An　improved　online　error　minimized-　extreme　learning　machine　(IOEM-ELM)　adaptive　control　method　is　proposed　by　introducing　the　adding　and　pruning　mechanism　of　hidden　nodes　to　realize　the　control　of　a　kind　of　MIMO　system　with　multiple　operating　modes.　The　strategy　to　handle　the　multiple-operating-modes　problem　is　analyzed　by　the　idea　of　multiple　model　adaptive　control.　Further,　considering　that　the　ground-granulated　blast-furnace　slag　(GGBS)　production　process　is　a　complex　system　with　the　characteristics　of　multiple　operating　modes,　high　nonlinearity,　strong　coupling,　and　high　uncertainty,　a　data-driven　intelligent　control　scheme　is　designed　based　on　the　proposed　IOEM-ELM　neural　network.　By　analyzing　the　numerous　production　data　produced　in　normal　and　abnormal　situations,　three　typical　operating　modes　are　extracted　to　fully　depict　the　actual　production　process　as　a　testing　platform.　As　the　network　structure　adjusts　dynamically　using　the　IOEM-ELM　method,　model,　and　controller　are　designed　to　deal　with　the　GGBS　production　process　operating　among　multiple　modes.　The　example　shows　that　the　proposed　method　can　handle　changing　modes,　and　reduce　the　computation　of　GGBS　production　process　effectively.