KPI Curve Waveband Clustering for Sliding Window Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for setting the width of a sliding window for scanning KPI curves are inefficient, leading to increased calculation amounts, introduction of transient noise, and omission of knowledge due to short waveform segments, while longer window settings result in exponential cluster formation and prolonged data processing times.
Innovation Solution
A method that segments the KPI curve into equal-length wavebands, clusters them based on non-time dimensions, extracts fundamental waves, compares waveband data to these waves, and determines the grouping boundary lines to establish the width of the sliding window, facilitating efficient clustering and classification of wavebands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the sliding window is set to have a time length of 1 s, then the KPI curve can be segmented into several segments with the length of 1 s for recognition and comparison, but the waveform segments used for recognition are too short, which directly increases the calculation amount of a label exponentially and introduces transient noise
Solution Approach 1:
The patent changes the time length parameter of the sliding window from the conventional 1 second to an optimized value (0.5-2 seconds) based on the specific characteristics of the KPI curve and business requirements. This parameter optimization balances the segmentation granularity with the calculation burden, avoiding both overly fine segmentation that increases computation exponentially and overly coarse segmentation that loses detail information.
2Loss of information
If the sliding window is set to have a time length of 1 period, then the wavebands can be clustered and grouped in each window, but a plurality of clusters are formed with many short different fundamental wave types, increasing the calculation amount exponentially
Solution Approach 1:
The patent merges adjacent wavebands with similar characteristics into unified clusters by comparing the similarity between wavebands and cluster centers. This merging approach reduces the number of clusters formed, preventing the exponential increase in calculation amount while retaining important waveform patterns that would otherwise be lost in separate short segments.
Solution Approach 2:
The patent optimizes the sliding window time length parameter to balance the number of clusters formed. By adjusting this parameter, the system avoids forming too many small clusters (which increases computation) while ensuring that important waveform patterns are not merged into a single large cluster (which loses information).
3Productivity
If continuous waveform segments are not jointly used as one fundamental wave type, then the KPIs can be classified directly, but the pattern recognition of a whole waveband in the KPI curve is lacking, leading to omission of knowledge
Solution Approach 1:
The patent performs preliminary clustering of wavebands into clusters with similar characteristics before the final classification of KPIs. This preliminary action groups continuous waveform segments that share common patterns, enabling subsequent classification to operate on meaningful aggregated units rather than isolated segments, thus preventing knowledge omission while maintaining classification efficiency.
4Reliability
If the sliding window width is not optimized, then the KPI curve can be processed, but the time from data generation to system alarm is prolonged due to large calculation amount
Solution Approach 1:
The patent optimizes the sliding window time length parameter to a specific range (0.5-2 seconds) that balances computational efficiency with detection reliability. This optimized parameter setting reduces the calculation amount exponentially compared to conventional settings, thereby shortening the time from data generation to system alarm while maintaining reliable anomaly detection capability.
Data Source
AI summary
A KPI curve data processing method is disclosed, involving the following steps: segmenting a KPI curve into several equal-length wavebands, clustering these wavebands based on a non-time dimension, and extracting a fundamental wave for each cluster. The method compares the similarity between each waveband's data and the fundamental wave of its cluster, identifies the cluster's grouping boundary, and groups the wavebands accordingly. It calculates the total time length of consecutive wavebands of the same type in each cluster, using the maximum length as the width of a sliding window. Scanning the KPI curve with this window allows consecutive clusters to be quickly grouped into a single waveform category, reducing computation and enabling integral classification of the wavebands. This process forms distinct waveband chains, aiding in clustering and classification, while minimizing knowledge omission