Health intelligent evaluation based on knowledge graph multi-set pooling
To extract more comprehensive features in time domain and space domain from equipment sensor monito-ring data and other related heterogeneous data,a health intelligent evaluation method based on knowledge graph multi-set pooling was proposed.A Health Temporal Knowledge Graph(HTKG)was constructed to fuse spatiotem-poral features of monitoring data,component data and priori knowledge.The overall spatiotemporal features of HT-KG were embedded into graph-level representation vectors with a graph pooling network,which included node fea-ture learning,first level graph pooling,self-attention feature learning and second level graph pooling.The health e-valuation was transformed into a graph classification problem based on representation learning.The proposed meth-od had been evaluated on public engine datasets.Experiments results showed that the method could achieve high e-valuation accuracy and also shows good stability in few-shot situations.
health evaluationgraph neural networkknowledge graphspatiotemporal featuresgraph pooling
NETL
NSTL
万方数据
