Improvement of probabilistic models of failure prediction of elements of railroad power supply infrastructure

The article examines the technique of designing diagnostic system of infrastructure of electrical railways based on use of bayesian networks for prediction of probabilities of failures. To achieve maximum effectiveness of diagnosis we should minimize the number of input parameters, while maintaining the required accuracy. It is proposed to create a mathematical model of the diagnostic system, that will allow to evaluate the influence of each parameter on the accuracy of prediction of failures. To compensate the lack of source data we can use the advantage of bayesian networks - the opportunity to generate network structure by the method of expert evaluations. Generated bayesian network will perform the failure probability calculation with limited information.