With the rapid and advanced development of high-speed railway (HSR) in China, enhancing the coordinated adaptability of HSR train stop planning and timetabling while improving operational precision has become increasingly critical. To address this, this paper tackles the limitations of existing collaborative optimization approaches. Most conventional methods rely on two-stage models-an approach prone to goal misalignment and local optima. Therefore, this paper resolves the aforementioned issues by integrating the two stages into a single optimization phase. Specifically, it takes into account the impacts of train types and stop-start additional time on the optimization scheme, adopts inter-station origin-destination (OD) accessibility to indirectly characterize the demand for passenger flow exchange, and formulates departure time rules to maintain the stability of the train timetable structure. A multi-objective collaborative optimization model is developed, considering constraints such as station service frequency and safety intervals, with objectives including minimizing total train travel time, total number of stops, and the deviation between actual and expected departure times at the origin station. An empirical study was conducted on the 58 downbound trains of the Wuhan-Guangzhou HSR in June 2025 ,using solver Gurobi Results indicated reductions in total train travel time and total stops by 2.6% and 3.5%, respectively, and a reduction of 13 overtaking events. This optimization improves accessibility between lower-tier nodes while maintaining service stability at higher-tier nodes, leading to enhanced operational efficiency, service quality, and safety. It provides valuable guidance for high-speed railway operators in developing integrated stop plans and timetables.

Federated learning can break the data silos while ensuring the participants control of the data, thus effectively protecting the data privacy of intelligent edge nodes. Existing federated learning techniques usually upload the intermediate parameters of model training to the parameter server to realize model aggregation, but there are two problems in this process: on the one hand, the federated learning process generates a large amount of communication overhead due to multiple learning. On the other hand, malicious nodes may upload false parameters or low-quality models, thus affecting the aggregation process and model quality. To address the above two problems, an enhanced data sharing method based on blockchain and random forest is proposed. The method consists of three parts, which are the tree algorithm based on CART tree, the federated forest model based on CART tree, and the learning algorithm based on CART tree and Bagging. Firstly, a blockchain is constructed and the data is stored on the blockchain. Secondly, the data is distributed to the nodes involved in learning, each node stores the corresponding data, and multiple nodes are constructed into a tree. Then the central server is responsible for collecting all the trees constructed by the nodes to generate a complete tree. Finally, the model is used to make predictions. Experiments using real datasets show that this method can not only enhance the mutual trust between nodes participating in federated learning, but also reduce communication overhead, and finally obtain a trusted federated learning model with enhanced privacy protection.