In　order　to　solve　the　imprecise　problem　of　radial　basis　function　neural　network　(RBFNN)　to　the　sample　output　for　the　sensitive　system,　a　self-augmented　RBFNN　(SA-RBFNN)　is　designed　to　improve　the　accuracy　of　the　model.　Firstly,　the　network　structure　is　constructed　by　using　the　correlation　between　input,　output　and　hidden　layer　neurons　to　increase　the　sensitivity　of　the　network.　Secondly,　an　accelerated　gradient　algorithm　is　used　to　train　SA-RBFNN,　which　improves　the　speed　and　the　precision　of　neural　network　training.　Finally,　the　proposed　SA-RBFNN　is　evaluated　through　a　benchmark　experiment　and　a　practical　problem　in　wastewater　treatment　process.　The　results　indicate　that　the　proposed　SA-RBFNN　can　quickly　converge　to　the　optimal　solution　and　has　a　good　effect　on　more　sensitive　systems.　©　2021　IEEE.