Reinforcement learning-enabled genetic algorithm for school bus scheduling

In this paper, we focus on a bi-objective school bus scheduling optimization problem, which is a subset of vehicle fleet scheduling problems to transport students distributed across a designated area to the relevant schools. The problem being proven as NP-hard in the literature, we propose an algorithm that seamlessly integrates a reinforcement learning approach with a genetic algorithm. Our proposed algorithm utilizes the processed data supplied by our intelligent transportation system framework to decide the genetic algorithm parameters on-the-fly with the aid of reinforcement learning. With the active guidance of reinforcement learning, the efficiency of the genetic algorithm is improved, and the near-optimal schedule can be achieved in a shorter duration. To evaluate the model, we conducted experiments on a geospatial dataset comprising road networks, trip trajectories of buses, and the address of students. Results indicate that the genetic algorithm improves the travel distance and time compared to the existing schedule. Reinforcement learning-enabled genetic algorithm improves the performance and the objective function significantly, furthermore with a fewer number of generations compared to various state-of-the-art evolutionary algorithms. The saving by reinforcement learning-enabled genetic algorithm compared to the schedule by initial state generation process is 8.63% and 16.92% for the travel distance for buses and students, respectively, and 14.95% and 26.58% for the travel time for buses and students, respectively.

Automated summary

In this paper, a bi-objective school bus scheduling optimization problem is addressed by proposing a hybrid algorithm integrating reinforcement learning and genetic algorithm. Experimental results demonstrate that the proposed algorithm improves the travel distance and time for buses and students compared to existing algorithms, while also achieving better performance with fewer generations.