Online Framework Models Time-Series as Dynamic Mixtures of Time-Delay Systems
An innovative online framework conceptualizes streaming time series as evolving mixtures of time-delay systems, tackling issues in adaptive modeling. This approach enhances the robustness of model tracking while minimizing memory consumption by condensing previous regimes into a fixed-length representation that encapsulates system dynamics and input-output delays. By utilizing the Markov parameter series of the system, it creates a summary system tensor that reflects both dynamic behavior and delay traits. This research has been made available on arXiv under the ID 2605.26191.
Key facts
- The research addresses adaptive modeling in time-series data streams with clear input-output relationships.
- Rapid system changes (regime shifts) caused by environmental factors or input delay changes degrade model performance.
- The trade-off among accuracy, robustness, and memory usage arises when using multiple small models for each time-series pattern.
- The framework treats streaming time series as dynamic mixtures of time-delay systems.
- It maintains robustness of model tracking and reduces memory usage.
- Past regimes are summarized using a fixed-length representation capturing system dynamics and input-output delays.
- A summary system tensor is constructed using the system's Markov parameter series.
- The paper is available on arXiv with ID 2605.26191.
Entities
Institutions
- arXiv