Scholarlink

<p>The incorporation of advanced Machine Learning (ML) techniques into power grid security has become increasingly vital in ensuring the reliable operation and resilience of critical infrastructure. Critical infrastructures are increasingly vulnerable to a range of security risks, such as cyberattacks and operational abnormalities, due to their increased interconnectedness and reliance on sophisticated communication technology. Efficiently managing and securing the vast amounts of data generated by power grid systems is paramount for maintaining their integrity and functionality. This study proposed a novel method for detecting anomalies in power grids by utilizing a hybrid model of Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) that has been enhanced by a Genetic Algorithm (GA). The proposed CNN-LSTM architecture efficiently captures both temporal and spatial features of security data, while the GA ensures optimal hyperparameter selection, enhancing the model's performance. We compare our method with conventional techniques and show significant enhancements in anomaly identification accuracy, recall, precision, and F1 score employing an Industrial Control System (ICS) Dataset for smart grid anomaly identification. The experimental outcomes demonstrate that the proposed CNN-LSTM-GA model attains a recall at 96.5%, accuracy at 98.5%, precision at 97.2%, and an F1 score of 96.8%. These metrics significantly surpass those of traditional models such as SVM (accuracy: 92.3%), RF (accuracy: 94.1%), LSTM (accuracy: 96.0%), and Extreme Gradient Boosting (XGBoost) (accuracy: 95.5%). This approach not only improves the detection of malicious activities and operational anomalies but also enhances the overall security posture of power grid infrastructures. Despite the promising results, the proposed model's performance may be affected by variations in data quality and real-time operational conditions. Future work will focus on enhancing model robustness through additional optimization techniques and evaluating its performance under diverse and dynamic smart grid scenarios.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced security data management in power grids using deep learning with genetic algorithm-based hyperparameter optimization

  • 期刊论文

  • 发表日期：

    2025年3月7日

查看全文
  • Xianqi Cao

摘要

The incorporation of advanced Machine Learning (ML) techniques into power grid security has become increasingly vital in ensuring the reliable operation and resilience of critical infrastructure. Critical infrastructures are increasingly vulnerable to a range of security risks, such as cyberattacks and operational abnormalities, due to their increased interconnectedness and reliance on sophisticated communication technology. Efficiently managing and securing the vast amounts of data generated by power grid systems is paramount for maintaining their integrity and functionality. This study proposed a novel method for detecting anomalies in power grids by utilizing a hybrid model of Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) that has been enhanced by a Genetic Algorithm (GA). The proposed CNN-LSTM architecture efficiently captures both temporal and spatial features of security data, while the GA ensures optimal hyperparameter selection, enhancing the model's performance. We compare our method with conventional techniques and show significant enhancements in anomaly identification accuracy, recall, precision, and F1 score employing an Industrial Control System (ICS) Dataset for smart grid anomaly identification. The experimental outcomes demonstrate that the proposed CNN-LSTM-GA model attains a recall at 96.5%, accuracy at 98.5%, precision at 97.2%, and an F1 score of 96.8%. These metrics significantly surpass those of traditional models such as SVM (accuracy: 92.3%), RF (accuracy: 94.1%), LSTM (accuracy: 96.0%), and Extreme Gradient Boosting (XGBoost) (accuracy: 95.5%). This approach not only improves the detection of malicious activities and operational anomalies but also enhances the overall security posture of power grid infrastructures. Despite the promising results, the proposed model's performance may be affected by variations in data quality and real-time operational conditions. Future work will focus on enhancing model robustness through additional optimization techniques and evaluating its performance under diverse and dynamic smart grid scenarios.