Lead-Acid Battery Lifetime Estimation using Limited Labeled Data for Cellular Base Stations

Determining battery lifetime used in cellular base stations is crucial for mobile operators to maintain availability and quality of service as well as to optimize operational expenses. Generally to achieve this, technicians must regularly visit cellular sites to manually measure the maximum battery capacity. However, this is a time consuming, costly, and challenging task. To tackle with these problems, it is vital to develop an automatic battery lifetime estimation framework. Generally, the existing battery lifetime estimation strategies depend mostly on time series and large labeled battery datasets which are costly and almost impossible to obtain in telecommunication scenarios. Therefore, in this paper we propose a data-driven battery lifetime estimation framework, based on a non-time series and limited labeled battery dataset. Apart from other studies, we mainly extract features from network alarm data and use a simple but effective automatic strategy to generate a small labeled battery set. To extend the number of labels, we propose an active semi-supervised SVM method with minimum human efforts. Finally, the obtained labeled dataset is fed into random forest algorithm to estimate battery lifetime in cellular base stations. The experimental results show the robustness and effectiveness of the proposed framework.