Machine Learning based Energy Saving Scheme in Wireless Access Networks

With the deployment of the 5G, lower latency and higher frequency brings not only the users' satisfaction but also the dense deployment of base stations which causes large energy consumption. Thus, energy saving is a challenging issue. Machine learning techniques, which could extract features from raw data in the wireless access network and predict the future state by the previous features, are useful to energy saving. In this paper, we proposed the energy saving scheme based on machine learning techniques to achieve the balance between system performance and energy efficiency. Cell switching-off by load prediction using different algorithms, and dynamic threshold adjustment improve the energy efficiency and thereby reduces the operation cost. By demonstrating the results of experiments in wireless environment, the switch-off duration could reach up to 176% increase, and the energy scheme can save up to 2, 223 RMB electricity saving per week per cell.