A knowledge graph-based hazard prediction approach for preventing railway operational accidents.

• A railway operational hazard knowledge graph modelling method is developed. • A knowledge graph embedding-based hazard prediction method is proposed. • A procedure of formulating targeted hazard control measures is shown. • An application to real accident data demonstrates the effectiveness of the proposed approach in predicting hazards and assisting in making accident prevention decisions. Railway operational accidents are usually caused by the domino effects of hazards. Predicting such hazards before accidents is an essential way to prevent operational accidents of railways. Various railway operational hazards constitute a heterogeneous relationship network due to their interactions. It is useful to predict hazards from such a network. In this paper, a novel knowledge graph-based hazard prediction approach is proposed, aiming to capture hazards in advance for blocking potential accident causation paths. This approach serves as a powerful supplement to classical ways of predicting railway accident information. Its originality lies in the application of knowledge graph embedding-based link prediction theory to railway operational hazard prediction, by means of a translation-based embedding method adapting to the relational features of hazards. It also provides a feasible way to construct the railway operational hazard knowledge graph. The outcomes of this approach could offer railway operators the basis of decision regarding accident prevention. An example of application to a set of real railway operational accident data in the UK is presented. The results show that this approach is effective in terms of predicting hazards and assisting in developing targeted hazard control measures. [ABSTRACT FROM AUTHOR]