Energy efficiency in data center operations has become a critical issue in the digital era, including for public utilities such as Regional Drinking Water Companies (PDAM). One of the key indicators of data center efficiency is Power Usage Effectiveness (PUE), with PDAM recording a daily average of 2.11. Significant spikes in PUE values during certain months suggest operational dynamics that are not yet fully understood, highlighting the need for a comprehensive analysis of the contributing factors. This study employs Artificial Neural Network (ANN) and Long Short-Term Memory (LSTM) algorithms to forecast PUE values. The research process includes data understanding, preprocessing, modeling, evaluation, and forecasting, conducted under three time-based split scenarios: 60%:20%:20%, 40%:20%:40%, and 20%:20%:60%. Among these, the third scenario demonstrated the best performance. The ANN model achieved a Mean Absolute Percentage Error (MAPE) of 10.10%, a Root Mean Square Error (RMSE) of 0.2565, and an accuracy of 85.34%. Meanwhile, the LSTM model recorded a MAPE of 10.62%, RMSE of 0.2600, and accuracy of 85.14%. These results fall within the 10%–20% error range, indicating reliable forecasting performance. The predicted PUE values were 2.38 for ANN and 1.85 for LSTM, both categorized as “Average” in terms of energy efficiency, suggesting room for improvement. The highest energy loads were attributed to cooling systems, specifically air conditioning (AC), air conditioning and mechanical ventilation (ACMV), and uninterruptible power supply (UPS) systems. Therefore, future infrastructure optimization efforts should prioritize improvements in these areas
Copyrights © 2026