With　the　attractive　features　of　high　efficiency,　low　harmonic,　modularity　and　scalability,　the　Modular　Multilevel　Converter　(MMC)　is　suitable　for　a　wide　range　of　high-voltage　large-capacity　applications.　As　an　advanced　control　strategy,　the　Model　Predictive　Control　(MPC)　can　control　multiple　variables　through　a　cost　function,　which　also　has　advantages　of　direct　modeling　and　fast　dynamic　response　performances.　The　conventional　Model　Predictive　Control　method　for　MMC　is　applied　by　calculating　all　the　candidate　switching　states　to　achieve　control　objective　optimization,　which　limits　the　application　of　MPC　along　with　the　submodule　increasing.　An　Optimized　Model　Predictive　Control　(OMPC)　was　proposed　to　control　the　submodule　voltages,　ac　currents,　circulating　currents　and　the　switching　frequency,　meanwhile　reducing　the　computation　load　from　C2NNto　N+1　for　N+1　level　MMC.　In　addition,　as　the　number　of　the　submodule　increases　to　hundreds,　the　proposed　Grouping-Sorting　algorithm　combined　OMPC　(GSOMPC)　can　further　reduce　the　computation　load　from　N+1　to　2X+M+3(N=M×X)　to　eliminate　the　strict　hardware　requirements.　Experiment　results　based　on　a　2.7　kV/60　kW　of　MMC　with　23-level　back-to-back　dynamic　test　system　verify　the　correctness　and　effectiveness　of　the　OMPC　and　GSOMPC　strategy.　©2014　Chin.　Soc.　for　Elec.　Eng.