Journal of Electrical Engineering and Automation

With the large-scale deployment of 5G base stations, energy consumption issues have become increasingly prominent. Traditional energy-saving solutions suffer from strong temporal fluctuations in service traffic, insufficient prediction accuracy, and poor generalization of energy-saving strategies in heterogeneous base station scenarios, making them difficult to adapt to dynamic service demands and prone to degrading network service quality. To address this, this paper proposes an adaptive base station energy-saving system based on Multi-Feature Long Short-Term Memory neural networks (MF-LSTM) and K-means clustering. The MF-LSTM integrates multi-dimensional features integrates temporal features (weekday/hour) and traffic data to enhance prediction accuracy, while K-means clustering categorizes base station energy consumption patterns to dynamically match optimal energy-saving strategies. Experimental results demonstrate that the MF-LSTM explains over 92% of traffic fluctuations, and the adaptive energy-saving strategy achieves an 11.71% reduction in base station energy consumption, providing a more effective approach for green communication.

Base station energy conservation; Long Short-Term Memory neural network; Adaptive algorithm