Enhancing Energy Efficiency in BBUs: Predictive Analysis with Long-Term and Granular Data

presentation, poster

English

This paper investigates optimizing Baseband Unit (BBU) energy consumption through predictive modeling using both long-term and granular datasets. We evaluate various machine learning models, including a hyperparameter-optimized Convolutional Neural Networks- Long Short-Term Memory (CNN-LSTM) architecture, to predict traffic volume, user equipment (UE) downlink latency, and BBU power consumption. Our results show that while granular data offered detailed insights, the long-term dataset consistently provided better results for traffic volume prediction. Despite this, we focused on the granular dataset for its ability to capture short-term fluctuations, which is crucial for real-time network management. The CNN-LSTM with hyperparameter optimization demonstrated superior accuracy compared to baseline models such as LSTM and GRU, especially for predicting downlink latency and power consumption. The combined use of long-term and granular datasets added complexity without significantly improving predictive accuracy compared to individual datasets. These findings are instrumental in advancing the development of Predictive Energy Saver for Baseband Units (PESBiU) 2.0, focused on further optimizing energy efficiency in 5G networks. Keywords: 5G

5G networks, energy efficiency, Baseband Unit (BBU), machine learning, hyperparameter optimization, power consumption, traffic volume

KOLÁČKOVÁ, A.; PHAN, V.; YOUNESIAN, E.; HOŠEK, J.; JEŘÁBEK, J.; KHATIB, N.

7. 11. 2024

The 9th International Conference on Advanced Engineering - Theory and Applications

1