데이터센터 열환경 예측 및 적응형 제어알고리즘 개발

Abstract

Purpose: As cooling energy accounts for 50.5% of the data center total energy use, it is necessary to reduce cooling energy with optimal control method. Thus, this study aimed at developing an adaptive control algorithm for data center thermal environment based on Artificial Neural Network (ANN). Method: A data center thermal environment model was built to obtain variable data which is used to train and evaluate performance of the prediction model and algorithm. The thermal environment prediction model was developed using ANN and optimization process was conducted by the Bayesian Optimization Algorithm. The adaptive control algorithm, which embedded the prediction model, adopted the Sliding Windows method and was optimized to maximize the control performance. Result: The performance evaluation of the developed algorithm was conducted compared with non-adaptive algorithm. As a result, the adaptive algorithm presented better performance than the non-adaptive with 0.45 of RMSE. Therefore, the developed algorithm secured the stability and accuracy and will be applied to supervisory control platforms.