Machine Learning Model for Real-time Prediction of Influx of Critical Corrective Maintenance Work on the Grid

Given the nature of emergency corrective work, there is a great deal of planning that goes into the resources needed to remediate them so that utilities avoid a significant event. Nevertheless, asset failures are inevitable, and emergent corrective maintenance will affect grid reliability regardless of how well predictive maintenance strategies perform. The integration of distributed energy resources (DERs) and the changing weather patterns, variable load demands, and accelerated equipment stress have challenged how asset investments are determined to improve grid reliability and avoid unplanned outages. Transformers have been particularly sensitive to these conditions, resulting in increased thermal loading and a higher-than-anticipated failure risk. This paper aims to explore the use of two machine learning models to predict peaks in emergency corrective maintenance (ECM) occurrences by analyzing weather forecasts and

PJM (Pennsylvania-New Jersey-Maryland Interconnection) load demand and seeks to improve responsiveness to uncontrollable environmental impacts on the grid while accounting for the added stress from distributed energy resources, which further burden transformers and increase maintenance needs. The paper compares regression and classification models built on the same variables to determine their value for proactive resource optimization and improved maintenance response.