Energy Theft Detection in IoT-based Smart Grid Using Deep Learning
The development of smart devices and the integration of Internet of Things (IoT) technology into modern energy systems have enabled real-time acquisition of high-frequency smart meter data, which supports the development of sustainable and efficient power grids. This real time data monitoring enhances operational efficiency and optimal utilization of energy. However, these smart devices, such as smart meters, increase the complexity of the system, which introduces challenges in identifying anomalies such as energy theft, meter malfunctions, and irregular consumption patterns, which negatively impact energy sustainability. Energy theft is one of the major concerns because it leads to unnecessary non-technical losses and inefficient utilization of electrical resources. This paper presents a method to detect energy theft with the help of machine learning and deep learning techniques. The techniques implemented are Support Vector Machine (SVM) and Convolutional Neural Network (CNN). These techniques are used to analyze real-world energy consumption patterns. The available dataset was imbalanced due to the scarcity of energy theft cases. To address the imbalance in smart meter datasets, synthetic data is generated by assuming certain conditions that may represent the energy theft scenarios. The models that are proposed here are trained on one year data taken from smart meters to accurately learn and predict load patterns reflecting actual consumer behaviour. To validate these proposed models and to evaluate the performance of models (SVM & CNN), quality indices such as the confusion matrix, accuracy, precision, and recall metrics are utilized. The experimental results demonstrate that the CNN model presents better results than the SVM, particularly in detecting minority classes associated with energy theft. The accuracy observed for the CNN model is 98%, and a recall of 87 %. These results validate the CNN model for detecting energy theft cases, which is actually a non-technical loss, hence supporting sustainable, reliable, and environmentally responsible smart grid operations.