Extension of Biased Random-Key Genetic Algorithm With Q-Learning for Dial-a-Ride Problem

With the widespread adoption of delivery services and advancements in autonomous driving technology, route optimization problems are being intensively studied. In particular, the expansion of mobility services, such as car-sharing and ride-sharing, is driving the demand for efficient urban transportation. Ride-sharing services are essential for optimizing urban transportation; however, their implementation is limited by multiple factors. In this study, we propose a hybrid algorithm that combines the biased random-key genetic algorithm (BRKGA) and Q-learning (QL), referred to as BRKGA-QL, for the dial-a-ride problem (DARP). The proposed algorithm utilizes a time-expanded network to address time window constraints during search processes. Furthermore, the solution quality is enhanced by controlling multiple neighborhood searches through chromosomal representation and incorporating search results into the genetic information. Performance evaluation of the proposed algorithm using benchmark tests demonstrates that it achieves shorter total travel distances than conventional algorithms for many test cases. These results confirm the effectiveness of the proposed approach for solving the DARP.