Agent Based Transport Modelling Applied to One Way Car Sharing

Agent Based Transport Modelling Applied to One Way Car Sharing


Helen Porter, Peter Davidson Consultancy, Peter Davidson, Peter Davidson Consultancy, Rob Culley, Peter Davidson Consultancy


We describe the development of travel itineraries for a synthetic population of agents and the optimisation of the operational system to meet this demand efficiently.


An innovative one way electric car sharing system, ESPRIT, allows vehicles to be coupled into road trains to improve operational efficiency and overcome problems with vehicles accumulating away from areas of demand. It is anticipated that by lowering the costs associated with vehicle redistribution the overall operating costs are reduced and individual vehicles are used more efficiently. Activity-based modelling is necessary for the evaluation of innovative and disruptive mobility services and policies effects, because it is able to provide a precise representation, with fine spatial and temporal granularity, of the population under study. This allows for examples to realistically take into account the service access time, which is fundamental in the customer’s choice but could not be accurately captured by traditional models. Similarly, the ability to detect chains of trips is fundamental when assessing who can realistically use shared transportation modes. This paper applies a new approach for an actvity based framework which implements agent based modelling within transport demand models as well as describing advanced methods for optimising the location of car sharing infrastructure using mathematical algorithms.
Household and travel diary data was used to develop the synthetic population for Lyon of about 1.4 million agents. The characteristics of the agents living in each zone is consistent with a set of control variables, in this case distribution of household sizes, household type and car ownership category. The demand model includes mode and destination choice, the coefficients for which were estimated from revealed preference travel diary data using simultaneous nested logit estimation. The demand model uses logit based probabilities for decision choice, but takes individual agents rather than origin destination matrices as input. We describe the challenges of computing with a large number of individual agents rather than conventional origin-destination matrices, and how this approach was exploited to dramatically improve run times. We also discuss the advantages of this approach for more detailed market segmentation whereby each agent can have several attributes and their choices can be followed individually to check rationality.

Individual decisions from the demand model are translated into a set of travel itineries in xml format, one for each agent, which are passed to the simulation model. The itineries give information about the origin, destination and mode for each travel leg. Mulit-mode journeys, for example where car sharing is combined with a Public Transport trip are split into different journey legs to allow the simulation model to interpret the itineries.
Redistribution of vehicles can be achieved by encouraging users to travel to, for example, travel to an alternative destination location or to tow a second empty vehicle. These choices are constructed by an “offer” model. This offer model is only meaningful in an agent based context where individual vehicles are considered. One of the most widely used general-purpose, state-of-the-art agent-based transport simulators available for transport researchers is MATSim ( Thus, we have developed a modular, expendable and easily customisable model of car sharing systems in MATSim, which considers all the main characteristics of existing and future car sharing services. Key features booking services, fleet redistribution and customer management, but also support for electric vehicles and smart charging policies.

Several problems related to the deployment and operations of car sharing systems have yet to be fully addressed. For instance, station-based car sharing systems, in which users are required to pick up and drop off shared vehicles only at dedicated stations, require significant capital investments for deploying the necessary station infrastructure, which undermines the economic viability of the car sharing service. Thus, when deploying a station-based car sharing system it is crucial to strike the right balance between the costs for the operator and the quality of service provided to the customers. To this end, we have defined an optimisation problem for the deployment of its stations. The goal of this problem is to find the minimum cost deployment (in terms of number of stations and their capacity) that can guarantee a pre-defined level of service to the customers (in terms of probability of finding an available car/parking space). Preliminary results obtained using realistic traces of pickup and drop-off events from existing car sharing operators, show that the proposed solution is able to strike the right balance between cost minimisation and quality of service.

A model for Lyon has been developed which we describe here, showing the results of the model and the implications for the ESPRIT system. This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 653395.


Association for European Transport