Short-term Prediction Model of Data Center Cooling Energy Consumption Based on SEN-MH-BiLSTM

With the continuous expansion of data center (DC) scales, the issue of energy consumption has become increasingly prominent. Accurate short-term prediction of cooling energy consumption in DCs is of great significance for optimizing DC energy management, reducing operational costs, and promoting the green development of the DC industry. This paper proposes a fusion model that combines SENet, multi-head attention mechanism, and BiLSTM for short-term prediction of cooling energy consumption in DCs. The model utilizes SENet to enhance the feature representation capability, captures long-term dependencies in time series data through BiLSTM, and introduces multi-head attention mechanism to fully explore the intrinsic correlations between energy consumption and electric power, temporal feature variables, thereby improving prediction accuracy. Experimental analysis is conducted based on actual data from a DC in Gansu Province, and the results demonstrate the effectiveness of the proposed prediction method. Comparative test results with various deep learning prediction models show that the proposed method effectively improves the prediction accuracy of DC cooling energy consumption.