A local area traffic model

Abstract

In recent times a need has developed for methods and models which may be used to simulate vehicle movements and travel demands in smallscale networks. These networks might represent a small part of an urban area in which particular aspects of traffic movements and congestion, environment and/or transport planning, and traffic regulation policies might require investigations. Typically such investigations would consider a relatively short time duration, such as a peak period. This thesis describes an attempt to construct a traffic model for use in such networks. In the body of this thesis the model is termed the Local Area Traffic Model. The development and construction of the model may be divided into a number of phases. Firstly, reviews were made to reveal: (i) the development of traffic assignment models in general, and the assumed path selection mechanisms used in such models (Chapter 2); (ii) studies of the observed behaviour of drivers and their route selection criteria (Chapter 3); (iii) some existing traffic assignment models used for small-scale networks (Chapter 4); and (iv) various problems, theories and algorithms concerned with the determination of paths through networks (Chapter 5). From the knowledge revealed by the reviews of previous work, a set of requirements for the local area traffic model was proposed. Additional work was also conducted to provide clarification of some of the proposed model components. A study of observed variabilities of travel times for particular journeys by an individual was undertaken. Results concerning relationships between measures of variability and mean travel times, and the distributions of travel times, were suggested. Some new results about the determination of minimum cost paths from link to link in a network with turn penalties were also suggested. The local area traffic model was then formulated based on a general hypothesis that competing paths for a particular trip were chosen on a probabilistic basis. The probability of choice of a given path was based on the perceptions of travellers about the costs involved in using the path. These costs, and thus the path probabilities, were dynamic and could vary throughout the simulated time period, in response to changing travel conditions and traffic congestion. The basic model was defined with two unknown global parameters in the representation of the path selection process. These parameters were measures of (i) the variability of individual travel times on a path segment; and (ii) the spread of trips between competing paths of different path travel costs. The local area traffic model included queueing and delay sub-models, for various intersection types, and travel time - flow rate relationships for particular classes of urban streets. Some new results concerning mean delays and queueing at signalized intersections, and travel timeflow relationships on surface streets, were required for the traffic model. Once the model was established, a method to test its applicability was suggested. Suitable data for model testing was then required. A test area was selected, and data collection was undertaken to yield two independent sets of data, consisting of peak period traffic demands and observed street volumes for each of two separate peak periods in the study area. The traffic model was then used with the test data, and its ability to reproduce observed travel conditions in the test area was investigated. Optimization techniques were used to find best-fit values for the global parameters for each data set. As the data sets for each peak period were collected independently, comparative tests of the optimized parameter values could be made with a technique described as dual samp.ling. Investigations of the robustness and sensitivi.ty of the parameters were then undertaken. Model errors were examined to test for possible biases in the model. Certain conclusions about the suitability of the model for applications to small-scale, detailed traffic networks were reached as a result of the model tests. It is suggested that the general model is a suitable tool for examining vehicle trips through a small network. The path selection process in the model appears capable of synthesizing observed trip movements in a dynamic traffic demand situation. Investigations of the observed data enabled representative values of the two global model parameters to be determined for the test area. The model was capable of reproducing observed traffic volumes inside the test area to a high degree of accuracy. Some areas of future and related research are also suggested.