State Classification Using Persistent Homology Attractors

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for classifying states based on multidimensional time series data, such as invariant analysis and subspace methods, are inadequate for non-linear time series and can lead to false classifications, especially when all variables change simultaneously, making it difficult to sense changes accurately.

Innovation Solution

The proposed solution involves generating an attractor from multidimensional time series data using a persistent homology process to create Betti number sequences, which reflect the topological features of the data, allowing for accurate classification and change detection by comparing these sequences over time.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If invariant analysis is used to monitor time correlation of multidimensional time series data, then the method can be practiced easily, but it cannot sense changes when all variables change in the same direction simultaneously

Engineering Contradiction:
Improveease of practiceVSAvoidchange detection accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent transforms the time series data into a higher-dimensional space by creating delay coordinates (x(t), x(t-τ), x(t-2τ), ...) to form vectors in n-dimensional space. This dimensional transformation allows the system to detect changes that are invisible in the original time domain, particularly when all variables change simultaneously, by observing changes in the geometric structure (attractor) of the data in the expanded phase space.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Measurement precision

If subspace method is used to generate orthogonal basis from multidimensional time series data, then it enables detection of change of subspace in density, but it is not suitable for non-linear time series

Engineering Contradiction:
Improvesubspace change detection capabilityVSAvoidapplicability to non-linear data
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent changes the fundamental parameter of analysis from linear orthogonal bases to topological invariants (Betti numbers) that characterize the geometric structure of the attractor. By using persistent homology to compute Betti numbers across different scales, the method adapts to non-linear time series while maintaining the ability to detect subspace changes, as Betti numbers capture the essential topological features regardless of the underlying dynamics linearity.

Inventive Principle:
Principle #35Parameter changes

3Reliability

If traditional analysis methods are used on multidimensional time series data with unknown properties, then false classification may occur, but checking data properties in advance requires additional work

Engineering Contradiction:
Improveclassification accuracyVSAvoidtime for property checking
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent creates a universal analysis framework based on topological data analysis that can handle multiple types of time series data (linear, non-linear, periodic, chaotic) with a single methodology. The attractor-based approach with persistent homology serves as a multi-functional tool that automatically adapts to different data characteristics without requiring preliminary property assessment, eliminating the need for separate analysis methods for different data types.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP3425561B1State classifying program, state classifying method, and state classifying device
Publication Date: 2023.10.11 FUJITSU LTD
  • EP3425561B1 patent drawingFigure 1~3
  • EP3425561B1 patent drawingFigure 4~5
  • EP3425561B1 patent drawingFigure 6~7

AI summary

A state classifying method includes generating an attractor containing a plurality of points that correspond to a plurality of sets of time series data, coordinate values of each of the plurality of points being values corresponding to the sets of time series data; generating Betti number sequence data by applying a persistent homology process on the attractor; and classifying a state that is represented by the plurality of sets of time series data based on the Betti number sequence data.