Accounting for driver-passenger matching decisions in a ridesharing simulation platform

Abstract

10 This paper presents a new ridesharing simulation platform that accounts for dynamic driver 11 supply and passenger demand, and complex interactions between drivers and passengers. The 12 proposed simulation platform explicitly considers driver and passenger acceptance/rejection 13 on the matching options, and cancellation before/after being matched. New simulation events, 14 procedures and modules have been developed to handle these realistic interactions. The 15 capabilities of the simulation platform are illustrated using numerical experiments. The 16 experiments confirm the importance of considering supply and demand interactions and 17 provide new insights to ridesharing operations. Results show that larger matching window 18 could have negative impacts on overall ridesharing success rate. These results emphasize the 19 importance of a careful planning of a ridesharing system.

21 Keywords: ridesharing; simulation; driver supply; passenger demand

1 1. Introduction

2 Innovative shared mobility, namely ride-sourcing services, have reshaped our urban 3 transportation. Ride-sourcing companies, such as Uber, Lyft, and Didi, provide convenient 4 mobility services with lower fares, by utilizing drivers’ own vehicles instead of company fleets 5 to provide services (Wang and Yang, 2019). One type of these services is ridesharing, in which 6 peer drivers serve more than one passenger in each ride. Ridesharing services potentially can 7 reduce vehicles kilometer traveled (VKT), compared to other service types. This is because 8 peer drivers in ridesharing, who are assumed to perform activities other than only pick up and 9 drop off passengers (as is the case of dedicated ride-hailing drivers), have their designated 10 destinations and do not cruise in the network (Wang and Yang, 2019). Despite various policies 11 implemented for encouraging ridesharing (e.g., HOV lanes), the market share of ridesharing is 12 still relatively low (Hensley et al., 2017). 13 From a planning perspective, designing and evaluating the effectiveness (expected “real-world” 14 performance) of ridesharing systems is challenging. Many studies addressed the operational 15 decisions of service providers, specifically, dynamic ridesharing driver-passenger matching 16 problems (e.g., Agatz et al., 2012; Alonso-Mora et al., 2017). 17 In a realistic ridesharing setting, peer drivers and passengers may also make various matching 18 decisions that affect ridesharing operations. For example, as new passenger requests continue 19 to emerge during ridesharing, existing schedules might be modified to serve these new 20 passengers. Consequently, for passengers already waiting for pickups, they may cancel their 21 trips due to delayed pick-up after being assigned a driver (He et al., 2018; Wang et al., 2020). 22 Similarly, for drivers and on-board passengers, they may reject the assigned matchings due to 23 extra detours (Chu et al., 2018; Rosenblat et al., 2017). Other decisions could be passengers 24 cancelling their trips because of a long wait before being matched with a driver (Wei et al., 25 2020), and passenger reorder and rebooking after cancellation. In terms of evaluating a 26 ridesharing system, if these possible driver and passenger actions were not considered in the 27 ridesharing models, the system performance could not be properly estimated. We summary the 28 existing ridesharing models in Table 1.

30 Table 1 A comparison between the proposed and existing ridesharing models.

Dynamic Cancellation Cancellation Multiple Matching option Literature supply before being after being passengers acceptance/rejection demand matched matched

Network assignment models Xu et al. (2015)  Di and Ban (2019)  Ma el al. (2020)  Li et al. (2020)  Wei et al. (2020)  

Dynamic Cancellation Cancellation Multiple Matching option Literature supply before being after being passengers acceptance/rejection demand matched matched

Market equilibrium models He et al. (2018)  Wang et al. (2020)  Ke et al. (2020)  Simulation models Djavadian and Chow (2017a,   b) Wang et al. (2017)    Beojone and Geroliminis    (2021) Shen et al. (2018)   Nahmias-Biran et al. (2019)  Thaithatkul et al. (2019)   Linares et al. (2016)    Nourinejad and Roorda    (2016) This paper     

2 As shown in the table, there are still gaps in the literature for developing a comprehensive 3 ridesharing simulation platform, with explicit considerations of driver and passenger matching 4 decisions. We propose to develop a novel simulation platform to capture these complex 5 dynamic interactions, to provide new insights, and to handle more realistic scenarios.

6 2. Methodology

7 The proposed ridesharing simulation platform captures the complex dynamic interactions 8 between passengers and drivers in a ridesharing system, and it is able to handle: a) Dynamic 9 passenger demand and driver supply; b) Acceptance/rejection on matching options; c) Order 10 cancellation and no-show. We outline the overall ridesharing simulation process in Figure 1.

2 Figure 1 Flowchart for ridesharing simulation process

3 Our ridesharing simulation platform primarily incorporates the driver-passenger matching 4 decisions in the Ridesharing Matching Event and the Agent Staying/Leaving Decision 5 Making Event. In the following subsections, these two simulation events, and the 6 corresponding modules will be explained in detail.

1 2.1. Ridesharing Matching Event.

3 Figure 2 Sequence diagram for Ridesharing Matching Event routine 4 During Ridesharing Matching Event execution, the Passenger and Driver agents first 5 synchronize their matching status with the Service Providers. Then, the Service Providers 6 perform ridesharing matching between these Drivers and Passengers, and provide information 7 on the matching options. 8 Upon receiving these matching options, Driver and Passenger agents compare the utilities 9 among various options, and choose to accept/reject the provided options. Later, after Service 10 Provider receiving these responses, and performing necessary modifications on the matching 11 schedules, these finalized matchings are sent to the Drivers and Passengers. 12 The Driver is set to perform the finalized matching, by register one of the possible following 13 events: Pickup Event, Drop-Off Event, Agent Termination Events, or Re-schedule Event. 14 Finally, if there is any remaining event in the event queue, another Ridesharing Matching Event 15 will be scheduled at next matching window. 16 In the following paragraphs, we illustrate examples of Agent Matching Response Module and 17 Matching Option Responses Handling Module.