The　modeling　for　fermentation　process　has　important　significance　in　achieving　control　and　optimal　control　of　the　fermentation　process.　Generalization　capability　of　the　model　based　on　global　learning　support　vector　machine　was　not　strong,　so　according　to　local　learning　theory　the　method　of　establishing　the　fermentation　process　dynamic　model　was　proposed　in　this　paper.　The　dynamic　of　the　fermentation　process　model　was　realized　through　establishing　the　fermentation　process　dynamic　sample　sets.　The　method　was　used　in　the　penicillin　fermentation　process　and　the　process　of　Escherichia　cob　producing　interleukin-2.　Experimental　results　showed　that　the　established　dynamic　model　had　a　higher　accuracy　and　a　better　dynamic　adaptability　compared　with　the　static　SVM　model.