Aiming　at　the　uncertainty　caused　by　large-scale　wind　power　integration,　a　multi-stage　data-driven　robust　unit　commitment　model　is　proposed.　The　non-parametric　confidence　band　of　cumulative　distribution　function　for　the　wind　power　forecast　error　is　obtained　with　the　Dirichlet　method,　and　the　conservative　degree　and　the　correlation　control　parameters　are　introduced　to　construct　the　dynamic　ambiguity　set　of　the　high-dimensional　uncertainty.　On　this　basis,　combined　with　the　sequential　decision-making　characteristics　of　the　power　system　scheduling　process,　the　single　period　during　the　real-time　dispatch　is　regarded　as　a　scheduling　stage　and　a　multi-stage　robust　unit　commitment　model　based　on　the　above　dynamic　ambiguity　set　is　established.　According　to　the　multi-layer　nested　structure　of　this　model,　a　linear　affine　rule　is　developed,　and　then　the　model　is　transformed　into　a　mixed-integer　linear　programming　problem　to　be　solved　by　adopting　the　robust　counterpart.　The　simulation　is　implemented　on　the　power　systems　with　different　scales.　The　decision-making　by　the　traditional　two-stage　robust　unit　commitment　model　is　taken　for　comparison.　Moreover,　the　effects　of　the　operation　parameters　of　the　power　grid　and　generation　units,　the　control　parameters　of　conservative　degree　and　the　correlation　are　considered　to　verify　the　effectiveness　of　the　proposed　model.　©　2022,　Power　System　Technology　Press.　All　right　reserved.