Improved machine classification algorithm for electric rail circuits in train warning systems

Abstract

There are known algorithms that implement the classification of code signals in an electric rail circuit. These algorithms, however, have some disadvantages in the form of either relatively complex implementation or reduced accuracy in the presence of noise in a code signal. In this article, we present an improved classification algorithm, which combines the simplicity of implementation and accuracy. The algorithm is based on a neural network trained with cyclically shifted learning examples. We explore the optimal size of the neural network for this type of training set. At the cost of the increased size of the neural network we streamline the classification process and preserve its accuracy.