Scientific Seminar on Mathematical Modeling, Data Analysis, and Information Security of Complex Systems — Invited Talk
Performance Time Series Analysis in the Multiple Baseline Framework
Abstract
Performance time series (PTS) encode information on system state evolution, anomaly patterns, and structural changes, and are therefore essential for proactive system maintenance and reliability assessment. Real-world systems often operate across multiple normal modes, each corresponding to a distinct performance baseline, rendering the data-generating process structurally non-stationary. The core challenge lies in distinguishing normal mode switching from genuine anomalous behavior.
Given the continuous and streaming nature of performance indicators, this work argues that sequence analysis provides a natural methodological foundation for performance time series analysis. PTS analysis research should focus on three core components: (1) multi-baseline modeling and switch detection; (2) multi-baseline-aware anomaly detection mechanisms; and (3) interpretable continuous health scoring for system state quantification.