Recently,　Gravitational　search　algorithm　(GSA)　was　considered　as　one　method　for　optimizing　functions　and　solving　real　problems.　For　the　sake　of　better　adjust　the　values　of　recurrent　RBF　neural　network　(RRBFNN)　to　make　the　network　achieve　better　performance,　the　MGSA　is　essential　in　this　article.　The　advised　work　achieves　a　better　compromise　between　exploration　and　development.　At　the　same　time,　by　increasing　the　guidance　of　the　global　optimal　particle,　the　problem　that　the　gravitational　search　algorithm　converges　slowly　in　the　later　iteration　is　solved.　The　Experiment　found　that　the　network　has　better　convergence　speed　and　better　test　accuracy　than　the　RRBFN　optimized　by　the　conventional　optimization　algorithm.　©　2018　IEEE.