Blog

Car2work: A shared mobility concept to connect commuters with workplaces

Robert Regué

Over the last decade there has been a surge of short-term rental schemes such as ZipCar, Car2go or Drive Now, among others. Furthermore, ride hailing apps such as Lyft, Uber or SideCar are becoming more popular. These new mobility concepts are focused on improving mobility in urban areas, with rather large population densities, providing an alternative to public transit and private cars. However, these concepts often fail to serve the regular transit commuter in sprawled areas that need a ride to/from transit hubs. Car2Work is a concept that intends to address this issue. Car2Work is a clean hybrid system falling between a traditional vanpool and a carsharing system, leveraging the advantages of both: commuters, if the system matches them, have a ride home guaranteed, and the vehicles are shared in a short-term rental fashion while idling on the stations.

In this work, we have modeled the system under a simulation framework that at its core relies on a variation of the multi-hop peer-to-peer ride-matching problem.  Commuters announce their trips and the model finds the optimal schedule, including transit connections. A variety of scenarios can be modeled, including the use of autonomous vehicles, different fleet splits, multiple transit modes, or varying commuter preferences. The impacts of a wide-scale deployment can also be modeled; for example, in assessing the impacts on the current transportation network of a Car2Work implementation along the Metrolink Inland Empire Orange County (IEOC) line.

More info in here.

On May 1st at 10AM, Dr. Vrugt will give a seminar on:

The iterative research cycle – On detection of model structural errors

In the past decades, Bayesian methods have found widespread application and use in environmental systems modeling. Bayes theorem states that the posterior probability, P(H|D) of a hypothesis, H is proportional to the product of the prior probability, P(H) of this hypothesis and the likelihood, L(H|D) of the same hypothesis given the new/incoming observations, D, or P(H|D) is proportional to the product of P(H) and L(H|D). In science and engineering, H often constitutes some numerical simulation model, D = F(x) which summarizes using algebraic, empirical, and differential equations, state variables and fluxes, all our theoretical and/or practical knowledge of the system of interest, and x = \{ x_1,…,x_d } are the d unknown parameters which are subject to inference using some data, D of the observed system response. The Bayesian approach is intimately related to the scientific method and uses an iterative cycle of hypothesis formulation (model), experimentation and data collection, and theory/hypothesis refinement to elucidate the rules that govern the natural world. Unfortunately, model refinement has proven to be very difficult in large part because of the poor diagnostic power of residual based likelihood functions. In this talk I will introduce the elements of a diagnostic approach to model evaluation and improvement.

This diagnostic approach uses signature behaviors and patterns observed in the input-output data to illuminate to what degree a representation of the real world has been adequately achieved and how the model should be improved for the purpose of learning and scientific discovery. Several case studies are used to illustrate the proposed methodology.

Dr. Vrugt is an assistant professor at the Civil & Environmental Engineering department that leads a research group that combines numerical modeling (deterministic, stochastic) and/or analytic solutions with small and large-scale measurement (direct and indirect observations), and inverse modeling (parameter estimation, data assimilation, model averaging, etc.) to improve theory, understanding and predictability of complex Earth systems. We engage in all aspects of the iterative research cycle and regularly develop new numerical, computational, statistical, and optimization approaches to reconcile complex system models with observations for the purpose of learning and scientific discovery and, thereby, enhancing the growth of environmental knowledge. We use distributed computing to permit inference of CPU-intensive forward models.

For more information, please visit http://faculty.sites.uci.edu/jasper/

 

Short-term Traffic Flow Prediction with Mixture Autoregressive Model

Zhe Sun (Jared)

This study aims to address the problem of short-term traffic prediction on freeways with a mixture auto-regressive model (MAR). Short-term traffic prediction plays an important role in the traffic control system and provides valuable information to commuters and decision makers. It is known that, on urban freeways, traffic flow is mainly contributed by the commute trips and exhibits transitions between on and off-peak. However, most of the existing short-term prediction models ignore the transition of traffic modes and thus mischaracterize the traffic dynamics as being time invariant.  Through simulation and empirical study on the real dataset, we show that the proposed mixture model is able to explain the heteroscedasticity in traffic flow data and explicitly account for the switching of modes.

Financing, Timing, and Capacity of a New Intercity Highway under Demand Uncertainty: The BOT Case

Ke Wang (Kevin)

The 2013 ASCE infrastructure report card gave roads nationwide a “D”. While capital investments reached $91 billion annually for all levels of government, this falls way short of the $170 billion that FHWA estimates are needed annually to significantly improve road conditions and performance. Given the public’s reluctance to increase revenues for transportation, it is urgent to revisit Public-Private Partnerships (PPP) to attract capital and engineering expertise from the private sector. This study proposes a real-options framework for analyzing public-private partnerships that could be used to fund roads; it includes demand uncertainty, endogenous tolls, endogenous road capacity, and accounts for the lag between the beginning of a project and its completion. The competition between the new and the existing road is modeled explicitly, and traffic congestion is accounted for using a BPR function.

It is well known that applying a standard cost-benefit analysis (which is static and deterministic) to uncertain projects could be highly misleading because it ignores both uncertainty and irreversibility. This study derives analytical results for the optimal timing and capacity of a new Build-Operate-Transfer (BOT) highway project between two cities when the demand between these cities follows a reflective geometric Brownian motion (RGBM). A numerical illustration with realistic parameter values shows that there is a monotonic relationship between demand volatility and investment threshold: ignoring demand uncertainty will lead to invest prematurely. Moreover, the value of the option to defer a BOT project can be substantial.

This week student seminar presenter is Kyungsoo Jeong, PhD Candidate in Transportation Engineering.

California Vehicle Inventory and Use Survey Pilot Study

The purpose of Cal-VIUS is to provide updated and relevant detailed truck inventory and activity data from trucks that have operations within the State of California to government agencies, businesses and academia who are interested in commercial vehicle activity and freight transportation. It will serve an update to the California portion of the national VIUS effort, which was last conducted in 2002. The pilot study was conducted to design the sampling framework and the survey instrument, and then provide guidelines for the actual Cal-VIUS. A stratified random sampling is adopted to capture key population characteristics for commercial vehicles with a variety of attributes. The survey instrument is an online questionnaire with questions designed to gather key data needed by stakeholder agencies.