p12 / Conference
Towards Detecting Non-Technical Losses in Electrical Power Systems with Machine Learning
Energy Transition in the Mediterranean Area (SyNERGY MED 2024), 1-5 · 2024
In plain language
This study uses Greek electricity-distribution data to distinguish normal consumption from patterns associated with non-technical losses.
Research context
Learning from traditional-meter records can support better targeting of loss-detection efforts.
Approach
Preprocessing and feature engineering address an imbalanced HEDNO dataset. Models include Naive Bayes, logistic regression, trees, random forests, XGBoost, and CatBoost.
Findings
Ensemble methods perform best in the reported experiments, particularly CatBoost.
Summary based on abstract.