AbstractIn the field of modern urban logistics, the development of door-to-door freight transport through rail–road combined transportation is a necessary approach to achieve modernization and smartness of railroad freight transportation. As the last step of door-to-door rail–road joint transportation, station-to-door transportation determines the quality and efficiency of services. Meanwhile, the operation time of goods assembly is uncertain in the freight center station, freight handling station, and in transit, which largely limits the efficiency of rail–road combined transportation delivery at the stage of station-to-door. To address the aforementioned problems, we proposed a forward-looking matching strategy (FL) that jointly considers the set of goods orders that can be fulfilled in the current decision stage and the set of goods orders that can only be fulfilled in the future stage to improve the matching effect. Then, we built a two-stage stochastic dynamic programming model that jointly considers matching between goods orders and distribution path optimization. At the same time, we simplified the complex model by using a Bayesian approach to update the goods’ operation time in real time. Finally, we designed an improved differential evolution algorithm based on order similarity and distribution for solving the optimization. The algorithm we designed reduces 34.69% in transportation cost and 31.37% in waiting time cost compared with the actual delivery plan implemented.