邻接矩阵
计算机科学
可解释性
图形
推论
算法
有向图
理论计算机科学
图论
人工智能
数据挖掘
数学
组合数学
作者
Nan Liu,Wei Wang,Chong Wang,Jian Liu
标识
DOI:10.1049/icp.2023.1679
摘要
This paper introduces a novel approach to fault diagnosis in high-speed train braking systems through the integration of graph neural networks and causality analysis. Specifically, a fault diagnosis model is formulated based on a graph diffusion model. Cyclic causality is extracted from the data using Function Causality Inference (FCI), leading to the construction of a causality graph. Subsequently, the adjacency matrix is transformed into a diffusion transfer matrix using the graph diffusion model, enabling the aggregation of higher-order information. A graph classification model, employing Graph Attention Networks (GAT), is then established to evaluate the effectiveness of the proposed fault diagnosis model. Through comprehensive comparative analysis, the performance enhancement of the high-speed train fault classifier achieved by this approach is verified. The results underscore that our fault diagnosis model not only enhances the interpretability and reasoning capabilities of cyclic causality, but also adeptly captures the high-dimensional structural information inherent in the graph. This amalgamation leads to a more precise and pragmatic fault diagnosis model.
科研通智能强力驱动
Strongly Powered by AbleSci AI