Synopsis

Origin-Destination (O-D) matrices are used every stage of transportation planning process. All the policies evolved out of transportation studies therefore are directly dependent and sensitive to O-D matrices used. The traditional way of determining O-D matrices from population home-interview survey data is a difficult task as the required data collection is cost intensive and time consuming .the high cost of surveys combined with a shift from long term transportation planning to short term strategic with limited budgets researchers to estimate O-D matrices from a variety of other data sources which can be collected easily.

Various data have been used till data in O-D matrix estimation include registration plate survey rods side interviews screen line flows marginal trips totals and link volume counts .Out of these the method based on marginal trip totals and link volume counts are highly promising and received a lot attention. In spit of the large amount of work done on the methods based on the above two scores of information there is still scope for improvement .In this dissertation two methods for O-D matrix estimation are developed. Pone method is based on marginal trips totals and the other is based on link volume counts. First the proposed method related to O-D matrix estimation using marginal trip totals is presented. The O-D matrix estimation using link volume count is discussed later. Before proceeding further it may be noted that this dissertation concentrates more on the first methodology, which forms the primary part of the dissertation. The second methodology is basically an improvement over certain existing methodologies and forms the second part of the dissertation.

One of the primmer issues in the methodologies using marginal trip totals is how well they can incorporate information on trip dispersion while estimating O-D matrices. Some methods like the intervening opportunities model and the gravity model do not have any mechanism by which all types of information on trips dispersion can be incorporated during the O-D matrix estimation process. Other methods like mathematical programming based ones can incorporate a variety of information regarding trip dispersion. However these methodologies necessarily objective functions, which lose all physical meaning in the absence of, target O-D matrices.

Further the type of information used so far to specific trip dispersion is restricted to cardinal information on O-D matrix elements and include target O-D matrix trip length frequency and range on O-D matrix elements. Surprisingly no attempts has been made so far use ordinal information on O-D matrix elements to specify trip dispersion .In this dissertation a methodology is developed which can utilize ordinal information like number of trips between a pair of zones is more than number of trips between another pair zones called as well as cardinal information on O-D matrix elements to specify trip dispersion. In this discretion a methodology is developed which can utilize ordinal, information like number of trips between a pair Of zones is more than the number of trips between another pair zones (called orders) as well as cardinal information on O-D matrix elements as sources of information on trip dispersion .The proposed methodology is based on a mathematical formulation which is constructed such that ordinal information as well as cardinal information can be used to obtain various estimates of O-D matrix elements which satisfy the observed trip dispersion pattern (specified through ordinal/ cardinal information) and marginal trip totals (feasible estimates ) . The methodology unlike earlier mathematical programming based methods avoids the use of an objective function, which loses all physical meaning in the absence of target O-D matrices. Further the type of information used so far to specify trip dispersion is restricted to cardinal information on O-D matrix elements and include target O-D matrix trip length frequency and range on O-D matrix elements. Surprisingly no attempt has been made so far to use ordinal information on O-D matrix elements to specify trip dispersion .In this dissertation a methodology is developed which can utilize ordinal in formation like number of trips between a pair of zones is more than the number of trips between another pair of zones (called orders) as cardinal information on O-D matrix elements as sores of information on trip dispersion. The proposed methodology is based on a mathematical formulation which is constructed such that ordinal information as well as cardinal information can be used to obtain various estimates of O-D matrix elements which satisfy the observed trip dispersion pattern (specified through ordinal/ cardinal information) and marginal trip tools (feasible estimates) .The methodology unlike earlier mathematical programming based methods avoids the use of an objective function which artificially ranks the various feasible estimates .The proposed methodology in effect provides a tool which can be used to sample the feasible search space defined by the marginal trip totals and ordinal and /or cardinal information on O-D elements (alternatively the observed trip dispersion pattern). The methodology bank on repeated sampling of this search space in order to obtain improved and consistent estimates of the O-D matrix elements.

Experiments are conducted to determine the effectiveness of order as well as the effectiveness of the proposed methodology in estimating O-D matrix elements. Studies are also conducted to illustrate the sensitivity of the accuracy of O-D matrix estimates with respect to various influencing factors like the numbers of orders used to specify trip dispersion. Results from these experiments using various different real worlds as well as synthesized O-D matrices confirm that orders are quite effective in improving the accuracy of O-D matrix estimates. Further the idea of sampling the feasible search space used in the proposed methodology is found to be more effective in utilizing orders while estimating O-D matrices than the traditional idea of utilizing an artificial objective function to obtain a unique estimate. Results from the sensitivity analysis also reveal important facts related to the efficient use of information to specify trip dispersion pattern.

The second methodology (which it may be pointed out is not related to the first methodology) developed in this dissertation utilizes as mentioned earlier the link volume counts in estimating O-D matrixes .O-D matrix estimation using link volume counts is based on the relationship between travel demand (O-D matrix) and traffic flow on links of the network. Hence for any methodology in this category .The traffic assignment technique (used to assign the travel demand to obtain traffic flow on links of the network) forms an important constituent of the entire O-D matrix estimation process.

Many methods which use link volume counts in O-D matrix estimation assume that the choice of a route by a traveler is independent of the flows on the links of the network and thus a proportional assignment technique can be used to represent the route choice behaviour of travelers. All-or-nothing and Logic are some of the commonly used proportional assignment techniques. The assumption of proportional assignment technique however is at best valid for uncongested sparse networks. A flow department traffic assignment technique is more appropriate when congestion in the network is significant. In the literature equilibrium assignment technique is suggested for such case .The equilibrium assignment technique is used in two different ways in O-D matrix estimation; (i) solving the mathematical programming (MP) formulation of equilibrium assignment technique for O-D matrix elements and (ii) using equilibrium assignment technique to obtain route choice properties which define link flow constraints. However the methods based on solving the MP formulation of equilibrium assignment technique require a target O-D matrix and flows on all links of the network. These requirements are generally difficult to meet .The methods where route choice properties are obtained equilibrium assignment technique the route choice proportions are not unique and this is not a desirable property. Moreover the debatable assumption that traffic equilibrium exists on a network is also made in this method s. These drawbacks motivated the development of a new assignment technique for use in O-D matrix estimation.

In this dissertation a flow dependent multi-path fuzzy inference based assignment technique is proposed for route choice .The proposed assignment technique is used to obtain the route choice proportions which along with the observed link flows are then used in the O-D matrix estimation. Since the proposed assignment technique is flow dependent the entire O-D matrix estimation process becomes a bi-level optimization problem and is solved iteratively .The positive features of proposed model include (i) incorporation of the dependence of choice off a link on the flow on that link and (ii) the assumption that information on the performance of a link is perceived "vaguely"; further the proposed procedure does not require (i) flows on all links of the network, and (ii) a target O-D matrix .

Experiments are conduced to determine the impact of the choice of an assignment technique on the accuracy of O-D matrix estimates for a given traffic flow pattern. Studies are also conducted to determine the effectiveness of the proposed assignment technique when used to estimate O-D matrices. Results from various experiments on a test network using for different assignment techniques namely (i) incremental capacity restraint technique (ii) Dial's STOCH technique (iii) equilibrium technique and (iv) proposed (fuzzy inference based) technique show that O-D matrix estimates using the proposed assignment technique are good for a variety of flow patterns observed on the network. Open the other hand he other three assignment techniques are less versatile when used in O-D matrix estimation and produce good matrix estimates only when the observed flow pattern satisfy the implicit assumption of the assignment techniques. Results from experiments using larger networks shows that the proposed methodology performs equally well for larger networks.