Non-technical loss detection in power distribution networks using machine learning.

Area of Science:

• Electrical Engineering
• Data Science
• Machine Learning

Background:

• Non-technical losses (NTL) in power distribution, including electricity theft, result in billions of dollars in annual financial losses for utility companies.
• Detecting NTL is challenging due to imbalanced datasets, where instances of NTL are significantly outnumbered by legitimate consumption data.

Purpose of the Study:

• To evaluate and compare various machine learning algorithms and data balancing techniques for effective NTL detection.
• To identify the optimal combination of algorithms and data balancing methods for maximizing detection accuracy and precision.

Main Methods:

• Seven data balancing techniques were applied: ADASYN, Random Over Sampling, Random Under Sampling, Near Miss Under Sampling, Borderline-SMOTE, SMOTE-ENN, and SMOTE-Tomek.
• Seven classification algorithms (Decision Tree, Logistic Regression, XGBoost, Random Forest, SVM, Naïve Bayes, KNN) were tested in a two-stage model, with the second stage incorporating data balancing.
• Performance was evaluated using accuracy, precision, recall, F1 score, and Matthews Correlation Coefficient (MCC) at a 70%-30% training-testing ratio.

Main Results:

• The Random Forest algorithm combined with Random Over Sampling demonstrated superior performance.
• This combination achieved 98.03% accuracy and 99.02% precision, outperforming existing methods in the literature.
• All performance improvements were statistically validated, showing significant enhancements at the 95% confidence level.

Conclusions:

• Machine learning, particularly the Random Forest algorithm coupled with Random Over Sampling, offers a highly effective solution for detecting non-technical losses in power distribution.
• The proposed method significantly improves the accuracy and precision of NTL detection, addressing the challenge of imbalanced data.
• This approach provides a robust framework for utilities to mitigate financial losses caused by electricity theft and other NTLs.