OMCI model similarity computation based on graph neural networks.

The optical network unit management and control interface (OMCI) is a crucial protocol for interconnectivity between the optical line terminal (OLT) and optical network unit (ONU) in Gigabit- capable passive optical networks (GPON) systems. When addressing OMCI interoperability issues, developers often need to conduct exception analysis on OMCI service models. However, due to the complexity of OMCI domain knowledge, directly analyzing OMCI service models can be extremely challenging, time-consuming, and laborious for inexperienced developers. To address these challenges, the OMCI model exception analysis method based on graph neural networks (GNNs) is proposed. This method leverages graph similarity computation algorithms to search for similar OMCI models from a database as references, compares the differences, and identifies anomalies. Firstly, real OMCI data is structured into graph data. Subsequently, the similarity graph neural network (SimGNN) is improved by integrating graph isomorphism networks and self-attention pooling for fast graph similarity computation. Finally, the similarity scores between each graph in the OMCI graph database and the anomalous graph data are calculated, and the most similar OMCI service model graphs are recommended based on the score ranking. Experimental results show that the improved graph similarity computation model outperforms the baseline model on the OMCI dataset used in this study. Moreover, it proves effective in practical applications, offering valuable assistance in analyzing OMCI interoperability issues. [ABSTRACT FROM AUTHOR]