<![CDATA[The decreasing availability of clean water has motivated the use of Smart Atmospheric Water Generator (SAWG) systems as an alternative water source, but their electrical energy consumption fluctuates with ambient conditions and operating patterns. This study develops a predictive model of SAWG energy consumption (kWh) using Extreme Gradient Boosting (XGBoost) and demonstrates a prediction-based operational optimization scheme for energy-efficient scheduling. The SAWG logging dataset (1,601 rows, 9 variables) is preprocessed through missing-value handling, numeric conversion, and noise/outlier detection, resulting in 1,313 usable records. The feature set includes environmental parameters, electrical signals, and time features: hour of day, day of week, and month. Modeling employs chronological time-based splits (80:20 as the main configuration and 60:40 as a robustness check), Time Series Cross-Validation on the training block, and hyperparameter tuning via GridSearchCV. Evaluation on the hold-out test sets shows that the model’s performance in a strict time-series setting remains limited: for the 80:20 split, the test results are approximately MAE = 23.16 kWh, MSE = 648.93 kWh², and R² = −0.22, while for the 60:40 split they are MAE = 27.21 kWh, MSE = 932.17 kWh², and R² = −1.75. Although the model cannot yet explain the overall variance of energy consumption satisfactorily, it can still be used to rank hours by predicted energy. In the prediction-based operational optimization stage, hourly model outputs are fed into a Greedy Scheduler that selects H = 8 operating hours with the lowest predicted energy. Compared with a naive schedule, which yields a total predicted energy of 47.493 kWh over the simulation horizon, the greedy schedule achieves 43.134 kWh, corresponding to an estimated saving of about 9.18%. These results indicate that prediction-based scheduling can reduce SAWG energy consumption without modifying the device hardware and can be further developed as a decision-support component for SAWG operation.]]>
