks. Energies 2014; 7: 4116–4139. https://doi.org/10.3390/en7074116

42.

Yang

Jeon

. Effects of correlated colour temperature of LED light on visual sensation, perception, and cognitive performance in a classroom lighting environment. Sustainability 2020; 12: 4051. https://doi.org/10.3390/su12104051

43.

Idrus

Rahim

Hamzah

, et al. An alternative approach in assessing visual comfort based on students’ perceptions in daylit classrooms in the tropics. Civil Eng and Arch 2020; 8: 801–813. https://doi.org/10.13189/cea.2020.080507

44.

Ayoosu

Lim

Y-W

Leng

, et al. Daylighting evaluation and optimisation of window to wall ratio for lecture theatre in the tropical climate. J of Daylight 2021; 8: 20–35. https://doi.org/10.15627/jd.2021.2

45.

DHW

Chen

, et al. Simple mathematical models to link climate-based daylight metrics with daylight factor metrics and daylighting design implications. Heliyon 2023; 9: e15786. https://doi.org/10.1016/j.heliyon.2023.e15786

46.

Kar

Kumar

Shareef

, et al. An intelligent lighting control system for individual visual comfort and energy savings in buildings. J Reliab Intell Environ 2023; 9: 385–398. https://doi.org/10.1007/s40860-022-00189-y

47.

Petkovic

Bajovic

Vukobratovic

, et al. Smart dimmable LED lighting systems. Sensors 2022; 22: 8523. https://doi.org/10.3390/s22218523

48.

Wijntjes

Eisemann

, et al. Quantifying the spatial, temporal, angular and spectral structure of effective daylight in perceptually meaningful ways. Opt Express 2023; 31: 8953. https://doi.org/10.1364/oe.479715

49.

Walkenhorst

Luther

Reinhart

, et al. Dynamic annual daylight simulations based on one-hour and one-minute means of irradiance data. Sol Energy 2002; 72: 385–395. https://doi.org/10.1016/s0038-092x(02)00019-1

50.

Al Mughairi

Beach

Rezgui

. Post-occupancy evaluation for enhancing building performance and automation deployment. J Build Eng 2023; 77: 107388. https://doi.org/10.1016/j.jobe.2023.107388