A demand responsive feeder-system service mandatory and clustered, optional bus stops

Currently, most bus services are ine cient due to their rigid nature. Ser-vice providers are not well-equipped to handle the constantly changing demandfor transportation and lack the tools and the required knowledge to serve thisdemand e ciently. This may result in frustration and client dissatisfaction. How-ever, with the rise of smart cities in the near future, it will be possible to collectdata about the potential passengers, like their requested arrival time and theiractual origin location, in real-time. Service providers will also be able to commu-nicate bus arrival times and other information in real-time to its passengers anddrivers. This will create a two-way communication system between the clientsand the service providers, which o ers a number of opportunities to increase theservice quality.One crucial advantage of traditional transport services (TTS) is their low op-erational costs, arising from their ability to transport large groups of passengerscollectively at a time. Furthermore, TTS are usually e cient in high demandareas as they can su ciently meet the constant demand of all passengers withoutrequiring routes or timetables to be customized. The predictable nature of TTSdue to xed scheduling makes them highly accessible to most commuters. How-ever, the xed nature of TTS can also be viewed as a limitation, since it rendersTTS inadequate in settings where demand for transportation is sparse and con-stantly changing. Consequently, passengers utilizing TTS frequently experiencelong travel times. There is thus a need for Demand Responsive Transportation(DRT) services to deal with this ever-changing demand for transportation. DRTservices are services like a Taxi or Dial a Ride (DAR) services [1]. On the onehand, DRT services operate only when there is a demand for transportation andcan thus meet the passenger's expectations accordingly [2]. On the other hand,DRT is often quite expensive and are not applicable on a mass-scale due to thehighly customizable routing of the vehicles and the high complexity of schedulingthe passenger's requests for a ride. These services also do not o er a solutionto deal with the unknown demand for transportation, e.g. passengers that donot request a ride, because they are unfamiliar with the system, but could still1bene t from such a ride. As such, this research will focus on developing a trans-portation service situated between the TTS and DRT services, integrating thepositive characteristics from TTS as well as from DRT services.In this research, a feeder system is developed. A feeder system transportspassengers from a low-demand area, like a sub-urban area, to a transportationhub, like a city center. All passengers will thus have the same destination, butdi erent origins. This system can have one or multiple bus-lines and within abus-line there will be two sets of bus-stops: mandatory stops, and clustered op-tional stops. The mandatory stops need to be visited by each bus of a certain line.The optional stops can be visited by a bus, but they will only be visited when aclient, with origin within walking distance, needs to be picked-up. This impliesthat the buses can have di erent routes and di erent timetables according to thedemand, while there is still a factor of predictability with the mandatory stops.The mandatory stops, can thus serve as a safety-net for the unknown demand.In this feeder system, potential passengers make a request for transportationto the transportation hub, by stating their current location and their latest arrivaltime. In this rst stage of the research, a static problem and a single line will beconsidered. Here, the requests from the passengers are known beforehand for agiven period of time. To optimize the performance of the system, we develop aMixed Integer Problem (MIP). The MIP is implemented in C++ and solved withthe commercial solver CPLEX 12.9. However, it was found that the model couldnot be solved within a reasonable time for large, realistic instances. Thereforea heuristic approach is developed as well. We implement a Large NeighborhoodSearch (LNS) heuristic and compare the results of the heuristic with those of theexact model. It was found that the heuristic yields gaps between 0.3% and 4%for small instances while computational times are reduced considerably. Evenso, these models and algorithms are still in the early stages of development andfurther research is needed.

Jaar van publicatie:2020