As the scale and complexity of data centers continue to increase, intelligent operation and maintenance (O&M) has become one of the important technologies to ensure high performance and availability. Among them, Key Performance Indicator (KPI) data reflects the health status of data center systems and plays a key role in intelligent O&M. Accurate KPI time series prediction is essential for detecting KPI anomalies and improving the reliability of data center systems. However, traditional time series prediction models have low accuracy when dealing with KPI data. Therefore, this paper proposes a time series prediction model based on contrastive learning and frequency domain attention mechanism to more fully capture the temporal features of KPI data, improve prediction accuracy and reliability. The experimental results show that the proposed model exhibits strong competitiveness in KPI time series prediction.