Efeitos da dependência espacial em modelos de previsão de demanda por transporte
Data
2017-11-15
Autores
Título da Revista
ISSN da Revista
Título de Volume
Editor
Biblioteca Digital de Teses e Dissertações da USP
Universidade de São Paulo
Escola de Engenharia de São Carlos
Universidade de São Paulo
Escola de Engenharia de São Carlos
Resumo
Descrição
A dependência espacial para análise de dados de demanda por transportes, que está entre as principais questões analítico-espaciais consideradas na análise de transportes, constituiu o foco deste trabalho. Ignorar questões de análise espacial pode invalidar os resultados da análise, levar a previsões inadequadas e, conseqüentemente, a um planejamento ineficiente. Em virtude disso, admitiu-se que a introdução de indicadores de dependência espacial na modelagem de demanda por transportes deveria produzir resultados mais precisos e, desta forma, mais confiáveis dos que os obtidos com modelos tradicionais. Neste sentido, o principal objetivo deste trabalho foi comparar a projeção de demanda por transportes, especificamente na fase de previsão de viagens produzidas de base domiciliar, realizada através de modelos convencionais e de modelos alternativos, que introduzem indicadores para medir a dependência espacial. O trabalho é todo desenvolvido em ambiente SIG (Sistemas de Informações Geográficas), através de ferramentas de análise e estatística espacial, assim como ferramentas de planejamento de transportes de um SIG-T (SIG para Transportes). As ferramentas de análises espaciais serviram tanto para produzir os indicadores de dependência espacial (variáveis espaciais) como para avaliar os resultados dos modelos. Aplica-se o método, que avalia a introdução de indicadores globais e locais de dependência espacial nos modelos alternativos, através de um estudo de caso na cidade de Porto Alegre - RS, que tem por base dados de pesquisa de origem e destino (O-D) obtidos através de entrevista domiciliar (EDOM) em dois períodos distintos (1974 e 1986). Estas informações correspondem aos dados necessários do ano base, que foram utilizados na calibração dos modelos, e do ano meta, que constituíram as informações necessárias para análise dos resultados de estimativas futuras. Conclui-se que a introdução de variáveis espaciais é importante, uma vez que os melhores resultados foram obtidos com modelos alternativos, tanto na etapa de calibração e diagnóstico dos modelos como na etapa de validação (estimativas futuras). No entanto, a dinâmica apresentada pelo desenvolvimento urbano, como é o caso de Porto Alegre, acarreta alterações nas relações entre as diferentes variáveis com o fenômeno estudado, modificando, inclusive, os padrões espaciais. Esta conclusão é dada pelo fato que, o modelo mais ajustado para os dados do ano base não foi o que apresentou os melhores resultados para estimativas futuras. Isto conduz à hipótese, a ser explorada em trabalhos futuros, de que a análise desta dinâmica e o estudo de formas de considerá-la nos modelos de demanda por transportes pode produzir resultados ainda melhores
The degree of spatial data dependence, which is among the issues of spatial analysis that should be considered in transportation planning, is the focus of this study. Ignoring this particularity of data can: produce wrong estimates; jeopardize the results of analyses; and, as a consequence, lead to unsuccessful planning. Therefore, the basic assumption of this work was that the inclusion of spatial dependence indicators can produce more accurate and reliable estimates than those obtained with traditional model structures. In order to test this hypothesis, the main objective of this study was to compare demand predictions produced by traditional models with those produced by alternative models that include indicators of spatial dependence. The study was limited to home-based production trip models, which are part of the trip generation phase of the traditional four-step modeling approach. All work was conducted in a GIS (Geographic Information System) environment, making use of spatial statistics and analysis tools, as well as transportation planning tools available in a GIS-T (i.e., a dedicated GIS for Transportation). Spatial analyses tools were used to generate the spatial dependence indicators and to evaluate the results of the application. A case study was carried out in the city of Porto Alegre, which is the capital of the brazilian state of Rio Grande do Sul, for evaluating the impacts of the addition of global and local indicators of spatial dependence in the models. Two O-D surveys carried out in the years 1974 and 1986 provided the data needed for calibration and validation. The first one was taken as the base year and the second one as the goal year. The results of the application showed that the performance of the models can be improved in both calibration and validation phases with the insertion of spatial variables. However, the urban growth observed in a very dynamic context, such as in the city studied, may dramatically change the relationships between variables, including their spatial patterns. That aspect was responsible for the fact that the model with the best performance in the calibration phase was not the one producing the most accurate forecasts. It raised the hypothesis, to be explored in future research, that the analysis of those dynamic processes and their consideration into transportation demand models are also needed to improve even further the performance of the models
The degree of spatial data dependence, which is among the issues of spatial analysis that should be considered in transportation planning, is the focus of this study. Ignoring this particularity of data can: produce wrong estimates; jeopardize the results of analyses; and, as a consequence, lead to unsuccessful planning. Therefore, the basic assumption of this work was that the inclusion of spatial dependence indicators can produce more accurate and reliable estimates than those obtained with traditional model structures. In order to test this hypothesis, the main objective of this study was to compare demand predictions produced by traditional models with those produced by alternative models that include indicators of spatial dependence. The study was limited to home-based production trip models, which are part of the trip generation phase of the traditional four-step modeling approach. All work was conducted in a GIS (Geographic Information System) environment, making use of spatial statistics and analysis tools, as well as transportation planning tools available in a GIS-T (i.e., a dedicated GIS for Transportation). Spatial analyses tools were used to generate the spatial dependence indicators and to evaluate the results of the application. A case study was carried out in the city of Porto Alegre, which is the capital of the brazilian state of Rio Grande do Sul, for evaluating the impacts of the addition of global and local indicators of spatial dependence in the models. Two O-D surveys carried out in the years 1974 and 1986 provided the data needed for calibration and validation. The first one was taken as the base year and the second one as the goal year. The results of the application showed that the performance of the models can be improved in both calibration and validation phases with the insertion of spatial variables. However, the urban growth observed in a very dynamic context, such as in the city studied, may dramatically change the relationships between variables, including their spatial patterns. That aspect was responsible for the fact that the model with the best performance in the calibration phase was not the one producing the most accurate forecasts. It raised the hypothesis, to be explored in future research, that the analysis of those dynamic processes and their consideration into transportation demand models are also needed to improve even further the performance of the models
Palavras-chave
análise de demanda por transportes, autocorrelação espacial, estatística espacial, SIG, analysis of transportation demand, GIS, spatial autocorrelation, spatial statistics