A Combined LSTM-CNN Model for Abnormal Electricity Usage Detection

Non-technical losses remain a significant challenge for electricity providers. The advent of smart grids and sophisti-cated measurement infrastructures has enabled the use of data-driven approaches to identify unusual power usage patterns, thereby mitigating these losses. Various machine learning models have been successfully applied to detect anomalies in electricity usage. However, the sophistication of tactics like power theft and the rapid growth of consumption data present ongoing challenges to anomaly detection. To address this, we introduce an combined LSTM-CNN model designed to identify abnormal electricity usage. This hybrid model, consisting of both LSTM and CNN components, adeptly processes time-series electricity usage data. Our experiments show that the proposed LSTM-CNN surpasses current methods, with a precision of 93.9%, a recall of 95.6%, an F1-score of 0.947, and an accuracy of 95.3% on the SGCC dataset.