Periodicity Detection in Time Series Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing periodicity detection methods in time series analysis face challenges due to the non-stationary nature of time series, dynamic and complex periodic patterns, noise, and outliers, leading to poor performance and false positives, especially when the number of underlying periodic patterns is unknown or exceeds one.
Innovation Solution
A periodicity detection system that processes input time series data by removing trends, normalizing, and decomposing it using methods like maximal overlap discrete wavelet transform, then detects periodic components in the frequency domain and validates them in the time domain using statistical tests and autocorrelation analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sophisticated periodicity detection methods are used to handle unknown number of periodic patterns, then detection accuracy may improve, but computational complexity increases and adaptability decreases
Solution Approach 1:
The patent segments the time series data into multiple candidate periodic patterns and processes each separately through a standardized detection pipeline. This allows the system to handle an unknown number of periodic patterns by dividing the complex detection task into manageable segments, improving both accuracy and computational efficiency.
Solution Approach 2:
The patent creates a universal periodicity detection framework that can detect multiple types of periodic patterns (different periods, amplitudes, phases) using the same core algorithm. This multi-functional approach eliminates the need for different sophisticated methods for different patterns, reducing overall system complexity while maintaining high detection accuracy.
2Adaptability or versatility
If multiple periodic patterns are detected simultaneously, then comprehensive analysis is achieved, but false positives increase
Solution Approach 1:
The patent performs preliminary validation of each candidate periodic pattern against statistical criteria and data characteristics before final detection. This preliminary action filters out spurious patterns early in the process, preventing false positives from propagating through the multi-pattern detection system and compromising reliability.
Solution Approach 2:
The patent implements feedback mechanisms where detected periodic patterns are validated against the original time series data and previously detected patterns. This feedback loop allows the system to adjust detection thresholds and parameters dynamically, reducing false positives while maintaining the ability to detect multiple genuine periodic patterns simultaneously.
Data Source
AI summary
Input time series data of a particular time length may be received, and pre-processed to obtain processed time series data. The processed time series data may then be decomposed into a plurality of pieces of time series data of the particular time length at different levels, and a detection of whether a piece of time series data at a particular level includes a periodic component candidate in a frequency domain may be performed. In response to detecting that the piece of time series data at the particular level includes the periodic component candidate in the frequency domain, a validation may be performed to determine whether the periodic component candidate in the piece of time series data at the particular level is a true periodic component of the input time series data in a time domain.