In　this　paper,　a　data-driven　predictive　controller　(DDPC)　with　a　fast　closed-loop　response　and　able　to　deal　with　failed　H-bridge　cells　is　proposed　for　multilevel　cascaded　H-bridge　inverters.　The　proposed　strategy　uses　multiple　dynamic　structures　(MDS)　and　the　measured　current　value　for　closed-loop　control.　Each　dynamic　data　is　a　control　signal　associated　with　two　current　responses　and　derived　by　a　PI-controller.　In　a　classic　DDPC,　all　dynamic　data　are　stored　in　a　single　dynamic　structure　(SDS)　independently　of　the　operating　condition.　This　results　in　a　mixing　of　dynamics　and　increases　the　risk　of　using　dynamic　data　at　the　wrong　operating　condition.　In　addition,　without　an　adequate　initialization　at　a　slow　update　rate,　the　controller　will　require　a　period　of　time　to　modified　SDS　and　characterize　the　new　operating　condition　after　the　fault　occurs,　which　results　in　a　large　transient　time.　To　obtain　a　fast　closed-loop　response,　an　offline　initialization　algorithm　is　used　to　define　MDS.　With　a　reliable　MDS,　simulation　results　show　that　the　proposed　DDPC　based　MDS　achieved　a　fast　transient　and　balanced　steady-state　current　response　during　startup　and　post-fault　operations.　©　2021　IEEE.