To　solve　the　problem　that　most　methods　based　on　graph　convolutional　network　(GCN)　only　use　topological　graph　and　ignore　the　structural　information　in　the　feature　space　in　deep　clustering,　a　node　clustering　method　was　proposed　by　introducing　feature　graph　to　make　full　use　of　the　structural　information　in　the　feature　space.　First,an　auto-encoder　(AE)　was　used　to　learn　the　potential　representation　of　node　features,　and　the　node　embeddings　were　obtained　at　the　three　levels　of　feature　graph,　topology　graph　and　node　attribute　at　the　same　time.　Then,　a　fusion　mechanism　was　used　to　fuse　the　learned　node　embeddings.　Finally,　the　network　was　trained　by　self-supervision　to　implement　node　clustering.　A　large　number　of　expiments　on　six　benchmark　datasets　show　that　the　proposed　method　significantly　improves　the　clustering　accuracy.　©　2023　Beijing　University　of　Technology.　All　rights　reserved.