Synthetic Time-Series Signals for Privacy-Safe ML Prognostics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge is to develop machine-learning techniques for prognostic-surveillance operations on time-series data from power plants and associated power-distribution systems without the complications of dealing with privacy contracts associated with such data.
Innovation Solution
A system that decomposes original time-series signals into deterministic and stochastic components to produce synthetic time-series signals, which are statistically indistinguishable from the originals, allowing for the development and evaluation of machine-learning techniques without the constraints of privacy contracts, while maintaining the same serial-correlation structure, stochastic content, and spike characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If real time-series data from power plants and distribution systems is used to develop machine-learning techniques, then the evaluation of prognostic-surveillance operations can be performed with authentic data characteristics, but privacy contracts and approval processes complicate and delay the research process
Solution Approach 1:
The patent creates synthetic copies of real time-series sensor data that replicate the statistical properties, temporal patterns, and degradation characteristics of authentic power plant data. These synthetic datasets enable ML research without requiring access to protected real data, thus eliminating privacy approval delays while maintaining evaluation reliability
Solution Approach 2:
The synthetic data generation system acts as an intermediary between the need for authentic data evaluation and the constraints of privacy contracts. By generating intermediate synthetic datasets that bridge these requirements, the system enables ML development without direct access to protected real data
2Adaptability or versatility
If synthetic time-series signals are generated to avoid privacy contracts, then data access becomes unrestricted for ML research, but the synthetic signals must be statistically indistinguishable from real signals to ensure valid ML evaluation
Solution Approach 1:
The patent systematically adjusts and matches statistical parameters (mean, variance, skewness, kurtosis) and temporal characteristics (autocorrelation, cross-correlation, spectral density) of synthetic signals to replicate real data properties. This ensures synthetic data maintains measurement precision while enabling unrestricted access
Solution Approach 2:
The system generates synthetic data with matched statistical properties across multiple dimensions (first-order statistics, second-order correlations, spectral characteristics) to ensure sufficient fidelity for ML evaluation without requiring exact replication of every data characteristic
Data Source
AI summary
The disclosed embodiments relate to a system that facilitates development of machine-learning techniques to perform prognostic-surveillance operations on time-series data from a monitored system, such as a power plant and associated power-distribution system. During operation, the system receives original time-series signals comprising sequences of observations obtained from sensors in the monitored system. Next, the system decomposes the original time-series signals into deterministic and stochastic components. The system then uses the deterministic and stochastic components to produce synthetic time-series signals, which are statistically indistinguishable from the original time-series signals. Finally, the system enables a developer to use the synthetic time-series signals to develop machine-learning (ML) techniques to perform prognostic-surveillance operations on subsequently received time-series signals from the monitored system.