Time Series Data Search Device Using Segmented Sample Sets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for searching time-series data are computationally intensive and often result in increased approximation errors, making it challenging to ensure that the approximation error remains within an acceptable range, especially when using sample segments for similarity searches.
Innovation Solution
A time-series data search device that generates a sample segment set by dividing the initial segment set into initial temporary segment sets and selecting sample segments based on a condition radius, ensuring that the approximation error is within a specified range, thereby reducing computational complexity and improving search efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all combinations of training segments and test segments are compared to find singularities, then detection accuracy is improved, but computational complexity increases considerably
Solution Approach 1:
The patent divides the training time-series data into multiple training segments and the test time-series data into multiple test segments. By comparing segments rather than entire datasets, the computational complexity is reduced from O(N*M) to O((N/W)*(M/W)) where N and M are the lengths of training and test data, and W is the segment width. This segmentation allows exhaustive comparison to be performed on manageable chunks while maintaining detection accuracy.
Solution Approach 2:
The patent performs exhaustive comparison of all segment combinations within each temporary segment set, which is a partial application of the full exhaustive search. By limiting the comparison scope to segments within the same temporary segment set (rather than all possible combinations across the entire dataset), the method achieves sufficient detection accuracy with reduced computational burden.
2Device complexity
If sample segments are generated by integrating similar segments to reduce computation, then computational complexity is reduced, but approximation error increases
Solution Approach 1:
The patent pre-generates training temporary segment sets by integrating similar training segments before the actual singularity detection process. This preliminary action creates a compressed representation of the training data that reduces the number of segments to be compared during testing, thereby reducing computational complexity while controlling approximation error through the use of multiple temporary segment sets.
Solution Approach 2:
The patent changes the parameter of segment representation by creating multiple temporary segment sets with different integration levels. Each temporary segment set contains segments that have been integrated to different extents, allowing the system to balance between computational complexity and approximation error by selecting appropriate temporary segment sets for comparison.
3Device complexity
If the number of sample segments is narrowed down to reduce computation, then computational complexity is reduced, but approximation error cannot be ensured within allowable range
Solution Approach 1:
The patent segments the training data into multiple temporary segment sets, where each set contains a manageable number of integrated segments. This segmentation allows the system to work with a reduced number of segments in each comparison operation while maintaining overall accuracy through the collective coverage of multiple temporary segment sets.
Solution Approach 2:
The patent uses the structure of temporary segment sets to provide feedback control over approximation error. By organizing segments into temporary sets with specific integration properties, the system ensures that the approximation error remains within acceptable bounds while achieving computational efficiency.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An initial segment set generation unit (120) generates an initial segment set (F) from training time-series data (S). An initial segment set sorting unit (130) sorts initial segments included in the initial segment set by the order of feature quantities. A sample segment set generation unit (140) generates a sample segment set (E) from the initial segment set. A sample segment set sorting unit (150) sorts sample segments included in the sample segment set by the order of the feature quantities. A search result distance calculation unit (160) calculates a search result distance (Z) for each of test segments by using the sample segment set.