#71 Data-driven Network Models for Analyzing Multi-modal Transportation Systems
- Principal Investigator
- Sean Qian
- Status
- Completed
- Start Date
- July 1, 2017
- End Date
- June 30, 2018
- Project Type
- Research Applied
- Grant Program
- FAST Act - Mobility National (2016 - 2022)
- Grant Cycle
- 2017 Mobility21 UTC
- Visibility
- Public
Abstract
We propose to establish a new theory and open-source tool for modeling both passenger and vehicle flow in a sophisticated transportation network. The model takes input of data collected from various source, and models individual travel activities and behavior on roadway systems, transit systems and parking systems. The multi-modal network model is the key to systematic planning and operations of transportation infrastructure. Planning, operational strategies and policies can be fully examined in the network model in terms of system delays, reliability, vehicle-miles traveled (VMT), fuel consumption and emissions. This research will use real-time and historical roadway, parking and transit data, in both Philadelphia and Pittsburgh Region, to test and validate the network model. This research can ultimately help the regional MPOs and DOTs quantitatively analyze and optimize various management strategies for sustainable mobility.
Description
We propose to establish a new theory for modeling both passenger and vehicle flow in a sophisticated transportation network. The model takes input of data collected from various source, and models individual travel activities and behavior on roadway systems, transit systems and parking systems. The multi-modal network model is the key to systematic planning and operations of transportation infrastructure. Planning, operational strategies and policies can be fully examined in the network model in terms of system delays, reliability, vehicle-miles traveled (VMT), fuel consumption and emissions. We will interview regional MPO and RPO modeling groups including the Southwestern Pennsylvania Commission (SPC), and Delaware Valley Regional Planning Commission (DVRPC) to better understand their needs related to improving their regional planning models. MPO's are required by law to have and maintain a regional planning model, which describes the region's travel demand and traveler route choices. The MPO models are able to estimate this, but many MPO models currently rely on ADT (average daily traffic) data, which does not directly capture the time-of-day demand and travel behavior. Those network models do not consider multiple transportation modes as a whole, and do not reproduce the actual observation in the network. The current models can be expanded by adding finer temporal granularity through data that already exists from sensors. Over the last decade, new technologies and innovations in transportation systems have produced massive amounts of data, which has enabled us to better monitor, evaluate and manage our transportation systems. The rich data from various sources provides unprecedented opportunity for the transportation industry to understand travel behavior and to propose efficient management strategies. For instance, traffic counts and speed measurements are collected years along at the resolution of 5 minutes and at various locations in the network. How can those day-to-day high-resolution data help estimate the travel behavior and predict traffic flow? The scope of this research is to tackle those problem from the perspective of probabilistic choice theories by utilizing large amount of data. Many MPOs also maintain a Congestion Management System (CMS) will collects and analyzes information to determine congested highways and planning needs to address these issue. These CMS systems are also a source of information that will be reviewed. We propose innovative data-driven methodology to integrate various sources of traffic data into the process of improving and integrating the planning models into the integrated multi-modal transportation systems. The core of dynamic network analysis linking supply and demand models lies in traffic flow models and travel behavioral models. Travel behavioral models essentially examine travelers' choices in departure time, routes, parking locations, and modes. Traffic flow models capture the spatio-temporal flow prorogation in the network. Behavioral models and flow models are well connected as they are inter-dependent. Travelers' choices are dependent on their respective traveling experience quantified by the individual utility (also known as generalized travel cost). On the other hand, a realization of travelers' choices leads to a consequent spatio-temporal flow which largely determines individual's utility. Oftentimes a critical question is: would there exist a solution of individuals' choices such that the resultant network flow is consistent with the travel times/cost that those choices are made upon. If so, how reliable is this solution in terms of predicting next-hour traffic flow (real-time) and day-to-day traffic flow (long term)? Task 1. Develop an on-line network model for within-day nonrecurrent traffic management. The on-line network model adopts the historical traffic demand and their pre-scribed route/departure time choices. Furthermore, the real-time model receives real-time traffic data feeds (such as real-time INRIX speed data and counts data) and calibrates the en-route route choices in the real time, corrects the forecast of variation of recurrent flow, and the forecast of incident-induced traffic congestion in the next hour. For the purpose of efficient traffic management, the on-line BEATS also computes the optimal traffic diversion ratios for pre-determined detour routes. Task 2. Develop a generalized statistical network model for day-to-day recurrent traffic management. This task develops a generalized statistical network model that estimates probability distributions of link and path flows for the recurrent traffic. The generalized statistical model generalizes classical deterministic User-Equilibrium (UE) or Stochastic UE traffic assignment models to characterize the statistical features of varying Origin-Destination (O-D) demands, link/path flow and link/path costs. Optimal decisions on roadway design, operations and planning can be made using estimated link/path flow distributions from the network model. It integrates three types of variations, demand variation, route choice variation (by travel costs perception error and by individual random choices), and measurement errors. We also build the Multivariate Analysis of Variance (MANOVA) framework for constructing hypothesis tests on traffic recurrence under the framework of BEATS. The generalized statistical network model can also be understood, from the day-to-day perspective, as a Markov Chain. On each day, each individual traveler decides what time to leave home and which route to take, namely the within-day choices. The within-day decision is made upon his perception or anticipation of the generalized utility of this trip that this traveler learns from day to day. In other words, the within-day choices are adjusted from day to day as travelers cumulate their traveling experience and knowledge. Every traveler attempts to minimize his expected generalized travel cost and its variation. The combined day-to-day and within-day travel choices are naturally a Markov Chain of order m where m is the number of days that it takes to cumulate sufficient traveling knowledge for decision making. Existing literature has examined the day-to-day route choice adjustment by formulating it into a stochastic process, in particular Markov Chain. However, transportation system dynamics within a day has been overlooked, which is unable to capture time-varying queuing spillover in the network. The stochastic user equilibrium has not been investigated in the real-world networks. This research aims to demonstrate how individual travelers make departure time and route choices from day to day, and as a result, how the transportation system evolve by time-of-day and from day-to-day. We shall model travel choices through Markov Chains and propose efficient algorithms to test the uniqueness of convergence of recurrent travel choices in large-scale real networks. Task 3. Develop a generalized model choice model among roadway, transit and parking systems We will integrate three networks, roadway, transit and parking networks. The multi-modal network model embeds a nested-logit-based modal split model among three transportation modes, the public transit, solo-drivers and carpoolers. It models the choice probabilities for each single user in the network. It also adopts multi-class LWR model to capture the dynamics of mixed vehicles and dynamic queuing on the network. The spatio-temporal multi-modal traffic flow can then be formulated in a Variational Inequality (VI) problem and solved efficiently. Task 4. Case studies for the Philadelphia and Pittsburgh Metro Areas. The Philadelphia and Pittsburgh Metropolitan Areas are traffic data rich comparing to other metropolitan areas in the U.S. Various data sets in these two regions, including traditional traffic sensors (loops, cameras, etc.) and cutting-edge sensors (Bluetooth, GPS probe, parking, etc.), are available and have been archived for a decade. The rich data sets allow us to learn travelers’ behavior accurately and develop an in-depth understanding of non-recurrent traffic in large-scale networks. In the Philadelphia area, I-95 is currently being reconstructed. The reconstruction narrows lane widths, which leads to heavy traffic congestion during peak hours. Delaware Valley Regional Planning Commission (DVRPC), PennDOT and the city are looking at ways to implement various strategies to achieve Integrated Corridor Management to improve the efficiency, reliability and safety of this I-95 corridor. The reconstruction work will involve considerable lane closures, detours and other traffic diversions. The I-95 reconstruction presents a great opportunity to conduct research on impact of critical infrastructure closures and non-recurrent traffic management. This research will using real-time INRIX data, historical counts and INRIX data, real-time and historical parking data, real-time and historical transit data to test and validate the two models proposed in Tasks 1-3. This research can ultimately help the regional MPOs and DOTs quantitatively analyze and optimize various management strategies for sustainable mobility. .
Timeline
2 years
Strategic Description / RD&T
Deployment Plan
Work with Southwestern Pennsylvania Commissions and Delaware Valley Regional Planning Commissions
Expected Outcomes/Impacts
1. An online and day-to-day multi-modal network modeling framework implemented in the two test beds: Philadelphia and Pittsburgh Metro Areas. 2. Experimental results and guidelines for incorporating large-scale data (historical and real-time) to facilitate the model calibration and validation processes. 3. Open-source tools implementing both online and day-to-day network models. The tools, programs and data will be made available in the public domain.
Expected Outputs
TRID
Individuals Involved
| Name | Affiliation | Role | Position | |
|---|---|---|---|---|
| seanqian@cmu.edu | Qian, Sean | CEE/Heinz | PI | Faculty - Untenured, Tenure Track |
Budget
Amount of UTC Funds Awarded
$70000.00Total Project Budget (from all funding sources)
$70000.00Documents
Match Sources
No match sources!
Partners
No partners!