Gradient Boosting Algorithms Application for Beamsteering in V2X Systems

This paper presents the simulation results of a machine learning application in the vehicle-to-infrastructure system aimed at controlling the radiation pattern of the phased antenna array based on a priory data about the path geometry and traffic model in the service area of the antenna system, as well as the received information from the message-request from vehicles. An application was created to generate a dataset that considers both the a priori data about the service area of the vehicle-to-infrastructure network object and the momentary data received from the vehicles. The operation of gradient boosting XGBoost, LightGBM, and CatBoost algorithms for positioning vehicles without reliance on global navigation satellite system coordinates is considered. The CatBoost algorithm had the highest accuracy of 96.8%, and the lowest delay of 0.056 ms in processing the input data obtained with the XGBoost algorithm. The algorithms were also tested for robustness to multi-beam signal propagation and possible solutions to increase it where indicated.