PMU Data Anomaly Detection Using Singular Value Decomposition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In power transmission systems, phasor measurement units (PMUs) generate data streams with errors, missing, or inaccurate data, which if not detected and corrected, can lead to improper operator actions and system instability.
Innovation Solution
A computer system that uses singular value decomposition (SVD) to detect anomalies in data streams from PMUs, calculates replacement values, and validates these values by comparing them with other data streams to ensure accuracy and reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data streams from PMUs are transmitted to PDC for system analysis, then operators can analyze system conditions and stability, but errors and anomalies may occur during collection or transmission leading to erroneous data
Solution Approach 1:
The system performs preliminary anomaly detection and correction before data reaches the operator decision-making process. By calculating expected data values using system models and comparing them with actual measurements, the system proactively identifies and corrects anomalies in advance, preventing erroneous data from affecting operator actions
Solution Approach 2:
The patent introduces an intermediary data validation layer between PMU data collection and PDC analysis. This intermediary system uses system models, historical data, and statistical methods to filter and verify incoming data, acting as a mediator that separates useful information from erroneous measurements
2Measurement precision
If operators manually verify data accuracy, then erroneous data can be detected, but response time increases and operator workload increases
Solution Approach 1:
The system implements self-service data validation where the data processing system automatically verifies its own data quality using embedded models and algorithms. The system autonomously detects anomalies, calculates corrected values, and validates results without requiring manual operator intervention, thereby maintaining high verification accuracy while eliminating time loss and reducing workload
Solution Approach 2:
The patent establishes continuous feedback loops where system models predict expected data values, compare them with actual measurements, and automatically adjust or flag discrepancies. This automated feedback mechanism enables rapid, continuous data verification without manual intervention, maintaining precision while minimizing time loss
3Manufacturing precision
If data correction algorithms are applied to fix bad data, then data accuracy improves, but false corrections may occur if validation is insufficient
Solution Approach 1:
The system performs preliminary validation checks before applying corrections by comparing corrected values against multiple criteria including system physical constraints, historical patterns, and alternative calculation methods. This preliminary validation ensures that corrections are physically plausible and consistent with system behavior before they are applied to the data stream
Solution Approach 2:
The patent introduces multiple intermediary validation layers between data correction and final output. These include model-based validation, statistical outlier detection, and cross-verification with related system measurements. Each intermediary layer acts as a checkpoint to prevent false corrections from propagating through the system
Data Source
AI summary
In one aspect, a computer system for managing occurrences of data anomalies in a data stream is provided. The computer system includes a processor in communication with the data stream. The processor is programmed to receive a first data stream from a phasor measurement unit. The processor is also programmed to calculate at least one singular value associated with the first data stream. The processor is further programmed to detect a first data anomaly within the first data stream using the at least one singular value. The first data anomaly occurs during a first time segment. The processor is also programmed to indicate the first time segment as containing the first data anomaly.