Modelos de generación de viajes considerando relaciones espaciales

Abstract

This doctoral thesis specifies trip generation models considering spatial relationships. Firstly, a multiple linear regression (MLR) model is specified using zonal data. These models are compared evaluating the hypotheses of each one along with their fits. Trip generation models considering spatial correlation are also proposed and the improvements found using these new models are analysed and compared to those which do not consider spatial correlation. The world is currently aware of the importance played by transport in everyday life. From this point of view and concentrating on the city of Santander, in the province of Cantabria, Spain, this thesis proposes the design of two types of multiple linear regression models , attraction/production models for different trip categories (overall, household based work related, household based, other reasons and not household based) using zonal data and other trip production models for two trip categories (household based for reasons of work and others) using household data. A thorough individualized and aggregated analysis is also carried out on the data obtained through their representation on plans showing the different zones located in the study area, as well as the comparison of the results obtained in certain trip categories, in this case using the concept of error percentages as well as plans. The conclusions drawn are extensive and developed throughout the work, however, it can be noted that zonal models obtain better estimations than those elaborated using household data, above all in the case of households based trips for other reasons, or that all the models are on the safe side in their trip estimations as they tend more on the side of over estimation than under estimation. An overall spatial correlation model is specified for a group examination of all the units making up the sample in order to determine if the spatial units are randomly distributed or, on the contrary, follow a determined pattern. In order to consider cluster situations in a determined area a local spatial correlation model is estimated to measure the spatial auto correlation for the location of each observation. The goodness of fit of different models is compared and a series of methodological hypotheses are checked based on the performance of the specifications which consider the existence of spatial relationships between observations. The models have been applied to the metropolitan area of Santander (Spain) with the goal of obtaining advanced trip generation models for this location. In the case of Santander, the models considering spatial dependency between observations show a substantial improvement in fit compared to the MLR models.