A Multi-Scale Collaborative Framework for Server Power Prediction in Data Centers: Long-Term Model with Enhanced NKAN-LTS

This paper presents a multi-scale collaborative framework for data center server power consumption prediction, which categorizes forecasting models based on application scenarios to meet grid demand response needs for both peak shaving and frequency regulation. Furthermore, it introduces the NKAN-LTS model, which enhances non-linear feature extraction and long-term forecasting by integrating NLinear's residual decomposition with Kolmogorov-Arnold Networks (KANs). Extensive experiments on public datasets show that the proposed approach significantly outperforms benchmark models in MAE and RMSE, demonstrating its potential for optimizing data center energy management and supporting grid demand response strategies.