A Multi-Task Scheduling Decision-Making Approach for SAEV Fleets: Integrating Stochastic Demand and Long-Term Benefit

Shared autonomous electric vehicles (SAEVs) are becoming increasingly popular in major metropolitan transportation systems. However, efficiently managing such fleets under uncertain travel demand remains a major challenge. To address this, a multi-stage decision-making framework is proposed for integrated scheduling optimization across multiple SAEV tasks. Firstly, vehicle charging and order-taking are formulated as bipartite graph matching problems. The long-term benefits, defined as the expected cumulative reward from future order assignments, are quantified using a temporal-difference (TD) algorithm with a long short-term memory (LSTM) network-based approximation. Secondly, vehicle repositioning is modeled as a stochastic program, and a chance constraint is employed to manage demand uncertainty by ensuring a predefined probability of order fulfillment. Thirdly, vehicle parking is modeled as an integer program to minimize vehicle idle mileage. A customized hybrid solution strategy combining the Jonker-Volgenant (JV) algorithm and the Gurobi solver is developed to solve the overall model. Finally, case studies based on real-world taxi trajectory data from Shanghai demonstrate that the proposed scheduling strategy outperforms alternative approaches in economic efficiency and control flexibility. Sensitivity analyses further reveal how key parameters affect system performance and identify their sensitive value ranges.