This project aims to enhance the understanding and modeling of traffic flow by developing a new theoretical framework that includes stochastic and non-static dimensions to the traditional traffic "fundamental diagram" (a concept that describes relationships among key traffic variables such as flow, density, and speed). The increasing adoption of connected automated vehicles introduces complexity to traffic dynamics that the current deterministic and steady-state model is not well-equipped to handle. The proposed framework will use empirical trajectory data from human-driven vehicles and those with automation features, such as adaptive cruise control, to develop computational models. These models will capture the stochastic behavior of various types of vehicles. Expected outcomes include better comprehension and description of heterogeneous traffic flow, improved design of automated vehicle control strategies, and smarter traffic management for enhanced throughput, stability, and resilience. 

Selected Paper:

• Stochastic Calibration of Automated Vehicle Car-Following Control: An Approximate Bayesian Computation Approach (Zhou et al. 2023)

This paper presents a stochastic calibration method based on Approximate Bayesian Computation (ABC). This method is applied to calibrate two car-following control models: linear control and model predictive control (MPC). The method is likelihood-free, where the likelihood function is replaced by simulation of the model followed by a comparison with the observed data to approximate the posterior distribution. This structure affords flexibility to calibrate posterior joint distributions of complex models, even those without analytical closed forms such as MPC. Two experiments were conducted to evaluate how well the proposed method reproduces:(i) marginal and joint distributions of model parameters, using synthetic data and (ii) vehicle trajectories (acceleration, speed, and position), using field data involving two commercial adaptive cruise control (ACC) systems. The results showed that the ABC method can reproduce marginal and joint distributions reasonably well, particularly for the simple linear controller. The method is also able to generate meaningful marginal and joint distributions for the complex MPC-based controller, which was previously infeasible. The method can also robustly characterize the commercial ACC behavior at the trajectory level, better than a deterministic solution, and reveal which controller fits better. The evaluation results using field data suggest that the commercial ACC vehicles (Civic and Prius in this study) tend to accelerate and decelerate mildly and that the simple linear controller best describes their behavior.

The increasing adoption of electric vehicles (EVs) provides an opportunity to create a new type of infrastructure system that can improve the resilience of our society. This project proposes a novel "energy sponge service" that would use EVs to store and transmit energy between transportation and power systems, helping to stabilize operations in both systems. The energy sponge service would be based on an EV-sharing cloud. This cloud would manage a fleet of EVs, and it would be able to dynamically allocate these EVs to provide mobility or energy services, depending on the needs of the system. The cloud would also be able to predict demand fluctuations in the transportation and power systems, and it would use this information to optimize the operation of the energy sponge service. The project advances the state of the art in harnessing interdependency to enhance system resilience and enable technology for "energy sponge" systems.

Selected Papers:

• Charging an electric vehicle-sharing fleet (He et al. 2021)

This research tackles the challenges of charging operations for Electric Vehicle (EV) sharing services using a queuing network model, drawing on real-world data from car2go's failed operation in San Diego. The authors propose a proactive charging strategy or charging EVs at the 40% energy level to boost profits by over 15%. They also emphasize the importance of concentrating charging resources at specific locations, ensuring adequate charger availability, especially when partnering with public charging networks, and raising charging power to overcome resource limitations. Moreover, they suggest that extending the per-charge range or using unmanned repositioning can increase profitability. The study provides valuable insights into EV sharing operations and offers planning and operational strategies beneficial for practitioners in the field.

• Joint infrastructure planning and fleet management for one-way electric car sharing under time-varying uncertain demand (Hua et al. 2019)

This work proposes an innovative framework to deploy a one-way Electric Vehicle (EV) sharing system that serves an urban area. For the first time, long-term infrastructure planning (charging station location and fleet distribution) and real-time fleet operations (relocation and charging decisions) are jointly optimized under time-varying uncertain demand. This substantially advances EV sharing system efficiency and yields a practical management strategy. We propose a multistage stochastic model to address the critical challenge of time-varying uncertain demand. An accelerated solution algorithm is developed to conquer the curse of dimensionality in integer infrastructure planning decisions. Meaningful insights are delivered through hypothetical numerical experiments and a realistic case study with EV sharing service in the New York City.

• Incentivized self-rebalancing fleet in electric vehicle sharing (Wu et al. 2021)

With the rising need for efficient and flexible short-distance urban transportation, more vehicle sharing companies are offering one-way car-sharing services. Electrified vehicle sharing systems are even more effective in terms of reducing fuel consumption and carbon emission. In this article, we investigate a dynamic fleet management problem for an electric vehicle (EV) sharing system that faces time-varying random demand and electricity price. Demand is elastic in each time period, reacting to the announced price. To maximize the revenue, the EV fleet optimizes trip pricing and EV dispatching decisions dynamically. We develop a new value function approximation (VFA) with input convex neural networks (ICNNs) to generate high-quality solutions. Through a New York City case study, we compare it with standard dynamic programming methods and develop insights regarding the interaction between the EV fleet and the power grid.