Machine Learning Model Development to Predict Power Outage Duration (POD): A Case Study for Electric Utilities.

Area of Science:

• Electrical Engineering
• Computer Science
• Data Science

Background:

• Climate variability and grid complexity challenge power outage management.
• Effective outage duration prediction is crucial for electric utilities.
• Real-time customer feedback is essential for managing outage impacts.

Purpose of the Study:

• To develop a novel predictive model for power outage durations.
• To enhance electric utility outage management and customer communication.
• To improve electric grid resilience and reliability through advanced analytics.

Main Methods:

• Utilized machine learning algorithms: decision tree (DT), random forest (RF), k-nearest neighbors (KNN), and extreme gradient boosting (XGBoost).
• Employed historical sensor and non-sensor outage data from a Turkish electric utility.
• Applied minimum redundancy maximum relevance (MRMR) for feature selection with XGBoost.

Main Results:

• The XGBoost model with MRMR achieved 98.433% accuracy in predicting outage durations.
• This represents a 12.922% improvement over state-of-the-art methods (average 85.511% accuracy).
• The model demonstrated adaptability to diverse grid structures and outage causes.

Conclusions:

• Machine learning offers a practical solution for enhancing power outage management.
• The developed model significantly improves prediction accuracy and grid reliability.
• This approach transforms electric utility responses and customer communication during outages.