Time-Series Condition Prediction Using Segmented Sequence Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting the remaining useful life of devices from time-series data face challenges in accurately capturing both long-term gradual tendencies and short-term changes, leading to difficulties in specifying necessary information for reliable predictions.
Innovation Solution
An information processing system and method that employs a trained model to extract features from time-series data by dividing it into partial data segments, generating multiple vectors through embedding and transformation, and ultimately transforming these vectors into a value representing the device's condition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If time-series data is divided into partial data and statistical values are extracted for prediction, then the prediction process becomes structured and manageable, but the ability to capture both long-term gradual tendencies and short-term changes is lost, reducing prediction accuracy
Solution Approach 1:
The patent divides time-series data into multiple pieces along the time axis and extracts features from each piece. This segmentation allows the system to systematically process both long-term trends (across multiple segments) and short-term changes (within each segment) while maintaining prediction accuracy.
Solution Approach 2:
The patent transforms extracted features into vector representations and processes them through multiple components that operate in different dimensional spaces. This dimensional transformation enables the model to capture complex patterns including both gradual tendencies and abrupt changes that cannot be represented in the original time-series format.
2Extent of automation
If feature values are extracted from time-series data for remaining useful life prediction, then the prediction can be performed using machine learning, but the difficulty in specifying necessary information increases when feature values appear in various forms
Solution Approach 1:
The patent creates a universal feature extraction framework that can handle various forms of feature values (long-term gradual tendencies, short-term changes, and intermediate patterns) through a single integrated system. The multiple components work together to process diverse feature types automatically, eliminating the need for manual specification of different feature forms.
Solution Approach 2:
The patent introduces intermediate vector representations as mediators between raw time-series data and final predictions. These vectors serve as a universal intermediate format that can encode various types of feature information, allowing the machine learning model to process diverse feature forms without requiring explicit specification of each feature type.
3Measurement precision
If multiple feature extraction methods are used to capture different time-scale patterns, then prediction accuracy improves, but the system complexity and computational burden increase
Solution Approach 1:
The patent merges multiple feature extraction operations into an integrated multi-component system. Instead of using separate independent methods for long-term and short-term feature extraction, the patent combines them into a unified architecture where components work together to extract and process features across different time scales simultaneously, reducing overall system complexity.
Solution Approach 2:
The patent ensures continuous processing of time-series data through its multi-component architecture, where each component processes features at different stages of the time scale continuum. This continuous approach captures patterns across all time scales without requiring discrete, separate processing steps, thereby reducing computational burden while maintaining accuracy.
Data Source
AI summary
An information processing system predicts a condition of a device from time-series data acquired from the device, by using a trained model. When predicting, the information processing system allows the trained model to extract features that depend on the sequence from pieces of partial time-series data obtained by dividing the time-series data along the time axis, generate first vectors in which the extracted features are embedded, each of the first vectors corresponding to each of the pieces of the partial time-series data one to one, generate a second vector in which the first vectors are embedded, extract features that depend on the sequence from the first vectors, generate a third vector in which the extracted features are embedded, generate a fourth vector in which the second vector and the third vector are embedded, and transform the fourth vector into a first value that represents a condition of the device.