Power　system　transformation　is　crucial　in　a　dual-carbon　context.　In　this　paper,　we　propose　an　optimized　Transformer　architecture,　Romember　model,　which　improves　the　accuracy　of　power　consumption　prediction　by　introducing　rotational　position　coding.　We　construct　a　multidimensional　evaluation　framework　to　provide　guidelines　for　the　continuous　optimization　of　the　model.　The　Romember　model　contributes　to　balanced　grid　dispatch,　optimal　load　allocation,　and　the　promotion　of　energy　green　transition.　©　2024　IEEE.