For efficient management of the railway direction, it was proposed to create a predictive model of the train operation. One of the components of this model is the train arrival module, designed to determine the arrival time of different trains to technical stations of the railway direction. The train arrival module is proposed to build based on a neural network, which using statistical information for prior periods and the train data obtained in real time, determines the train arrival time at the technical station. Since the train departure parameters (time and date of departure from the next technical station, train weight and engine type) have different measurement units and there are significant differences between the minimum and the maximum value of the same parameter, it was decided to encode data about train in binary form. The values of each factor were grouped by intervals of a certain value. As a result of experiments with different types of neural networks, it was found that using the perceptron, the structure and construction method of which is given in the paper provides the smallest error of the results obtained. The operation principle of such neural network is as follows. Train information is encoded and fed to the neural network input in binary form; the result of the neural network operation is also a binary output vector, the value of which is interpreted in a certain value of the train movement duration. Based on the movement duration values, the predicted arrival time of freight trains at the technical station is calculated. Experiments with the interval value at binary coding of individual factors have shown a significant effect of this parameter on the neural network operation quality and train arrival forecasting accuracy.