METHODS AND MODELS OF SELF-LEARNING AUTOMATED DETECTION SYSTEMS STATE OF HIGH-SPEED RAILWAY TRANSPORT HUBS

Abstract

The article contains the results of researches aimed at the further development of methods and models for self-trained automated detection systems (SADS) of nodes and aggregates of high-speed railway transport (HSRT) based on the clustering of failure signs. There has been developed SADS model of nodes and aggregates of HSRT and a method for its training, in which the procedure of fuzzy clustering of failure signs realization is applied. The procedure for decision rules correction is also considered, that will allow the creation of adaptive self-trained mechanisms for automated systems for detecting HSRT nodes and aggregates. It is proposed to use the modified information condition of functional effectiveness (ICFE) as an evaluation indicator of the training effectiveness of SADS. This condition is based on Kullback-Leibler information-distance criteria. There is considered the method of space fragmentation of failure signs realization of the HSRT nodes and aggregates into clusters during the implementation of the failure recognition procedure. Also there is considered the method of initial training of SADS. The method is an iterative procedure for finding the global maximum of ICFE. There were substantiated perspectives of decisions on the integrated evaluation of the detection results of the nodes and aggregates of the HSRT rolling stock based on the use in similar automated complexes for detecting models with fuzzy clustering algorithms of hundreds of the HSRT failure signs systems.