With the rapid progression in the development of energy-efficient white light-emitting diodes (LEDs) over the past two decades, the practice of road lighting has undergone a global paradigm shift. Considering the recent advancements in simulation techniques and machine learning methods, this study analyzed the performance parameters and appropriateness of an ensemble of machine learning models for the computation of photometric and energy efficiency parameters. Considering white LEDs of common power ratings (38 W to 131 W), simulations were performed for single-sided, opposite, and staggered luminaire layouts, and three datasets, each containing 10,000,00 samples, were generated. A program was written, and the ensemble model, comprising random forest (RF), gradient boosting regressor (GBR), and linear regression (LR) as the base models, was trained on all three datasets utilizing the five-fold cross-validation technique. The ensemble model performed satisfactorily for the prediction of average illuminance (Eav), overall uniformity of illuminance (U0), and installation energy efficiency (epsilon x), but was outperformed by the RF model. For some test cases, the maximum absolute prediction error of the RF model was lower than both the ensemble model and the baseline LR model. Feature analysis indicated that power/distance (tau, W/km), mounting height (h, m), power rating (P, W), pole spacing (S, m), luminous flux (phi, lm), and luminous efficacy (eta, lm/W) were the most important input features. This study and a developed universal serial bus (USB) flash drive-based lightweight tool may assist qualified facilitators in swiftly predicting relevant parameters in the pre-commissioning stage of municipal road lighting projects.
Addressing lighting conflicts in multigenerational households, this study prioritizes the lighting requirements of older adults and develops older-adult-oriented design methods grounded in visual gerontology. Through literature review, lighting parameters for four generational groups are quantified, establishing that older adults require 500-750 lx (2-3 times the 200-300 lux baseline for younger adults) and 30-50 lx nighttime lighting to reduce fall risks due to vision decline. A "zoning, time-based, and layered" lighting strategy framework is proposed based on intergenerational needs analysis. Using DIALux software, three schemes are simulated: traditional uniform lighting, simple zoned lighting, and multi-layered combined lighting. A 30m2 living room model with six single-generational and multigenerational scenarios evaluates illuminance distribution, glare control, and energy consumption. Results demonstrate that multi-layered combined lighting - featuring 750 lx/3000 K localized enhancement in older-adult reading areas with independent control - addresses the high-illuminance needs of older adults while avoiding visual interference and circadian disruption for other generations, achieving 32% energy savings compared to traditional uniform lighting. Field verification in three multigenerational homes measured illuminance and color temperature parameters over 7 days. Simulation accuracy reached R2 = 0.9883 with 4.3% average relative error, meeting CIE standards. This research provides quantitative parameter guidance and verification techniques for multigenerational housing lighting design, particularly those configurations prioritizing older adults, offering significant value for improving the residential environmental quality experienced by older adult occupants.
This paper presents a novel approach to expedite the simulation process in the early stages of design by leveraging machine learning for the evaluation of lux levels in diverse layouts under electrical lighting conditions. The study introduces a methodology wherein each layout is discretized into points (sensors), capturing illuminance levels from electrical lighting sources. A dedicated machine learning model is developed, capable of predicting lux levels for individual points based on a set of comprehensive predictors, including distances from walls to points (west, east, south, north), distances from luminaires to walls (west, east, south, north), luminaire types, wall reflectance, roof reflectance, and floor reflectance. The chosen Catboost regressor, with an impressive R square value of 0.92, serves as the cornerstone of this predictive modeling. Furthermore, a comparative analysis is conducted between the proposed machine learning model and simulations performed using Climate Studio. The comparison extends across three distinct lighting sources and diverse layouts, revealing a substantial similarity ranging from 0.40 to 0.99. This study not only introduces an innovative approach to accelerate the simulation process but also validates the effectiveness of the developed machine learning model in predicting lux levels, providing valuable insights for architects and designers in the early stages of layout design.
In this study, the accuracy of indoor lighting simulations using DIALux software was evaluated by comparing simulated illuminance with field measurements taken at a 1 & times; 1 meter resolution across a structured grid. This spatial alignment enabled a robust statistical comparison, revealing that both datasets followed an approximately normal distribution, although simulated values exhibited higher variability and a slight tendency to overestimate illuminance in central zones. Descriptive statistics showed that the mean measured illuminance was 488.91 lux (SD = 106.58), while the simulated mean was 487.31 lux (SD = 129.06). A paired t-test indicated a statistically significant difference (p < .05), yet the high coefficient of determination (R-2 = 0.92) confirmed a strong correlation between measured and simulated values. Visual analysis using heatmaps further supported this agreement, with both datasets showing higher illuminance near the center of the space and lower values at the edges. Despite this, both datasets failed to meet the EN 12,464-1 uniformity requirement (min/avg >0.6), suggesting uneven light distribution. To improve lighting performance and reduce energy consumption, alternative luminaire layouts were simulated, achieving an 18% reduction in energy use and improved uniformity (approximate to 0.7). Optimal desk placement zones were also identified by analyzing consistent compliance with lighting standards under various lighting scenarios. Finally, artificial neural networks (ANNs), including Multilayer Perceptron (MLP) and Radial Basis Function (RBF) models, were used to predict illuminance based on spatial coordinates. The RBF network outperformed MLP with lower RMSE (17.61 lux) and higher R-2 (0.9727), demonstrating superior ability to capture non-linear spatial variations. These findings validate the applicability of DIALux for early-stage lighting design and highlight the potential of ANN-based models for real-time lighting control and optimization in workspaces. The integration of simulation, measurement, and AI techniques offers a comprehensive approach to energy-efficient and user-centered lighting design.
Obstacle contrast is a crucial metric for improving visibility at night. This study investigated how lighting design parameters and obstacle spatial locations affect their contrast along a luminaire cycle, as a quasi-dynamic reconstruction rather than a continuous time-dynamic measurement. Field measurements and computer simulations were conducted on a two-lane road equipped with seven LED luminaires controlled via a Bluetooth solution. Five obstacle models, each model representing obstacles of different heights, were spatially positioned along lateral grids between two streetlights. A vehicle-mounted camera recorded high-dynamic-range (HDR) luminance maps, while the lumen output of streetlights was adjusted for two pole spacing configurations, both with and without headlights. HDR images provided detailed luminance data to evaluate factors influencing contrast. Contrast contour maps illustrated changes under varying lighting conditions, and a simulation model approximating the experimental setup was developed to examine overall trend behavior. Normalized comparisons between experimental and simulated contrast results showed similar relative patterns, indicating trend-level correspondence. Results showed a weak correlation between obstacle luminance and contrast for obstacles 1.0 m or shorter, but a strong correlation for obstacles 1.5 m in height. For very low obstacles, contrast decreased with increased lumen output, whereas for taller obstacles (1.5 m), contrast improved with higher lumen output. While reduced pole spacing increased pavement luminance, it did not increase obstacle contrast. A mixed-effects GLM statistical analysis identified target height, vehicle headlight status, and pole spacing as the dominant factors influencing obstacle contrast. This research offers insights for optimizing lighting design to enhance road safety.
