A Hotspot Prediction Scheme Based on Ensemble Learning With Adaptive Beamforming for Virtual Small Cell in 5G Networks

Effectively managing network congestion in 5G systems is crucial for ensuring seamless connectivity and optimal resource utilization. In this study, an operational scheme based on hotspot prediction for Virtual Small Cell (VSC) is presented, which aims to improve coverage, reduce delay and improve energy efficiency in 5G networks for congested locations. For this purpose, at first, using the location map of users, the location of hotspots is predicted by an Ensemble deep learning algorithm. Ensemble deep learning algorithm in this research combines the advantages of GRU and GRN networks and creates an efficient and accurate deep model that is more accurate than previous approaches that used single networks. Then, the information of hotspots is used to form VSCs and finally the adaptive beamforming operation is performed. To optimally allocate resources, genetic algorithm has been used so that the answer of the problem is not trapped in the local optimum. Simulation on the Telecom Italia dataset demonstrates a 10.13% RMSE, showing superior hotspot prediction accuracy compared to existing methods. Also, the numerical results of this study show that by using the hotspots prediction results, the energy efficiency of the system based on VSC operation and adaptive beamforming can be significantly improved.