Effect of Check Meter Quantity on Theft Detection in Distribution Networks with Rooftop PV and Net Metering

The increasing demand for renewable energy has led to the rise of rooftop photovoltaics (PV) and net metering (NM). However, these complicate energy theft detection, necessitating a more complex approach. This paper explored the effect of varying the number of check meters (CM) in distribution networks with rooftop PV and NM on theft detection algorithms, specifically for detecting meter tampering in households. Using the Ausgrid Dataset and IEEE European Low Voltage Test Feeder in OpenDSS and Python, power flow simulations were conducted with varying levels of PV and NM penetration, and the number of houses connected per CM. Features extracted from the simulated data include Gamma Deviance (GD), Log Cosh Loss (LCL), Percent Loss Error (PLE), and Poisson Deviance (PD). These were used as input to three machine learning algorithms: Support Vector Machine (SVM), Artificial Neural Networks (ANN), and Decision Tree (DT); their test accuracies were used to asses their performances. Kendall's Tau correlation test assessed the correlation between the theft detection accuracy and CM quantity. LCL and PD exhibited a moderate to high correlation between theft detection accuracy and CM quantity at a 99% confidence level, and are more robust to varying levels of PV and NM penetration for all algorithms. This provides insights into the development and implementation of theft detection systems, reducing electricity theft risk and promoting sustainable energy transition.