DEVELOPMENT OF THE METHOD OF CLUSTERIZATION OF TRAIN SITUATIONS

Abstract

The introduction of intelligent locomotive control systems requires better approaches to assessing and monitoring the current train situation than those used in modern traction rolling stock. Automatic detection of complex abnormal situations is currently not provided. For example, determining the inefficiency of the brakes, speeding, the presence of obstacles or people on the track, the deterioration of the traction properties of rolling stock, etc. relies solely on the driver of the locomotive. Given the important impact of these factors on traffic safety, it is proposed to include in the functions of automated and intelligent traffic control systems recognition of abnormal situations and notification of its occurrence. When driving a train, all objects of classification (train situations) are divided into a finite number of classes. A finite number of precedent objects are known and studied for each class. The task of pattern recognition is to assign a new recognizable situation to a class. The classifier or decisive rule is the rule of assigning the image of a train situation to one of the classes on the basis of its vector of features. An order of classification of train situations has been developed, which allows to allocate clusters of any complex shape, provided that different parts of such clusters are connected by chains of close to each other elements. The measure of difference is the square of the Euclidean distance.