Intelligent Approach to Solving the Problem Control over Railway Cars in the Marshalling Yard

This paper presents a new hybrid approach to solving the problem of wagon cuts monitoring in the classification bowl of a freight station. The approach combines vision technologies for identifying key objects in video camera images. It also involves modeling the state of the fleet based on data obtained from various sources of information about the state of the marshalling yard and classification bowl devices. The study of the problem under consideration is relevant as the solutions existing on the market do not provide accurate and comprehensive information about the movement of wagon cuts and locomotives in the classification bowl. The authors suggest combining an artificial deep learning neural network for recognizing wagon cuts to create a data-driven algorithm and a virtual model of the classification bowl based on multisensory data to build a knowledge-driven algorithm. This combination allows both to smooth out the errors of the first and second kind caused by the impossibility of obtaining all kinds of variations of wagon cuts in the frames of video cameras and to have the most complete static and dynamic picture of the marshalling yard. The paper presents the results of the algorithm and its advantages over existing solutions for the control of wagon cuts in the classification bowl.