This　paper　investigates　the　multi-dimensional　non-linear　exponential　smoothing　prediction　method　under　different　weight　coefficients,　so　as　to　realize　the　short-term　prediction　of　the　original　wind　power　output.　Based　on　the　current　research　on　wind　power　fluctuation　prediction,　this　paper　firstly　proposes　a　new　multi-dimensional　nonlinear　exponential　smoothing　prediction　method.　Based　on　different　weight　coefficients,　the　corresponding　short-term　forecast　volatility　is　obtained.　Then,　the　wavelet　packet　decomposition　method　is　used　to　further　realize　the　secondary　smoothing　of　the　prediction　wind　power　fluctuations.　Hybrid　energy　storage　system　(HESS)　plays　an　important　role　in　wind　power　fluctuation　suppression　from　different　frequency　bands.　However,　the　traditional　fixed　frequency　power　allocation　method　significantly　degrades　the　performance　of　the　accuracy　of　the　algorithm.　Therefore,　this　paper　proposes　a　frequency　conversion　entropy　strategy,　which　combines　energy　value　with　wavelet　packet　decomposition.　By　calculating　the　energy　components　of　different　frequency　bands,　the　fluctuation　can　be　divided　effectively.　Finally,　simulation　and　experiment　discussions　show　that　the　proposed　algorithm　in　this　paper　can　realize　the　power　allocation　effectively　compared　to　the　traditional　one.　And　it　further　verifies　the　feasibility　and　reliability　of　the　strategy　proposed　in　this　work.　©　2022　The　Authors.　IET　Renewable　Power　Generation　published　by　John　Wiley　&　Sons　Ltd　on　behalf　of　The　Institution　of　Engineering　and　Technology.