In multi-site tourist areas, tourist movements show trip-chain characteristics and dynamic travel demands, which can be satisfied by customized buses and ride-hailing. Existing studies mostly optimize these two travel modes separately. This paper establishes a bi-level model for the joint optimization of customized bus and ride-hailing services, aiming to maximize operators' total profit and minimize passengers' generalized travel cost. The upper level optimizes customized bus routes, frequencies, pricing together with ride-hailing fleet allocation and pricing. The lower level adopts a multinomial logit model to characterize tourists' mode-choice behavior. A heuristic algorithm combining genetic algorithm and simulated annealing with upper-lower iterative coordination is proposed to solve the mixed-integer nonlinear programming problem. Case study of Chongming Island, Shanghai is conducted. Results show that the joint optimization increases operator profit by 12.55% and reduces total passenger cost by 4.00% compared with independent optimization. Sensitivity analyses offer operational insights for multi-site tourist transportation systems.
