Recurrent　fuzzy　neural　network　(RFNN)　are　widely　used　with　nonlinear　system　modeling.　However,　the　modeling　ability　of　RFNN　is　usually　compromised　due　to　the　presence　of　uncertain　external　disturbances　and　changing　unknown　environments.　To　address　this　problem,　a　self-organizing　recurrent　fuzzy　neural　network　with　modified　Levenberg-Marquardt　algorithm　(MLM-SORFNN)　is　proposed　for　nonlinear　systems　modeling.　Firstly,　a　dynamic　adjustment　mechanism　of　the　network　structure　based　on　correntropy　is　proposed　to　improve　the　network　ability　to　adapt　to　uncertain　environments.　Secondly,　an　improved　LM　algorithm　with　adaptive　learning　rate　is　designed,　which　can　improve　the　modeling　accuracy　while　ensuring　the　convergence　of　the　network.　Finally,　the　experimental　results　demonstrate　the　superior　modeling　capability　of　the　MLM-SORFNN.　©　2024　IEEE.