Frans Tillema

Comparison of Neural Networks and Gravity Models in Trip Distribution

Transportation engineers are commonly faced with the question of how to extract information from expensive and scarce field data. Modeling the distribution of trips between zones is complex and dependent on the quality and availability of field data. This research explores the performance of neural networks in trip distribution modeling and compares the results with commonly used doubly constrained gravity models. The approach differs from other research in several respects; the study is based on both synthetic data, varying in complexity, as well as real-world data. Furthermore, neural networks and gravity models are calibrated using different percentages of hold out data. Extensive statistical analyses are conducted to obtain necessary sample sizes for significant results. The results show that neural networks outperform gravity models when data are scarce in both synthesized as well as real-world cases. Sample size for statistically significant results is forty times lower for neural networks.

Introduction to and Calibration of a Conceptual LUTI Model Based on Neural Networks

This paper deals with Land-Use-Transport-Interaction (LUTI) and presents the background and the calibration of a conceptual data driven LUTI modeling tool which is based on neural networks. A literature survey reveals the opinion of experts on the state of the art LUTI models: currently used land use transport models are too aggregate in substance to match travel demand models. Therefore research is conducted into the refinement of the models; resulting in comprehensive models. Unfortunately lack of theoretical frameworks results in these models not being operational on a large scale. This paper looks for an alternative approach and therefore addresses the following questions: (i) what are solution methods to make LUTI models more applicable; (ii) is there a sound way to put into operation these solution methods; (iii) what modeling technique is suitable to be used in this context; (iv) how does the conceptual LUTI then look like; and finally (v) can we calibrate and test this model. This leads to a conceptual model with three building blocks; (i) accessibility; (ii) household location choice; and (iii) employer location choice. Based on the demands and the previously mentioned lack of clear theories, it is concluded that a data driven approach, using Artificial Neural Networks (ANNs), is suitable to fit the framework. The auto calibration of ANN(s) ensures that complex relationships are found without a theoretical framework. The calibration of the ANNs in the model shows good results. Further research has to result in the actual implementation of the model. For the covering abstract see ITRD E129315.

Trip generation: comparison of neural networks and regression models

This paper deals with trip generation, examining the performance of neural networks (NNs) and commonly used regression models. The research reported herein aims to answer the question of whether NNs can outperform traditional regression models or not. The NNs are tested in 2 situations with regards to the data availability: 1) when data is scarce; and 2) when data is sufficient. Synthetic households, generated using travel diary data, are the basis for the research. These households are divided over a zone in varying complexities, from homogeneous without statistical deviation on the household characteristics, to inhomogeneous with a deviation on household characteristics. The use of synthetic data, without unknown noise, provides an opportunity to clearly determine the impact of complexity on the forecasting results. The question of whether NNs can be used in trip generation modeling is answered affirmatively. Overall, however, NNs do not outperform classical regression models in situations where data is scarce. Advantages over regression models are negligible.

A CONCEPTUAL LUTI MODEL BASED ON NEURAL NETWORKS

This paper deals with Land-use-Transport-Interaction (LUTI) and presents a conceptual model of a data driven LUTI modelling tool based on neural networks. A starting point is the general opinion of experts on the state of the art LUTI models; currently used land use transportation models are too aggregate in substance. Therefore to enhance the interaction between transport and land-use, researchers propose the refinement of the models; internalising more comprehensive relationships. However, the lack of a good theoretical framework impedes the development and consequently the use of these models on a large scale. This fact has fuelled questions: (i) does transport planning need these comprehensive descriptions of land use; (ii) is it necessary to disaggregate and refine the models further in order to be able to do consistent transport planning; (iii) which land use characteristics should at least be internalised; and (iv) which modelling method is suitable for implementation. Based on literature, the hypothesis is that the first question can be answered by ’no’ and no further refinements are needed. Literature shows that the main drivers for land-use changes are the location choice of both households and firms. Therefore only these two building blocks are used in relation with the transport component. Artificial neural networks (ANNs) will be used as the modelling technique. ANNs are data driven techniques that find relationships in data during an auto calibration process. Therefore ANNs can work without having a sound theoretical framework. This characteristic offers possibilities for LUTI modelling that lacks a sound theoretical framework. The research leads to a conceptual model. A literature review shows that the individual building blocks of the conceptual LUTI model can be modelled using neural networks. However, an integration of the building blocks has not been established yet. Further research has to result in the actual implementation of the proposed model.