Robust Optimization of Shared Bus Routing Based on Dynamic Open-Chain

The Shared Bus, also known as the Customer Bus (CB), is a network demand-responsive urban public transport travel mode. The efficiency and rationality of its scheduling strategy are crucial for public transport operators to enhance their operating profit and competitiveness. It provides environmentally-friendly, secure, and efficient on-demand travel services for urban residents, and has been implemented in several cities in China. Although the path optimisation problem regarding shared buses has become a research focus, the current research is predominantly based on static demand and deterministic scenarios, which is insufficient for the optimisation of real-time order-responsive service scheduling systems of networks with a high degree of freedom. It is evident that buses are susceptible to uncertainties during their journey, which has a direct impact on the output of the optimisation system. This, in turn, has an adverse effect on passenger satisfaction and increases vehicle operating costs. The VRP-like model is not suitable for shared buses with a high degree of freedom, which have no fixed routes and distribution centres. Consequently, this study proposes to transform the dynamic transport path planning problem into a variable-length open chain rolling optimisation problem. In order to enhance the resilience of the system in the face of extreme environmental conditions, a parallel robust scheduling optimisation strategy is employed to schedule all buses. Given the complexity of the multi-bus parallel robust optimisation structure in the dynamic open chain rolling framework, this study develops a meta-heuristic algorithm with high-dimensional objectives based on the NSGA-III framework. This algorithm employs solution coding methods of ‘combined load and delivery, (DCLD) and path node ordering. The experiment comprises five principal sections, each of which is evaluated in terms of the comparison results of the inverse generation distance evaluation metrics, IGD and Pareto frontiers, following repeated runs with different numbers of rolls. The results demonstrate that the DCLD-NSGA-III algorithm outperforms the other algorithms.