Redes Neurais de Classificação e Regressão Aplicadas no Contexto de Iluminação Pública

Abstract

The relevance of public lighting for various everyday aspects such as security and mobility motivates the use of technologies that improve service provision to the population in the many implementation phases. As a complement to software specialized in simulations, this work proposes to use learning machine techniques to ensure compliance with Brazilian Standard 5101. Exploring concepts related to public lighting and networks neural networks, Multilayer Perceptron (MLP) is implemented using backpropagation, for regression, to determine illuminance parameters and their uniformity (requirements partial compliance with the standard) and classification (which uses illuminance and its uniformity to define compliance to the standard). Models are trained and tested with real parameters of mesh configurations and lighting devices in simulations of projects to inform whether a project data set meets standards established by the standard. Regression and classification MLPs achieved MSE of 0.002 and 97.26% accuracy, respectively.