Dissertations and Theses

Date of Award

2022

Document Type

Dissertation

Department

Civil Engineering

First Advisor

Camille Kamga

Second Advisor

Kyriacos Mouskos

Keywords

TNDP

Abstract

This dissertation addressed a subset of the Transit Network Design Problem (Transit-NDP), where the main evaluation parameter is the impact of various transit projects – Bus Lines, Tram, Light Rail, Rail – on the Network Travel Time (NTT) of the vehicles. The NTT is estimated using the static four-step travel demand model (trip generation, trip distribution, modal split, and traffic assignment). A set of five metaheuristic models were developed and implemented for this specific Transit-NDP, namely, Simulated Annealing (SA)-VISUM, Simulated Annealing-Random Forest Regression Model (SA-RF), Tabu Search (TS)-Random Forest (TS-RF), SA-RF-TS-VISUM, SA-RF-VISUM. These models were evaluated using two test networks (Halle Salle with 6521 OD pairs and Karlsruhe with 527,076 OD pairs). An enumeration and sampling study provided insights into the impact of various combinations of transit projects on the NTT. At the same time, it also served as the dataset that the RF model was based on.

A set of demand-budget combinations were tested for each network – Demand (100%, 120%, 140%, 160%, 180%) and Budget (0%, 20%, 40%, 60%). The best solutions found in the Halle Salle network ranged between 100% demand, 48.23% NTT reduction to a 180% demand, 53.47% NTT reduction. Correspondingly, the best solutions found in the Karlsruhe Network ranged between 100% demand and 39.52% NTT reduction, whereas 180% demand resulted in a 53.99% NTT reduction. The principal conclusion for implementing these metaheuristics is that the utilization of an ensemble metaheuristic model encompassing a set of metaheuristics will be most beneficial as no dominating metaheuristics were found to outperform the others. The budget-demand sensitivity analysis produced the following significant outcomes: 1) the best solution found at higher budget levels does not necessarily include transit projects that were included at lower budget levels, 2) the best solution found at higher demand levels does not necessarily include transit projects that were included at lower demand levels. 3) The utilization of the RF model as a surrogate model to estimate the NTT based on the generated dataset of multiple tests runs – instead of the VISUM model - in synergy with the SA and TS resulted in reducing the corresponding computational time in the order of 0.07 hours (average) and a standard deviation of 0.01 hours for SA-RF, and 1.70 hours (average) and a standard deviation of 0.14 hours for TS-RF. 4) The metaheuristic models systematically found better solutions than the corresponding sampling method providing confidence in the utilization of metaheuristics to find “near-optimal” solutions. Under the Halle Salle network, almost all the metaheuristics implemented found solutions within 0.02% to 0.03% of the best solutions found by other metaheuristics under all tests conducted.

A significant outcome of this thesis is the following: Given a transportation network with a set of candidate transit projects being proposed by the major stakeholders, this research develops and implements an ensemble of models comprised of 1) a set of random test run combinations of candidate transit projects, and 2) a set test runs of various metaheuristics, to produce a set of the “best solution states,” where the main parameter is the corresponding NTT. This ensemble of models, which is envisioned to become self-calibrated utilizing the substantial traffic flow and control and infrastructure data, could be implemented and produce the corresponding results in a few weeks or months, depending on the network size. This ensemble of models will be able to provide the primary stakeholders with the impact of the best solutions found based on the various demand/budget combinations with regard to the NTT parameter in a consistent and systematic manner.

This thesis explored a set of widely used metaheuristics such as Simulated Annealing, Tabu Search, and Random Forest Regression. The successful implementation – both in finding good solutions and computational time - of these metaheuristics points to the development and implementation of additional metaheuristics, such as genetic algorithms and neural networks that are also utilized in the field of Artificial Intelligence (AI) extensively. In addition, they provide confidence in establishing a continuously updated AI-based model to evaluate a set of transportation network projects where additional parameters other than the NTT should be taken into consideration, as the NDP is a multi-objective problem.

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.