From Classical Models to AI: Forecasting Humidity for Energy and Water Efficiency in Data Centers

Soaring energy and water consumption in AI data centers necessitates better operational efficiency to prevent resource strain and hardware damage. Accurately forecasting humidity is crucial, as it allows operators to optimize cooling systems, reduce costs, and avoid downtime from condensation or static discharge. A comparative study evaluates various models, from classical ARIMA to advanced deep learning techniques, to find the most reliable forecasting method for a real-world dataset. The analysis emphasizes using conformal prediction to generate reliable prediction intervals, which offer a practical range of expected humidity levels rather than a single, potentially misleading point forecast.