Scholarlink

The accurate classification of the status of the Pressurized Water Reactor (PWR) Nuclear Power Plant (NPP) is crucial to ensuring its operational safety and efficiency. This study evaluates the performance of Deep Learning (DL) models alongside other Machine Learning (ML) models under different feature selection strategies. After data standardization, Principal Component Analysis (PCA) was applied for data visualization. PCA shows that the selection of eight principal components (PCs) retains nearly 90% of the cumulative variance ratio. Although PCA preserved most of the variability in the dataset, feature selection proved to be a more effective approach to ensure greater separability between normal and abnormal status, thus improving classification accuracy. Among the models evaluated, Multi Layer Perceptron (MLP) achieved the highest accuracy, followed by 1D Convolutional Neural Network (1D CNN). This work demonstrates a significant improvement in classification accuracy, achieving 97.99% using only the top seven selected features, compared to 93.98% reported in a previous study that employed an MLP with a different architecture and all available features. The results highlight the importance of selecting relevant features to enhance the performance of the ML models, and demonstrate the potential of the DL models for classifying the status of PWR NPPs while maintaining computational efficiency.

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

Deep Learning-Based Effective Anomaly Detection in PWR-Type Nuclear Power Plants

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Ihtesham Ibn Malek,
  • S. M. Shohorab Hossain Hasib,
  • Md. Farhan Shadiq

摘要

The accurate classification of the status of the Pressurized Water Reactor (PWR) Nuclear Power Plant (NPP) is crucial to ensuring its operational safety and efficiency. This study evaluates the performance of Deep Learning (DL) models alongside other Machine Learning (ML) models under different feature selection strategies. After data standardization, Principal Component Analysis (PCA) was applied for data visualization. PCA shows that the selection of eight principal components (PCs) retains nearly 90% of the cumulative variance ratio. Although PCA preserved most of the variability in the dataset, feature selection proved to be a more effective approach to ensure greater separability between normal and abnormal status, thus improving classification accuracy. Among the models evaluated, Multi Layer Perceptron (MLP) achieved the highest accuracy, followed by 1D Convolutional Neural Network (1D CNN). This work demonstrates a significant improvement in classification accuracy, achieving 97.99% using only the top seven selected features, compared to 93.98% reported in a previous study that employed an MLP with a different architecture and all available features. The results highlight the importance of selecting relevant features to enhance the performance of the ML models, and demonstrate the potential of the DL models for classifying the status of PWR NPPs while maintaining computational efficiency.