Accident risk tensor-specific covariant model for railway accident risk assessment and prediction

The safety-cusp catastrophe model can describe both the continuous changing process of system safety and the emergent property of accidents. However, the model framework needs to be developed in data fusion to realize real-time accident risk prediction. In this paper, based on the tensor analysis, an accident risk tensor field is derived from the safety-cusp catastrophe model. To dynamically assess the railway accident risk, an accident risk tensor-specific covariant (ART-SC) model is constructed based on the accident risk tensor field, where the ac-cident risk of railway systems is synchronously measured by using the concept of specific covariant (SC) risk. By combining the ART-SC model with the mixture density networks (MDN), the analysis results of the actual monitoring data of a railway system show that the Gaussian disturbance is related to the high accident risk. Accordingly, a specific covariant risk-Gaussian disturbance identification (SCR-Gaussian DI) method is proposed to realize the real-time prediction of the high accident risk of railway systems. The analysis results based on the real-world monitoring data prove that the ART-SC model is reasonable, and the SCR-Gaussian DI method can accurately predict the high accident risk faced by the railway system in the train running process in real time.

Automated summary

In this paper, an accident risk tensor field is derived from the safety-cusp catastrophe model using tensor analysis, and an accident risk tensor-specific covariant model is constructed to dynamically assess railway accident risk. By combining this model with mixture density networks, the high accident risk faced by a railway system can be accurately predicted in real time by identifying the Gaussian disturbance.