Considering　the　multivariable,　strong-coupling,　multi-conditions　complex　nonlinear　ground　granulated　blast-furnace　slag　(GGBS)　production　process,　this　paper　extracts　three　typical　working　conditions　based　on　massive　process　data.　Multiple　optimal　setpoints　are　obtained　by　resolving　the　multi-objective　problems　under　different　working　conditions.　For　each　condition,　a　data-based　model　is　established　using　the　recurrent　neural　network.　Correspondingly,　multiple　controllers　are　designed　by　the　adaptive　dynamic　programming　method.　Adopting　the　weighted　multiple　model　adaptive　control,　adaptive　control　of　the　GGBS　production　in　multiple　conditions　is　realized.　Integrating　cyber　resources　including　process　operating　optimization,　tracking　control　optimization,　communication,　industrial　Ethernet　and　physical　resource　of　GGBS　production,　a　optimal　control　system　of　GGBS　production　process　is　constructed　based　on　the　cyber-physical　system　(CPS).　Experiment　shows　that　the　proposed　multiple　model　adaptive　control　method　can　achieve　adaptive　control　of　the　GGBS　production　process,　reduce　system　overshoot　and　improve　the　control　quality.　Copyright　©　2019　Acta　Automatica　Sinica.　All　rights　reserved.