Time-Series Cycle Detection via Multiclustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining cyclical patterns in time-series data struggle with variations, leading to inaccurate threshold settings and false alarms, as they fail to account for both normal and abnormal yet expected fluctuations, especially in data sets with periodic events like stock market activities.
Innovation Solution
The method involves transforming and clustering time-series data into discrete time periods, comparing similar intervals to identify cycles, and forming multiclusters to determine cycles, which allows for the derivation of thresholds that account for both day-to-day and periodic variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional methods are used to determine cyclical patterns in time-series data, then the process is simple, but the accuracy of threshold settings is poor and false alarms increase
Solution Approach 1:
The patent segments time-series data into discrete time periods and further divides each period into clusters based on similarity. This segmentation allows the system to analyze patterns at multiple granularities (time periods, clusters within periods), improving threshold accuracy by capturing both macro cyclical trends and micro variations without overwhelming complexity through hierarchical organization
Solution Approach 2:
The patent introduces an additional dimension by organizing data not only by time sequence but also by similarity-based clustering within each time period. This creates a two-dimensional structure (time period × cluster), enabling the system to account for both temporal cyclical patterns and intra-period variations, thereby improving measurement precision while managing complexity through structured dimensionality
2Reliability
If traditional cycle detection methods are used, then the computational process is fast, but the system produces false alarms due to inability to account for expected fluctuations
Solution Approach 1:
The patent performs preliminary clustering and organization of time-series data into discrete time periods and clusters before actual cycle detection and threshold determination. This preliminary structuring pre-processes the data to highlight patterns and variations, making subsequent cycle detection more efficient and reliable while reducing false alarms, thus balancing reliability improvement with acceptable processing time
Solution Approach 2:
The patent creates simplified representations (clusters) that copy essential characteristics of complex time-series data within each time period. These clustered copies capture the dominant patterns and variations without retaining all original data complexity, enabling faster and more reliable cycle detection by working with condensed representations that preserve critical information
Data Source
AI summary
Cycles and other patterns within time-series data are determined. Time-series data are transformed into discretized sets of clustered data that are organized by time period. Comparison is made of the organized data to determine similar time periods and multiclusters of the similar time periods are formed. From the multicluster data, cycles are identified from which thresholds and other useful data may be derived, or the data used for other useful purposes.