Featuring　high　efficiency,　low　harmonic　distortion,　high　modularity　and　scalability,　the　modular　multilevel　converter　(MMC)　is　particularly　suitable　for　high　voltage　direct　current　transmission　applications.　As　an　advanced　control　strategy,　model　predictive　control　(MPC)　has　the　advantage　of　direct　modeling　and　fast　dynamic　response.　It　can　simultaneously　control　multiple　variables　through　an　appropriate　cost　function.　The　conventional　MPC　can　achieve　an　optimal　control　objective　by　evaluating　all　the　candidate　switching　states　for　the　MMC;　however,　with　increasing　number　of　submodules,　there　is　an　increasing　number　of　candidate　switching　states　that　place　an　enormous　burden　on　the　control.　In　this　paper,　a　grouping-sorting-optimized　MPC　(GSOMPC)　strategy　is　proposed　for　the　MMC　with　the　number　of　submodules　for　each　arm　increases　to　hundreds.　It　divides　all　submodules　of　each　arm　into　M　groups,　with　each　containing　X　submodules.　By　the　implementation　of　the　first　level　and　second　level　optimized　MPC　between　groups　and　submodules,　respectively,　the　computational　load　of　each　phase　decreases　from　C-2N(N)　to　2X　+　M　+　3(N　=　M　x　X).　In　addition,　to　reduce　the　strict　requirements　of　control　hardware　for　sorting　and　calculation,　the　proposed　strategy　is　able　to　simultaneously　control　the　submodule　voltage,　ac　current,　circulating　current,　and　switching　frequency.　Applied　to　a　2.7-kV/60-kW　MMC　back-to-back　dynamic　test　system,　experimental　results　verify　the　feasibility　and　effectiveness　of　the　proposed　GSOMPC　strategy.