Direct Current Load Prediction Using Machine Learning Model for Enhanced Telecommunications Reliability
Year: 2026
Type: article
Abstract: The reliable operation of Base Transceiver Stations (BTS) is essential for ensuring consistent mobile network performance, particularly in regions where internet usage continues to rise. One of the critical parameters in BTS power management is the DC Load, which reflects the current consumption of BTS equipment and is crucial for determining battery requirements during power outages. However, most BTS sites in Indonesia lack Power Monitoring Systems (PMS), resulting in the absence of real-time DC Load data needed for effective energy planning. This study proposes a regression-based machine learning approach to estimate DC Load by utilizing operational parameters such as BTS power consumption, traffic, connected users, and PRB utilization. Both single-feature and multi-feature regression models were developed using Linear Regression, Decision Tree Regressor, Random Forest Regressor, and XGBoost. The results show that BTS Power (kWh) has the highest correlation with DC Load, making it an effective single predictor for baseline modeling. The single-feature Linear Regression model performs best under high-load conditions, whereas the Random Forest Regressor provides superior accuracy under low-load scenarios, achieving an R² of up to 0.995. An adaptive hybrid model combining both algorithms was implemented to optimize prediction performance across varying load ranges. The proposed method was applied to approximately 3,500 BTS sites without PMS in Central Java, enabling more accurate battery sizing and improving operational reliability. Overall, this research demonstrates that machine learning can serve as a practical and efficient tool for DC Load prediction, supporting energy optimization and infrastructure planning in telecommunication networks.
Source: JURNAL INFOTEL
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $65
OpenAlex ID: W7236400908
