Estimating municipal road lighting parameters with linear and support vector regression models

Abstract

Contemporary public road illumination projects commissioned by municipalities and other local authorities install new light-emitting diode (LED)-based luminaires or retrofit existing conventional discharge lamp-based luminaires with LEDs. Such projects require conforming to minimum photometric and energy efficiency standards, and in early-stage planning, it is desirable to estimate illuminance and energy efficiency parameters for ensuring compliance with national and international codes of practice and modification of design parameters, if required. With a view of recent literature proffering linear regression models for the rapid estimation of photometric and energy efficiency parameters of road lighting systems, this study was conducted to perform a comparative assessment of various candidate models in terms of these statistical indices: coefficient of determination (R 2 ), root mean square error (RMSE), mean squared error (MSE) and mean absolute error (MAE). Photometric simulations of road lighting systems with single-sided pole arrangements were conducted in a proprietary software for five randomly chosen LED power ratings between 33 W and 78 W, and the generated extensive simulation data was utilized to develop and train linear and support vector machine (SVM)-based regression models for the prediction of average illuminance, overall uniformity of illuminance, and energy efficiency of lighting installations. Overall, the cubic SVM models demonstrated the most satisfactory performance (R 2 = 0.99, RMSE ≤ 0.51, MSE ≤ 0.26, and MAE ≤ 0.44) among all the developed models, and this was also discerned in several experimental cases. This study offers new insights for augmenting extant practices in municipal public road lighting design and commissioning.