PG&E is developing a machine learning model to improve prediction of contaminated insulators Insulators on transmission structures are susceptible to contamination from various sources, such as avian excrement, dust, emissions, smoke, and gases.

Under critical meteorological conditions, like dense fog or light precipitation, these contaminants can form a conductive pathway, leading to dry band arcing and flashovers. These flashover events can result in outages and ignitions, making the early detection and mitigation of contamination critical to reliability and wildfire risk reduction.

PG&E will illustrate how they developed a machine learning model utilizing on a broad set of input datasets based on physical processes to explore logistic regression, random forest, and XGBoost models that direct inspection teams to identify contaminated insulators and plan mitigation strategies for insulator washing.

PG&E session participants will learn about:

• The risks associated with insulator contamination.
• The inputs used to develop a machine learning model that predicts contaminated insulators.
• The top predictive features indicative to identifying contaminated insulators in the model.
• How the model improves reliability and reduces fire risk.