Designing metal halide-based road illumination systems in developing countries using regression and neural networks

• Performed extensive photometric simulation of metal halide-based public road lighting systems. • Propounded models based on multiple regression analysis and artificial neural networking (ANN). • Luminaire power per unit road length and standard specular factor were the most important predictor variables. • The 50-neuron ANN model had the lowest mean square error and highest coefficient of correlation. • ANN-based models performed much better than the multiple regression analysis-based model. Public road illumination systems shape the urban landscape and are ineluctable to ensuring the nocturnal safety of motorists, wayfarers, and pedestrians. In developing nations including India, various local bodies still rely on time-honoured discharge lamps including metal halide (MH) lamps for road lighting. This study aimed to develop practicable models for predicting luminance and energy efficiency parameters of MH-based public road illumination systems. Photometric simulations were conducted for 80,000 installation combinations considering single-sided arrangements of MH luminaires and candidate models of predicting luminance and energy efficiency parameters were propounded with multiple regression analysis and two-layer feed-forward artificial neural networks (ANNs). Luminaire power per unit road length (τ, W/km) and standard specular factor (S 1 ) were found to be the most important predictor variables for the prediction of photometric parameters and installation energy efficiency respectively. For some conventional design configurations, the performances of the models were assessed and the 50-neuron ANN model performed well with an error margin of – 1.45 % to + 1.71 % for the prediction of average luminance, of – 8.67 % to + 9.08 % for the prediction of overall uniformity, of – 8.31 % to + 16.40 % for the prediction of longitudinal uniformity and of – 1.56 % to + 1.52 % for the prediction of installation energy efficiency. Through extensive photometric simulation and data analysis, this study provides useful insights for the commissioning of road lighting projects especially pertaining to the usefulness of ANN models for the planning and optimization of public road illumination systems in developing countries.