Eco-Mobility-on-Demand Service with Ride-Sharing

Abstract

Connected Automated Vehicles (CAV) technologies are developing rapidly, and one of its more popular application is to provide mobility-on-demand (MOD) services. However, with CAVs on the road, the fuel consumption of surface transportation may increase significantly. Travel demands could increase due to more accessible travel provided by the flexible service compared with the current public transit system. Trips from current underserved population and mode shift from walking and public transit could also increase travel demands significantly. In this research, we explore opportunities for the fuel-saving of CAVs in an urban environment from different scales, including speed trajectory optimization at intersections, data-drive fuel consumption model and eco-routing algorithm development, and eco-MOD fleet assignment. First, we proposed a speed trajectory optimization algorithm at signalized intersections. Although the optimal solution can be found through dynamic programming, the curse of dimensionality limits its computation speed and robustness. Thus, we propose the sequential approximation approach to solve a sequence of mixed integer optimization problems with quadratic objective and linear constraints. The speed and acceleration constraints at intersections due to route choice are addressed using a barrier method. In this work, we limit the problem to a single intersection due to the route choice application and only consider free flow scenarios, but the algorithm can be extended to multiple intersections and congested scenarios where a leading vehicle is included as a constraint if an intersection driver model is available. Next, we developed a fuel consumption model for route optimization. The mesoscopic fuel consumption model is developed through a data-driven approach considering the tradeoff between model complexity and accuracy. To develop the model, a large quantity of naturalistic driving data is used. Since the selected dataset doesn’t contain fuel consumption data, a microscopic fuel consumption simulator, Autonomie, is used to augment the information. Gaussian Mixture Regression is selected to build the model due to its ability to address nonlinearity. Instead of selected component number by cross-validation, we use the Bayesian formulation which models the indicator of components as a random variable which has Dirichlet distribution as prior. The model is used to estimate fuel consumption cost for routing algorithm. In this part, we assume the traffic network is static. Finally, the fuel consumption model and the eco-routing algorithm are integrated with the MOD fleet assignment. The MOD control framework models customers’ travel time requirements are as constraints, thus provides flexibility for cost function design. At the current phase, we assume the traffic network is static and use offline calculated travel time and fuel consumption to assign the fleet. To rebalance the idling vehicles, we developed a traffic network partition algorithm which minimizing the expected travel time within each cluster. A Model Predictive Control (MPC) based algorithm is developed to match idling fleet distribution with the demand distribution. A traffic simulator using Simulation of Urban MObility (SUMO) and calibrated using data from the Safety Pilot Model Deployment (SPMD) database is used to evaluate the MOD system performance. This dissertation shows that if the objective function of fleet assignment is not designed properly, even if ride-sharing is allowed, the fleet fuel consumption could increase compared with the baseline where personal vehicles are used for travel.