Estimativa de geração de viagens de veículos de carga em áreas urbanas utilizando modelagem geográfica e modelo linear generalizado

• Ano de defesa: 2018
• Autor(a) principal: Victor Lima Migliorini
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Geração de Viagens de Veículos de Carga; Regressão Linear Generalizada; Regressão Linear Generalizada Geograficamente Ponderada; Logística empresarial; Modelos lineares (Estatística); Transportes; Transporte de mercadorias
• Link de acesso: http://hdl.handle.net/1843/BUOS-B64JGF

The urban freight transportation has gained importance in transport management and public policies in medium and large cities in Brazil and the world. Accurately estimating the number freight trips to supply retail stores is one of the needs highlighted by transport planners. The traditional process for estimating the number of truck-trips employs linear regression. Although suitable, they present limitation regarding the heterogeneity of the urban context as well as the nonlinearity of the variables involved in the problem. Recent investigations on generalized linear models applied other areas of transportation have delivered promising results, therefore not yet evaluated for urban freight. This dissertation introduces models to estimate freight-trip generation to supermarket in urban areas, comparing the results obtained through linear model, generalized linear model and geographically weighted model. Using data from freight-trip generation to markets and supermarkets in Belo Horizonte and socioeconomic data, scenarios were developed using the three models. Findings showed generalized linear models presenting relative gains when compared to the traditional linear models. The geographically weighted model also presented better results than the linear regression. Statistically, the generalized linear model presented slightly better results than the geographically weighted model. However, instead of a static number, the geographically weighted model outputs a continuous surface with local estimative of freight-trip per pixel. This can change the paradigm and innovate the way freight-trip generation is modeled. Finally, through the analysis of the independent variables used in the models, it was found that the higher the average income and the population density, density of jobs and density of customers, lower is freight-trip generation in the region.