Leveraging AI to Identify the Root Causes of PMU Data Latency and Loss for Control Room

The wide area measurement system (WAMS) of the Unified Power System of Russia includes over 1300 PMUs and PDCs. Significant operational experience has been gained with PMU data quality software (PDQ-software), which calculates data loss percentage, delivery latency, PMU frame drop count and monitors the dynamics of PMU data quality change. These metrics allow for the accurate determination of the causes of PMU data quality degradation. The paper describes the core operational principles of the PDQ-software and provides practical examples of data quality degradation identification. The constantly expanding scale of the data acquisition system, coupled with the multitude of mutually influencing nodes in the information transmission path, has necessitated the development of artificial intelligence (AI) tools, as manual diagnosis of anomalies has become exceedingly labour-intensive due to the high number of system elements.

The paper outlines the conceptual approaches to developing AI-tools for data quality monitoring. A key element of the AI-module is the application of recurrent neural networks, specifically LSTM (Long Short-Term Memory), for the deep analysis of time series quality parameters (loss, latency, drop count) for each data stream and for aggregated node metrics.

Additionally, machine learning techniques, such as clustering, are used to group similar anomalies. The Root Cause Analysis (RCA) module, utilizing anomaly data from the LSTM and other detectors, as well as the object’s topological model, not only generates and ranks hypotheses about the root causes—providing explanations for its conclusions and visualization of the degraded path—but also localizes the origin of the issue. Mechanisms for dynamic thresholding and Human-in-the-Loop learning are implemented to enable the system to adapt to the changing network infrastructure.

Specific practical examples demonstrate how the proposed AI-based approaches contribute to identifying the causes of data quality deterioration.