A Bayesian-Optimized BiLSTM Model for Day-Ahead Power Forecasting of Data Center

The expansion of data centers, driven by 5G and artificial intelligence adoption, significantly increases energy consumption. Note utilizing clean energy is crucial for data centers to reduce emissions and costs under carbon neutrality goals. However, an intrinsic conflict exists between clean energy intermittency, computing demand volatility, and power supply reliability requirements. Accurate load forecasting is therefore critical for optimizing power management and enhancing supply reliability. This research develops a high-precision load forecasting model employing a bidirectional long short-term memory neural network. Bayesian optimization further enhances prediction accuracy by tuning model hyperparameters. Case validation confirms that the proposed model achieves high-precision dayahead power forecasting at 15-minute resolution.