Asynchronous Adaptive Large Neighborhood Search Algorithm for Dynamic Matching Problem in Ride Hailing Services

In the recent years there has been an increasing interest in optimizing the vehicle matching problem in Ride Hailing (RH) services. The problem is closely related to the classical Dial a Ride Problem (DARP), for which various efficient metaheuristics exist in literature. Among these metaheuristics, the Adaptive Large Neighborhood Search (ALNS) algorithm has shown great results [1]. The vehicle matching problem is similar to the dynamic DARP (new requests arrive dynamically). However, only few works focused on the dynamic aspect of the problem so far.Furthermore, the majority of transportation studies that simulated RH services neither benefited from DARP procedures nor considered the asynchronous nature of real scenarios, i.e. the customers need quick responses and the vehicles keep moving while computing assignments.Therefore, in the current work we evaluate the performance of rolling horizon ALNS in an asynchronous real-time framework, where vehicle movements are kept in a separate CPU process. We simulate various percentages of trips from New York Taxi data for the study. Using the presented batching strategy, we show that the ALNS can not only significantly reduce the batching period without compromising the solution quality but also can be used in real-time with good solutions.