This　paper　basically　investigates　the　enhancement　of　a　modified　non-square　direct　matrix　converters　(MC)　for　renewable　microgrid　applications　using　generative　adversarial　networks　(GAN).　Such　a　deep　learning　technique　which　benefits　from　two　multi-layer　adversary　perceptions　is　used　to　enhance　the　security　of　the　data　in　the　microgrid.　This　paper　also　provides　a　new　application　of　the　non-square　direct　MC　in　the　microgrid　system　which　is　able　to　provide　balanced　output　with　any　desired　amplitude　and　frequency　under　unbalanced　condition　specifically　in　the　case　of　using　renewable　energy　sources　such　as　photovoltaics　(PVs)　and　wind　turbines.　In　addition,　a　new　modified　social　spider　optimization　(MSSO)　algorithm　is　introduced　to　help　improving　the　training　process　of　GAN.　Simulation　results　show　that　matrix　converter　based　on　GAN　makes　it　possible　to　convert　any　input　voltage　to　the　desired　output　voltage　which　leads　to　the　elimination　of　the　back　to　back　converter　of　wind　turbine.　Copyright　©　2020　IEEE.