Modeling and application of a customized knowledge graph for railway alignment optimization

Abstract

Railway alignment optimization is a complex process in which human knowledge and experience are extensively used. However, the related knowledge is usually unstructured, which is difficult for computers to recognize. Moreover, related knowledge is applied in fragmented ways in existing alignment optimization methods, which are thus difficult to update with actual advancements of human experience and knowledge. To solve the above problems, the first-known knowledge graph modeling method for railway alignment optimization is proposed in this paper. First, a hierarchical and categorized semantic network modeling approach for railway alignment design knowledge is devised. Based on this, a railway alignment design knowledge graph (RAD-KG) is constructed. Then, a rapid knowledge retrieval method is proposed for improving the querying efficiency from the RAD-KG during alignment optimization. Finally, the RAD-KG integrating multiple alignment design principles is successfully applied to a real-world case. It is verified that the alignment generated by the proposed method reduces costs by 9.2% compared with conventional manual work by experienced engineers. Moreover, the RAD-KG-assisted method can rapidly update alignment design guidelines during optimization and, hence, produce several alignment alternatives satisfying various complicated requirements, which confirms the flexibility of the proposed method.