Data centers realize the intelligence of operation and maintenance processes with the help of intelligent operation and maintenance technology, which contains key performance indicator (KPI) such as server workload, network traffic, page access, CPU usage, etc. Accurate prediction of data center KPI sequence data at a certain moment in the future will help data centers to reasonably allocate and dynamically use computing resources. Thus, the stable and healthy operation of the data center system is guaranteed. Due to the strong nonlinear characteristics of data center KPI sequence data, the accuracy of existing statistical-based methods, machine learning-based and deep learning-based methods is not high. In this paper, we integrate CNN, CBAM, GRU, FECAM, and KAN to propose a time series data prediction model named CCGFK for predicting data center KPI metrics at a certain moment in the future. Experiments are conducted using API access logs from a data center in Shanghai, spanning from March 1, 2023 to October 27, 2023, and the experimental results show that the CCGFK model outperforms the compared models in RMSE, MAPE, MAE, and R2 prediction and evaluation metrics, and relatively improves the R2 metrics by 3.62% compared to the current optimal model TS2Vec, and relatively reduces the R2 metrics by 33.33%, 26.53%, and 49.75% in MAE, RMSE, MAPE metrics, and MAPE metrics. MAPE metrics are relatively reduced by 33.33%, 26.53%, and 49.75%. The CCGFK model proposed in this paper provides an effective solution for sequence data prediction in data center intelligent O&M KPI.