Insulator contamination is a growing threat to utility reliability and wildfire prevention, particularly under critical rainfall conditions that create conductive paths leading to flashovers. At PG&E, analysis from 2015 onward revealed that contamination-induced flashovers account for over 10% of outages, 6% of ignitions, and 30% of wildfire risk.

Building on machine learning models that predict contamination and flashover risks, PG&E has implemented a robust MLOps pipeline to move from proof-of-concept to production. This session details how their MLOps framework continuously monitors data quality, model drift, and prediction performance across multiple data sources—including weather, asset properties, inspection records, and contamination history.

Learn how this scalable solution improves ROC-AUC and PR-AUC metrics, empowers inspection teams with real-time alerts, and guides targeted mitigation strategies such as washing schedules. This is a must-attend for those seeking to apply MLOps to operationalize AI for grid safety and wildfire prevention.

Takeaways:

• Understand the risk and scale of contamination-induced flashovers in utility systems.
• Apply machine learning techniques to predict insulator contamination and guide proactive maintenance.
• Implement a full MLOps pipeline to monitor data integrity, automate retraining, and maintain model performance.
• Leverage real-time analytics and dashboards to support field inspections and optimize washing schedules.
• Reduce wildfire ignition risk and improve reliability through data-driven decision-making and AI at scale.