A Multi-component Chiller Status Prediction Method Using E-LSTM

Abstract

With the development of intelligent information technology, chiller system composed by different interrelated components has been widely used in industry to cool products and machinery. Predicting the status of chiller system can effectively monitor energy consumption and reduce accident rate. In this paper, we propose an improved LSTM (E-LSTM) method to predict multi-component chiller status. Firstly, a mean filter method is used to preprocess the original multi-component time series data. Secondly, we adopt E-LSTM to extract hidden features from seven component-wise inputs, consisting of outdoor temperature, wet bulb temperature, outdoor enthalpy, L1 & L2 differential pressures, total power, and IT load. Finally, the learned hidden features are fed into a regression layer to predict three future chiller statuses, including PUE, cold source power, and refrigeration secondary pump power, respectively. Experimental results show that the proposed method outperforms the baselines, such as linear regression, SVR, RNN, GRU and LSTM, and hence demonstrate the effectiveness of our proposed method in the task of chiller status prediction.