Detection and confirmation of energy theft in advanced metering infrastructure using long short-term bidirectional memory and fuzzy inference system models
TL;DR
By combining machine learning, anomaly detection, and fuzzy logic, the BiLSTM-based framework showed an improvement in detection accuracy and reduced non-technical losses, making it an important tool for improving utility operations and ensuring fair energy distribution.
Abstract
Background: Energy theft is a major issue in modern power systems, contributing to significant non-technical losses, reduced utility revenue, and instability in grid operations. With the deployment of advanced metering infrastructure (AMI) in smart grids, large volumes of consumer data are now available; this creates opportunities to apply artificial intelligence techniques for detecting fraudulent electricity usage. Aim: The aim of this article is to develop an intelligent framework for detecting and confirming energy theft using bidirectional long short-term memory (BiLSTM) network, combined with anomaly detection and a fuzzy inference system to improve accuracy and reduce false positives. Methods: The developed methodology used smart meter data from the London dataset to train a BiLSTM model for time-series forecasting of household energy consumption. Prediction errors from the model were analyzed using an anomaly detection approach to identify suspicious patterns. The fuzzy inference system, incorporating AMI-related parameters such as intrusion detection signals and observer meter readings, was then applied to confirm instances of energy theft. Results: The BiLSTM model demonstrated superior performance compared to conventional LSTM models, achieving an improvement of 10.24% in terms of RMSE,a 0.83% improvement in MAE, and a 322.14% improvement in the coefficient of determination (R²). The integration of anomaly detection with fuzzy confirmation effectively improved detection accuracy while minimizing false alarms. Conclusion: The developed BiLSTM-based framework provides a reliable and efficient solution for detecting and confirming energy theft in smart grids. By combining machine learning, anomaly detection, and fuzzy logic, the system showed an improvement in detection accuracy and reduced non-technical losses, making it an important tool for improving utility operations and ensuring fair energy distribution.