Research On The Application Of Knowledge Graph Representation Learning In Vehicle Safety Risk Warning

With the continuous development of the Chinese economy and technology,China is developing from a large vehicle country to a strong vehicle country,vehicle safety risks are receiving more and more attention.The causes of vehicle safety risks are complex and time-sensitive,it is difficult to determine the causes of vehicle safety risk from individual aspects,such as whether a vehicle is at risk of running a red light based on driver behavior,vehicle operational status,and road status.To this end,the reasoning and early warning of vehicle safety risk trends have become an important research topic.Traditional data structure modeling can hardly take into account the timeliness and complexity of the causes of safety risks.As a large-scale graph database of recorded facts,the knowledge graph can model attributes in historical safety risks,reason about the development of risk patterns and provide early warning of safety risks.Therefore,we conducted an in-depth profiling study of the vehicle safety risk domain,designed a knowledge graph schema based on domain-related properties,we obtained structured and unstructured data from vehicle accident safety reports,used semi-automated methods to obtain the knowledge graph data,and finally constructed a vehicle safety risk knowledge graph based on the schema map.In the design of risk prediction models,we considered the timeliness and complexity of vehicle safety risk and proposed two knowledge graph representation learning models.For the timeliness of safety risks,considering that a large number of risks evolve in the real world and there are temporal relationships between risks,we extend the tensor decomposition to the fourth dimension to introduce a temporal dimension to the static knowledge graph,which largely improves the semantic information of the knowledge map,and propose a tensor decomposition-based linked prediction model for the temporal knowledge graph.For the complexity of safety risks,considering that risk may involve multiple factors in the real world,traditional knowledge graphs using triples to model such risks need to split a fact involving multiple groups into multiple triples,which simplifies the data representation but will inevitably lose a large amount of semantic information.While directly extending the traditional triple link prediction model to multiple group link prediction will result in a proliferation of the number of parameters.In this thesis,we propose a tensor decomposition-based multivariate group knowledge graph representation learning method,which allows the model to retain multivariate group semantic information at the same order of magnitude of the number of parameters,and improves the correct rate accuracy of vehicle safety risk knowledge map link prediction.Finally,based on the constructed vehicle safety risk knowledge map,we apply the algorithm model and implements an in-vehicle safety risk early warning system,which displays early warning of possible vehicle safety risks through the analysis of safety risk factors.In summary,based on the knowledge graph,we conduct an in-depth study on the vehicle safety risk warning method,constructs a knowledge graph in the field of vehicle safety risk,proposes relevant algorithms for the safety risk characteristics,and also has the ability to be applied in real scenarios.