Scholarlink

In this article, we propose a model for the structural analysis of Systems of Systems (SoS) by leveraging deep learning techniques, with a focus on Graph Neural Networks (GNN). The primary objective is to propose a Convolutional GNN model capable of analyzing the structural operability of interdependencies between component systems (CS) within the SoS and accurately predicting the critical CS and the interdependencies that degrade performance in terms of the structural operability of the SoS. In this study, we used the Convolutional GNN algorithm to analyze a comprehensive dataset by leveraging the features of the nodes (CS) and the relationships between them (weighted edges). The model was rigorously tested, yielding remarkable results: 79% accuracy during testing after training and 54% accuracy during evaluation. The results could help provide a better understanding of the advantages and limitations of this approach for this type of task and pave the way for new methods to improve and optimize SoS performance.

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

Prediction by Deep Learning - GNN of Interdependencies within the SoS

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Younes Koulou,
  • Anas El Attaoui,
  • Norelislam El Hami

摘要

In this article, we propose a model for the structural analysis of Systems of Systems (SoS) by leveraging deep learning techniques, with a focus on Graph Neural Networks (GNN). The primary objective is to propose a Convolutional GNN model capable of analyzing the structural operability of interdependencies between component systems (CS) within the SoS and accurately predicting the critical CS and the interdependencies that degrade performance in terms of the structural operability of the SoS. In this study, we used the Convolutional GNN algorithm to analyze a comprehensive dataset by leveraging the features of the nodes (CS) and the relationships between them (weighted edges). The model was rigorously tested, yielding remarkable results: 79% accuracy during testing after training and 54% accuracy during evaluation. The results could help provide a better understanding of the advantages and limitations of this approach for this type of task and pave the way for new methods to improve and optimize SoS performance.