The　effective　multi-step　forecasting　of　the　main　steam　flow　in　municipal　solid　waste　incineration　(MSWI)　can　help　the　steam　power　generation　module　to　make　decision　responses　in　advance　and　control　the　system　effectively.　As　the　main　steam　flow　is　affected　by　multiple　process　flows　and　the　strong　coupling　between　each　process　flow　in　the　MSWI　system,　there　are　numerous　variables　affecting　the　main　steam　flow,　and　the　variable　redundancy　is　high,　and　the　noise　is　high.　This　results　in　low　accuracy　of　steam　flow　prediction.　To　solve　the　above　problem,　a　multi-step　prediction　method　based　on　principal　component　analysis　(PCA)　and　Pyraformer　model　is　proposed.　Based　on　the　multi-modular　MSWI　process　and　low　coupling　between　modules,　we　divided　the　variables　affecting　the　main　steam　flow　into　four　parts　and　carried　out　PCA　dimension　reduction　treatment.　The　processed　data　was　input　into　the　Pyraformer　model　to　obtain　the　multi-step　predicted　value　of　the　main　steam　flow.　Pyraformer　is　a　time　series　prediction　model　based　on　Transformer,　which　can　capture　different　range　of　time　dependencies　and　has　better　computational　efficiency　and　prediction　effect　for　prediction　tasks　with　more　variable　input.　Experiments　show　that　the　proposed　method　has　high　prediction　accuracy　and　stability,　which　can　provide　a　better　basis　for　advanced　decision　making　of　steam　power　generation　tasks　and　other　processes　of　MSWI.