Event risks and scenarios for the 1999 Paddington accident

Abstract

The accident near London in October 1999 didn't just shocks the UK. The collision of the two trains showed the obvious inadequacy of the existing models of train control on the Great Western main line of the UK railways. Specialists from a wide variety of fields of knowledge repeatedly return to this problem, trying to find key ways to minimize risks in this very difficult transportation industry. In this paper, an attempt is made to create a model NS neural network, which makes it possible to predict the possible consequences of a variety of technologically programmed actions, and safe dispatching. The five-sync model is based on all known risk-forming events that contributed to the occurrence of an accident in one way or another. As a function of synapse neuron activation, the bifurcation dependence of the doubled accounting period was chosen due to the number and uncertainty of the content of the accounting indicators that affect the final result. This made it possible to take into account the uncertainties that accompany the occurrence of any intersynapse channels from each input event to the resulting protocol, which assumes only a binary mapping of the expected risk. Thanks to this model, potential and hidden sources of potential risks for any system of events equivalent to the Paddington system are identified, taking into account some of its drawbacks. It is shown that the prevention of such accidents using neural network models should is reduced to automatically tracking images of potential risk-forming events embedded in the trained NS and alerting at least one or two of the monitored ones. It should be noted that such an NS model can be useful not only for the conditions of the Great Western Main Line of the UK railways, but for other systems that are subject to the same operating rules as in our example