Exploring the effects of mixed request schemes for demand-responsive feeder services

Modern day communications mean travellers can request transport options, such as taxis and seats on demand-responsive vehicles, on the fly without having to plan ahead. However, last minute requests for transport could be inefficient for the operator. This paper explores the effects of mixed book-ahead/immediate request schemes in the context of a feeder service. Demand responsive transportation (DRT) combines aspects of both buses and taxis: passengers travel together, but not necessarily to or from the same locations. A simple form of demand-responsive service is that of a feeder bus, permitting passengers to connect to a mass transit service such as a train. A DRT service can run by getting passengers to book ahead or by allowing requests to arrive at the last minute, however the performance of the system may differ under different mixes of these requests. The ratio of the immediate (last minute) requests to total requests is known as the degree of dynamism. The impact of the degree of dynamism is measured by the total vehicle kilometres travelled (VKTs), the success rate (the ratio of requests that actually got served) and the waiting time for customers. This paper simulates a DRT service using an event-based model where requests arrive during a simulated day for booking of both book-ahead and immediate requests. A booking system assigns the request to an appropriate time slot if possible and then an optimiser finds efficient routes to pickup and drop-off the customers at the required locations. Finally a dispatching system notifies and updates the driver of the vehicle with the current route. To optimise the route, the customer demands are modelled as a series of vehicle routing problems that optimise the total distance travelled, with the current solution used as the starting point for the next iteration. A solution method using adaptive large neighbourhood search attempts to fit the new customer request into the existing routes while still allowing the feeder service to meet the scheduled train. Three scenarios are tested using multiple vehicles. Instances with realistic demand and 250 customers are generated using realistic distributions derived from the Victorian Integrated Survey of Activities and Travel. The first scenario varies the degree of dynamism across different instances. A second scenario varies the degree of dynamism with a spatial distribution of immediate requests reflecting distance to the train station. We also investigate the impact of train frequencies on the feeder service by altering the headway between trains. More frequent trains leaves less scope for optimisation and the longer headway has a higher success rate as the vehicle has time to pick everyone up. Overall we find that the most inefficient service occurs around 80% degree of dynamism.