Predicting wind-caused floater intrusion risk for overhead contact lines based on Bayesian neural network with spatiotemporal correlation analysis

Abstract

Wind-caused floater intrusion has posed enormous threats to the safety and resilience of overhead contact lines (OCLs) of electrified railway. In this paper, a Bayesian neural network (BNN) based prediction model is proposed, incorporating spatiotemporal correlations, uncertainty of extreme wind speed and direction, characteristic parameters of OCLs, environmental information and human factors into floater intrusion risk prediction. To select optimal candidates, the spatial-temporal correlation among wind data with respect to different OCL corridors are examined. Then, the probabilistic wind model is developed to capture the stochastic nature of wind events and account for the uncertainty in wind speed and direction. The spatiotemporal correlation-constrained environment sensitive parameter is formulated to investigate the impacts of wind, characteristic parameters of OCLs, environmental information and human factors on floater intrusion of OCLs. A BNN model is implemented into predicting wind-caused floater intrusion risk. Finally, the remarkable effectiveness and robustness of the proposed model are compared with some other advanced prediction methods. The experimental results demonstrate that the proposed model not only has the capability of uncertainty estimation, but also provides the confidence interval of floater intrusion risk prediction, which can play a significant role in preventive operational flexibility and resilience against weather-related risks.