A　novel　online　self-organizing　fuzzy　neural　network　(FNN)　based　on　the　improved　Levenberg-Marquardt　(ILM)　learning　algorithm　and　singular　value　decomposition　(SVD)　is　proposed　to　predict　the　effluent　total　phosphorus　(TP)　in　a　wastewater　treatment　process.　The　centers　and　widths　of　membership　functions　and　weights　of　output　layer　are　trained　by　ILM　learning　algorithm.　Meanwhile,　the　output　matrix　of　the　rule　layer　is　decomposed　with　SVD,　which　is　implemented　by　one-sided　Jacobi＇s　transformation.　The　neurons　of　rule　layer　are　adjusted　dynamically　with　growing　and　pruning　algorithms,　which　are　based　on　the　singular　values.　In　addition,　the　convergence　of　the　proposed　ILM--SVDFNN　has　been　proved　both　in　the　structure　fixed　phase　and　the　structure　adjusting　phase.　Finally,　the　validity　and　practicability　of　the　model　are　illustrated　with　three　examples,　including　typical　nonlinear　system　identification,　Mackey-Glass　time　series　prediction,　and　prediction　of　effluent　TP.　Simulation　results　demonstrate　that　the　proposed　ILM--SVDFNN　generates　a　fuzzy　neural　network　automatically　and　effectively　with　a　highly　accurate　and　compact　structure,　and　it　can　well　satisfy　the　detection　accuracy　and　real-time　requirements　of　the　prediction　of　effluent　TP.　©2017,　Editorial　Department　of　Control　Theory　&　Applications　South　China　University　of　Technology.　All　right　reserved.