Development of an automated software tool based on machine learning methods for solving problems of radio planning in subway sections

Abstract

Introduction: The modern approach to radio planning provides subway passengers with uninterrupted access to the Internet. This is achieved through the use of a special signal propagation model which calculates signal power loss during its propagation between a transmitter and a receiver on subway lines. The disadvantage of the model is the high computational complexity. Purpose: Using machine learning methods to develop an algorithm for predicting the signal power loss, the algorithm being characterized by high accuracy and low computational complexity. Results: The analysis of machine learning methods revealed that the maximum possible accuracy in solving the problem is provided by the random forest method. A data structure containing the parameters of a digital map of subway lines was developed to train the selected method and predict a signal power loss. While developing the final algorithm a number of assumptions were made, such as: the problem is solved as a classification problem, the predicted values are integers. A signal power loss prediction algorithm that does not directly use the propagation model was developed, which reduced the computational complexity and the execution time for solving radio planning problems, with high prediction accuracy maintained. Practical relevance: Due to the use of machine learning methods in developed algorithms the time for performing radio planning was reduced from several days to several hours, with accuracy preserved. This allows to process more radio planning orders or to reduce the working time for engineers to complete the same number of orders, which is a financial benefit.