Time-Series Periodicity Analysis for Accurate Subsequence Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Selecting an appropriate subsequence length in subsequence time series clustering is a time-consuming process, leading to incorrect clustering results and reduced estimation accuracy in post-processing analyses such as outlier detection and anomaly detection.
Innovation Solution
A periodicity analysis apparatus and method that calculates the periodicity of time series data using a periodogram and autocorrelation, extracts subsequences of set lengths, categorizes them based on feature values, and performs post-processing analysis to ensure accurate clustering and detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If trial and error approaches or user defined policy are used to select subsequence length, then the selection process can be completed, but it is time-consuming and may lead to incorrect clustering results
Solution Approach 1:
The patent applies preliminary action by calculating the periodicity of the time series data before performing subsequence extraction and clustering. The periodicity calculation unit computes the periodicity using periodogram and autocorrelation methods in advance, and this pre-calculated periodicity information is then used to automatically determine the optimal subsequence length, eliminating the need for time-consuming trial and error approaches.
Solution Approach 2:
The patent replaces the mechanical trial-and-error system with an automated computational system. Instead of manually adjusting subsequence lengths through repeated attempts, the system uses periodicity calculation algorithms (periodogram and autocorrelation) to automatically determine the optimal subsequence length, substituting manual mechanical adjustment with automated mathematical computation.
2Productivity
If incorrect window size is used for subsequence extraction, then the clustering process can be completed quickly, but incorrect clustering results are obtained leading to erroneous information in post-processing analysis
Solution Approach 1:
The patent performs preliminary periodicity calculation before subsequence extraction to determine the correct window size. By calculating the periodicity of the time series data in advance and using it to set the subsequence length, the system ensures that the clustering process uses the optimal window size from the beginning, avoiding both incorrect results and the need for repeated processing.
Solution Approach 2:
The patent implements feedback by using the calculated periodicity information to guide the subsequence extraction process. The periodicity calculation unit provides feedback about the optimal subsequence length to the subsequence extraction unit, ensuring that the extraction process uses the correct window size that reflects the actual periodic characteristics of the data, thereby improving clustering accuracy.
Data Source
AI summary
Provided an apparatus configured to calculate a periodicity of the time series data using at least one of a periodogram of the time series data and an autocorrelation of the time series data; generate a plurality of subsequences from the time series data, a length of each subsequence set to the periodicity, calculate feature values of the plurality of subsequences, and categorize the plurality of subsequences, based on the features values thereof, into one or more groups; and execute analysis of the time series data, based on the categorization result.