Based　on　dynamic　neural　network-Boltzmann　machines,　a　new　optimal　forecast　control　method　of　penicillin　fermentation　processes　is　proposed.　First,　according　to　the　input　and　output　data　of　the　fed-batch　fermentation　processes,　the　generalized　predictive　control　(GPC)　model　is　built　by　system　identification　tools.　Secondly,　adopting　the　dynamic　neural　network-Boltzmann　machine　as　an　optimal　controller,　combining　with　the　penicillin　fermentation　process　and　its　GPC　model,　this　paper　structures　the　receding　optimization　closed　loop.　This　method　implements　the　three　taches　of　GPC:　multistep　prediction,　recursive　optimization,　feedback　emendation　Simulation　experiment　results　show　that　using　this　method　the　outcome　concentration　can　increase　twenty-five　percent,　and　this　control　method　is　effectual.