In　the　municipal　solid　waste　incineration　(MSWI)　process,　it　is　critical　to　predict　furnace　temperature,　which　is　closely　related　to　the　incinerate　state　and　the　steam　production,　to　maintain　the　high　efficiency　in　the　incineration　process.　In　this　paper,　a　novel　self-organizing　TS　fuzzy　neural　network　with　an　improved　gradient　descent　algorithm　(SOTSFNN-IGA)　is　developed　to　predict　furnace　temperature.　Firstly,　to　get　a　suitable　network　structure　and　achieve　high-efficiency　computing　capability,　the　error　criteria　and　activity　intensity　are　employed　to　grow　and　remove　the　fuzzy　rules　of　SOTSFNN-IGA　automatically.　Secondly,　an　improved　gradient　descent　algorithm　is　employed　to　adjust　the　parameters　of　SOTSFNN-IGA.　Thirdly,　the　convergence　analysis　of　the　proposed　SOTSFNN-IGA　is　given　through　the　Lyapunov　theory.　Subsequently,　to　understand　the　influence　of　each　variable　on　the　furnace　temperature,　a　new　variable　importance　measurement　method　is　employed.　Finally,　the　proposed　SOTSFNN-IGA　is　verified　based　on　several　benchmark　nonlinear　systems　and　a　furnace　prediction　in　the　MSWI　process.　Experimental　results　demonstrate　that　the　developed　SOTSFNN-IGA　has　better　advantages　in　prediction　accuracy　than　other　algorithms,　which　prediction　accuracy　and　NSE　coefficient　are　as　high　as　99.85%　and　0.9827　respectively　in　the　furnace　temperature　prediction.