A Machine Learning Approach to Temporal Traffic-Aware Energy-Efficient Cellular Networks

With the rapidly increasing demand for cellular network services, operators have responded by deploying more base stations (BSs) which have greatly increased the total energy consumption of cellular network infrastructure. In this paper, machine learning is used to exploit temporal variations in cellular network traffic. Four machine learning algorithms and one conventional time series forecasting method are used for traffic prediction and then compared. Results show that random forest regression performs best with a coefficient of determination of 0.82 on the whole dataset and 0.84 on data of isolated days of the week. Based on the predicted traffic, three sleep mode schemes are applied to a homogeneous network. Simulation results show that the strategic sleep mode scheme performs best with an 87.4% energy saving gain over the conventional scheme and a 32% percent energy saving gain over the random scheme for a given day. In addition, the strategic scheme achieves an hourly average power saving of 3,836 W per kilometer squared, which proves that machine learning traffic prediction-based sleep modes are instrumental in achieving energy-efficient cellular networks.