Multivariate Sensor Time-Series Visualization via Similarity Weighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to efficiently combine and process multivariate time-series data sequences of varying lengths into a representative form, especially when resources are limited, leading to issues like information loss and reduced predictive performance.
Innovation Solution
A method that calculates similarity values between data sequences, weights them based on similarity, and concatenates segments to form a representative data sequence, using techniques like Pearson correlation and dynamic time warping to handle sequences of different lengths.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If multiple data sequences of varying lengths are combined using existing methods, then data integration is achieved, but information loss occurs and predictive performance decreases
Solution Approach 1:
The patent transforms the data sequences by changing their parameter representation - specifically, it standardizes sequences of varying lengths into a common dimensional space while preserving their essential characteristics through mathematical transformation, allowing meaningful comparison and combination without information loss
Solution Approach 2:
The patent performs preliminary processing of the data sequences before combination, including normalization and alignment operations that prepare the sequences for effective integration, ensuring that when they are combined, their predictive value is maintained rather than lost
2Reliability
If data sequences are processed in detail to preserve information, then predictive performance improves, but computational complexity increases
Solution Approach 1:
The patent extracts only the essential and relevant features from the complex data sequences, separating the critical information needed for prediction from redundant or less important data, thereby maintaining predictive performance while reducing computational burden
Solution Approach 2:
The patent divides the processing into distinct segments or stages - first transforming individual sequences, then combining them, and finally processing the integrated result - which allows each stage to be optimized independently, reducing overall computational complexity while preserving predictive accuracy
3Adaptability or versatility
If data sequences of different lengths are aligned, then comparability improves, but temporal correlations may be distorted
Solution Approach 1:
The patent employs dynamic transformation methods that adapt the alignment process to the specific characteristics of each data sequence, allowing the transformation to flexibly handle varying lengths while preserving the temporal relationships inherent in each sequence through dynamic scaling or warping techniques
Data Source
AI summary
In accordance with some embodiments, systems, methods, and media for processing and representing multivariate sensor information gathered from multiple sources are provided. In some embodiments, the method comprises: receiving a data sequences from respective sources; identifying a shortest data sequence that corresponds to a first interval; calculating, for pairs of data sequences, a similarity value over the first interval; forming a first segment of a representative data sequence that is a weighted combination of the data sequences over the first interval based on similarity values associated with the data sequences; truncating at least a subset of data sequences to exclude elements corresponding to the first interval; identifying a shortest data sequence corresponding to a second interval; forming a second segment of the representative data sequence based on similarity values associated with the data sequences; and concatenating the first segment and the second segment of the representative data sequence.