Modeling Ride Sharing Within State Of The Art Travel Demand Models
Corresponding Author: Klaus Noekel, PTV Group
Presented By: Klaus Noekel, PTV Group
New forms of mobility, many based on the principles of shared economy, are rapidly gaining popularity and will likely transform the modal mix in cities within the next five to ten years. This causes concern, because state-of-the-art travel demand models do not represent these modes and may therefore misjudge future benefits of investments into conventional transport. Much research is directed at the design and operation of vehicle and ride sharing systems, but so far little of it is applied within the context of travel demand models. During workshops at the 2015 TRB Annual Meeting, one of the most intensely debated questions was how to include the effects of ridesharing into a conventional travel demand model.
The proposed solution starts from an aggregate four-stage model and explicitly aims to retain as much as possible of the model structure. Ridesharing is introduced as one additional mode at the mode choice level. The required generalized cost skim matrices are produced in four steps:
1) Aggregate demand for OD pairs is disaggregated into a list of trip requests for which origin location, destination location, and desired departure time are randomly drawn from known or assumed distributions.
2) Based on the service specification of the ridesharing service, time windows for departure and arrival time are added to each trip request.
3) The dynamic Dial-A-Ride Problem (DARP) is solved, which optimizes tours for a given fleet of vehicles and matches trip requests to tours so that all requests are served with minimal cost, respecting time window constraints.
4) Generalized cost components are skimmed off the tour plan and aggregated to OD pairs. Optionally, vehicle movements from the tour plan are fed into private transport assignment as a pre-load, and congested travel times from the assignment fed back into the DARP optimizer.
The paper reports on a pilot application in a real travel demand model, using standard travel demand modeling software. A special case is the implementation of step 3) for which a tour optimization algorithm implemented in a library targeted at logistics applications was linked to the travel demand model.