Automatic search model of railway shunting route based on improved artificial neural network algorithm

To enhance railway shunting route search accuracy while reducing memory consumption, this study proposes an automatic railway shunting route search model based on an improved artificial neural network algorithm. The method constructs a time-varying railway network topology through dynamic topology modeling and conflict avoidance mechanisms. Signals and switches are abstracted as graph vertices, while track segments are transformed into weighted edges with occupancy markers. A 0–1 matrix determines track connectivity and functional matching, eliminating route conflict schemes from the feasible solution set. The search process is optimized using a binary tree structure that capitalizes on the structural similarity between station areas and binary trees, transforming route searches into binary tree traversal operations. A non-recursive first-order traversal algorithm minimizes redundant searches and enhances efficiency. The improved artificial neural network standardizes data and initializes weights using binary tree prior knowledge to optimize global train route decision-making. An ant colony algorithm dynamically updates path pheromones and generates real-time modifiable dynamic route tables, addressing the high memory consumption of traditional static tables. The experimental results demonstrate that compared to conventional route table searches and station structure searches, the proposed method improves route search accuracy by 9–6.8%, reduces memory usage by 40–60%, and increases throat switch utilization efficiency by 15–20%. These findings validate the method's efficiency and robustness in complex railway shunting scenarios, offering a novel approach to railway shunting automation that combines precision with real-time performance.