Graphical Auto-Encoder Based Fault Cases Clustering with Multi-Feature Fusion

Addressing the challenge of potential multiple and complex correlations between fault cases that are difficult to discern and thus fail to effectively guide similar fault handling, this study is dedicated to mining and analyzing the intrinsic connections between fault cases by employing advanced graph clustering techniques, and develops a highly efficient clustering model to improve the utilization of knowledge from historical fault cases. Firstly, a graph auto-encoder based on the characteristics of neighbor nodes is constructed, in which two GraphSAGE layers and an attention layer are introduced to fully integrate the neighbor information within two hops. Subsequently, a self-training clustering module is constructed, and the total objective function is defined through a joint optimization strategy to aggregate the cases with high similarity into the same cluster. Finally, comparative experiments are conducted to validate the model in the field of fault cases clustering effectiveness, and compared with the mainstream model, the accuracy (ACC), normalized mutual information (NMI) and adjusted Rand index (ARI) evaluation indexes are significantly improved.