Intelligent Data Center Safety Status Prediction based on Algorithm Ensemble

With the promotion of energy conservation in data centers, ensuring the indoor safety while reducing energy consumption has become a social concern. Excessive temperature in the data center may cause equipment damage, thus a technical method is needed urgently to ensure the environment temperature stays in a safe state. Based on the requirement for temperature warning, this paper proposed an intelligent data center safety prediction method based on model fusion. The experimental data comes from the intelligent data center project in China Telecom. Most recent works only used a single-model-based method such as SVR, LR or LGBM to predict temperature. This paper first uses fusion-model-based method, integrating three single regressors (SVR, LR and LGBM) with the best parameters by soft voting. The performance of fusion model has shown lower RMSE with 0.086, compared with the RMSE of the three single regressors. It finds out that the ensembled model reduces the impact caused by the misjudgment of single weak regressors, which is stable and reliable for intelligent data center safety prediction.