A decentralised genetic algorithm with sub-objectives for the transit network design problem

This paper introduces a novel approach to the transit network design problem. This technique, based on a genetic algorithm, uses multiple sub-populations in parallel, each prioritising different aspects of the optimal solution. These aspects include minimising transfers, reducing detours, and addressing specific parts of the demand. All of these aspects are important in minimising average passenger travel time, and artificially changing each weight factor in the objective function allows us to escape local optima efficiently. Our algorithm demonstrates improved performance when tested on well-known benchmark instances by diversifying the population and letting each sub-population focus on different parts of a qualitative solution. This algorithm reduces average travel times by more than 2% compared to the current state-of-the-art. Additionally, our approach generates a range of high-quality and distinct solutions, offering valuable insights for decision-makers. The algorithm is also tested on a large, complex instance based on a rural area in Belgium and shows the capability to find improvements over the current transit network.