The　traditional　model　predictive　control　uses　only　one　switching　vector　for　every　per　sampling　cycle,　which　causes　the　variable　switching　frequency　in　rolling　optimization　and　makes　the　filter　difficult　to　design.　In　this　paper,　the　problem　of　rolling　optimization　is　analyzed　in　detail　by　the　variable　switching　frequency　and　cost　function.　Some　optimization　methods　are　put　forward　according　to　the　performance　requirements,　which　provides　a　new　idea　for　the　development　of　model　predictive　control(MPC).　©　2019　IEEE.