THE POSSIBILITY OF USING ARTIFICIAL NEURAL NETWORKS WHEN PREDICTING TRAIN WORK TRAIN DIRECTIONS

Abstract

Effective operational planning of the transportation process should be based on accurate forecasts of revenue trains, wagons, locomotives, freight, etc. This prediction can be realized with the help of modern mathematical method – apparatus of artificial neural networks. Neural networks have such properties as adaptive learning, self-organizing, summarizing, calculating in real time and resistance to disruption. The main applications of neural networks are functions approximation, associative memory, data compression, pattern recognition and classification, optimization problems, the management of complex processes and forecasting. The constituent elements of a formal neuron are х1, х2, …, хn – the network inputs, each of which is characterized by its weight w1, w2, …, wn, respectively. The adder Σ summarizes the input signals, the activation function f describes the neuron transition rule when new signals are received. According to connections architecture the neural networks can be grouped into two classes: feedforward networks, which graphs have no loops, and recurrent networks or networks with feedback. By the neural network training is understood a setting of the network architecture and weights for the effective implementation of the task. The learning methods can be divided into two groups: with the teacher and without a teacher. Neural networks can be used for real-time forecasting of the train arrival at the technical stations on the basis of known departure parameters of the trains: information about the time of departure from the neighboring technical station, the train weight, the data on seasonal and weekly uneven. While designing the neural network to solve a particular problem should be considered several variants differing in, for example, the number of neurons in the layers, the number of hidden layers, etc. The main indicators, by which can be made a comparison of the selected architectures of neural networks can be such parameters as the minimum error, as well as a high correlation coefficient between actual and calculated data. Thus, the artificial neural networks can be widely used in the design of operational traffic control systems since allows to receive an accurate prediction based on many factors and is relatively simple to design