Bayesian Network for Base Station Energy Consumption Data Completion

Abstract

Base station energy consumption data loss is randomly caused when collecting and saving various data from massive base stations. At the same time, after the AI energy-saving system issued the 5G deep sleep strategy, the energy consumption of the base station is greatly reduced, but it is necessary to restore the real energy consumption during the energy-saving period. The integrity of base station energy consumption data plays a key role in the assessment of AI base station energy-saving system. The assessment of base station energy consumption can be used as an important feedback content for the AI training model (for prediction, decision-making, etc.) that realizes automatic energy saving for base station. Existing deep learning solutions or GAN (Generative adversarial network) are difficult to meet the requirements of a large amount of effective training data. In this paper, a method based on Gaussian Bayesian network high-dimensional tensor decomposition (GBHTD) is proposed. Extensive empirical studies on multiple areas base station energy consumption datasets demonstrate that GBHTD can outperform state-of-the-art baseline methods in base station energy consumption data completion performance. Specifically, GBHTD decreases MAPE (Mean absolute percentage error) by up to 11%, reducing training time by up to 99% compared to the baselines.