An Integrated Feature Selection and Deep Learning Method for Energy Efficiency Prediction in Data Center Cooling Systems

This study introduces a novel method to predict energy efficiency in data center cooling systems, combining feature selection with a deep learning model. The approach uses a three-step feature selection process—mRMR, XGBoost, and NSGA-II—to optimize input features and hyperparameters, enhancing accuracy while minimizing sensor data requirements. The resulting deep neural network (DNN) processes time-series data without steady-state assumptions. Validated with real operational data, the method achieves a test R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> score of 0.9838, surpassing traditional models like SVR and LSTM. Results confirm its robustness, generalizability, and potential to reduce sensor deployment, advancing practical implementation in energy-efficient cooling systems.