The　fermentation　parameters　are　not　usually　optimal,　which　easily　lead　to　low　fermentation　unit　and　high　output　fluctuation.　Therefore,　the　aim　of　this　article　is　to　develop　a　method　of　how　to　find　the　optimal　trajectories　of　fermentation　parameters.　Vitamin　B-12　is　a　necessity　for　human　body　and　animal,　however,　at　the　present,　it　is　produced　with　low　output　and　expensive　price;　as　a　result,　it　is　cried　for　increasing　output.　Optimization　can　be　obtained　through　precise　mathematic　model,　however　due　to　the　complexity　and　high　non-linearity　of　fermentation　process,　most　simple　mathematical　models　cannot　describe　the　characteristic　of　bio-systems　very,　well.　The　microbial　fermentation　usually　goes　step　by　step.　The　best　operating　conditions　of　each　step　are　different;　so　the　optimization　of　each　step　needs　to　be.　sought　respectively.　This　paper　founded　respectively　neural　network　models,　which　are　able　to　realize　multi-step　pre-estimate,　for　the　biomass　concentration,　the　substrate　concentration,　and　the　product　concentration　of　Vitamin　B-12　fermentation　process.　Based　on　the　models,　genetic　algorithm　gained　the　optimal　control　trajectories　of　fermentation　temperature　and　pH　by　setting　different　objective　functions　in　different　fermentation　phases　to　seek　optimization.　Putting　the　optimal　operating　conditions　into　practice　makes　the　fermentation　unit　an　obvious　increase.