Time Series Forecasting Dataset from Power and Business Services In TFMixformer

To better reflect real-world service forecasting scenarios, we utilized two practical time series datasets rather than relying on standard public benchmarks. Specifically, we selected (1) Alibaba Service Metrics, containing fine-grained metrics from over 20,000 microservices in a production cluster, and (2) Power System Service Logs, comprising operational request counts and response times from a large-scale energy infrastructure. Five datasets were constructed focusing on HTTP and RPC workloads, with one-minute sampling intervals. These datasets provide diverse, high-variability patterns suitable for evaluating time-frequency modeling in service-level forecasting tasks.