A Hybrid Deep Learning Framework for Detecting Electricity Theft Cyber-Attacks in Renewable Distributed Generation Systems

Abstract

The increasing integration of renewable distributed generation (RDG) into modern smart grids has significantly improved energy efficiency and sustainability. However, this transition has also introduced new vulnerabilities, particularly in the form of cyberattacks and electricity theft. Unauthorized manipulation of smart meter data and communication channels can lead to substantial economic losses, compromised grid stability, and reduced reliability of energy distribution systems. Traditional detection mechanisms are often inadequate due to their inability to handle large-scale, dynamic, and heterogeneous data generated by smart grids.This research proposes a hybrid deep learning-based framework for detecting electricity theft and cyber-attacks in renewable distributed generation environments. The system integrates multiple machine learning and deep learning techniques, including Feed Forward Neural Networks (DNN), Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), and an ensemble model combining CNN with Random Forest classification. The objective is to leverage the strengths of each model to improve detection accuracy and robustness.The proposed system processes smart meter datasets containing consumer energy usage patterns. Data preprocessing techniques such as handling missing values, label encoding, normalization, and feature selection are applied to ensure data quality and consistency. Temporal and spatial patterns in electricity consumption are captured using GRU and CNN architectures, respectively. The GRU model is particularly effective in identifying sequential anomalies in time-series consumption data, while the CNN model extracts spatial correlations among features. The hybrid CNN-Random Forest model further enhances classification performance by utilizing deep feature representations for traditional ensemble learning.Performance evaluation is conducted using standard metrics such as accuracy, precision, recall, and F1-score, along with Receiver Operating Characteristic (ROC) analysis. Experimental results demonstrate that the hybrid CNNRandom Forest model outperforms standalone models, achieving higher detection rates and reduced false positives. The system also provides a graphical user interface (GUI) for ease of interaction, enabling users to upload datasets, train models, visualize results, and predict electricity theft instances.The proposed framework offers a scalable and efficient solution for real-time electricity theft detection in smart grids. By incorporating advanced deep learning techniques and hybrid modeling strategies, the system enhances the security and reliability of renewable energy distribution networks. This research contributes to the development of intelligent energy management systems capable of mitigating cyber threats and ensuring sustainable energy utilization.