A Generalization of the MATSim simulation process to Simulate the influence of Social Contacts Spatial Distribution on Mobility Behavior

Abstract

In their daily life, individuals are frequently involved in joint decision making — situations where several individuals have to agree on the actions they will undertake to achieve a joint outcome. Examples in the context of mobility behavior include intra-household task allocation, intra-household vehicle allocation, choice of the time and venue for a dinner with friends or traveling together in the same private vehicle. In addition to being necessary to predict joint travel and car occupancy, it has been hypothesized that considering explicitly this kind of joint decision process for the case of leisure activities planning might help to improve the forecasts for the choice of the leisure destination, due to the often social nature of such activities (Axhausen, 2007). Evidences of the influence of social contacts meetings on activity durations, trip length, or type of activities could be found (Stauffacher et al., 2005; Carrasco and Habib, 2009; Habib and Carrasco, 2011; Moore et al., 2013). This has motivated the inclusion of social networks in simulation frameworks. For instance, (Hackney, 2009) experimented with the co-evolution of social networks and travel patterns in MATSim — without joint decision processes. Other examples include joint decision making (Han et al., 2011; Ronald et al., 2012; Ma et al., 2011, 2012). This research aims at including social behavior — including joint decision making and coordination — in a multi-agent transport simulation in a meaningful way. To this end, the MATSim simulation framework has been extended. MATSim provides a coevolutionary algorithm to search for a user equilibrium over daily plans. The process was modified to shift the solution concept: instead of searching an invidual-based equilibrium, it searches a state without blocking coalitions: a state where no group of agents can all improve the utility they get from their day by changing their plans together, including the posssiblity of joint trips or activities. Of course, adding the possibility to perform joint activities, including joint location choice, increases a lot the size of the search space. Different possible techniques to cope with this were experimented, and are discussed, in terms of a tradeoff between computationnal performance, quality of the results and code maintainability.