Data Center Application Traffic Load Prediction Based on DeTiDE: Decomposition Time-Series Dense Encoder

This paper introduces a precise traffic prediction model that combines a Time-series Dense Encoder with a decomposition architecture. It simplifies time series features, extracts temporal features, and predicts load traffic with a stacked linear decoder. The model also includes a globally decomposed linear connection to enhance accuracy. Trained on Ant Group's data, it outperforms other models in prediction error and reduces mean square error by 28%. It also offers faster training time, demonstrating high accuracy and efficiency. This proposed traffic load prediction model is expected to improve operational efficiency, thereby saving costs and reducing energy consumption in data centers.