23º SINAPE - Simpósio Nacional de Probabilidade e Estatística

Dados do Trabalho


Título

GAMLSS AND EM HELPING TO IDENTIFY VEHICLES WITH HIGH EMISSION OF POLLUTANTS

Resumo

To define criteria based on measurements with a remote sensing device (RSD) to identify Otto cycle LDV [light duty vehicles] with high emissions of carbon monoxide, hydrocarbons or nitrogen monoxide, we used data from 179,142 vehicles in metropolitan São Paulo, with complete measurements. We adjusted statistical models of GAMLSS - Generalized Additive Models for Location, Scale and Shape class to test the influence of fuel type, VSP [vehicle specific power] and Proconve [Brazilian vehicle emission control program] phases on measurements of CO, HC and NO emission rates. The emissions were then conceptually subdivided into two groups: vehicles with normal and abnormal emission, this for the various pollutants in vehicles of L3, L4 and L5 phases. Latent variables were defined to indicate the distribution of vehicles in relation to those groups and phases. The algorithm EM-Expectation - Maximization was employed to identify all the parameters of the distributions. To determine the limits values for vehicles with high emissions of pollutants and Proconve phase, we use the 98% percentiles of the distributions set for vehicles of groups with normal emissions, therefore, the type I error was set at 2% and this percentage has been established considering the type II error of indicating the vehicle as normal emission when in fact it is a High emitter. Indicative values were determined with high emissions vehicles according to the pollutant and Proconve phase.

Palavras-chave

Key words: Vehicle emission, on-road measurement, vehicle specific power, remote sensing devices, I/M program, high emitters, carbon monoxide, hydrocarbons, nitrogen monoxide.

Área

Estatística Aplicada em Engenharia e Ciências Exatas

Autores

ANTONIO CASTRO BRUNI, JOÃO VICENTE ASSUNÇÃO