元學(xué)習(xí)混合模型預(yù)測(cè)中國(guó)生豬價(jià)格方法研究

打開文本圖片集

中圖分類號(hào)：F323.7 文獻(xiàn)標(biāo)志碼：A 文章編號(hào)：2096-9902（2025）13-0001-05

Abstract：Accuratelypredictingpigpricefluctuationsisof greatsignificance tomaintaining thebalanceofmarket supply anddemand，optimizingproductiondecisionsandensuringthestableoperationoftheindustrialchain.Aimingattheshortcoming oftraditionalpredictionmethodsininsuffcientgeneralizationabilityinsmallsamplescenarios，thispaperconstructsahybrid predictionframeworkbasedonSeasonal-TrendDecompositionusingLOESS（STL）andModel-AgnosticMeta-Learning（MAML）. Specificall：First，STLdecompositionisusedtodecouplethetimeseriesdataofpigpricesintothreecomponents：trendterms， seasontermsandresidualterms，whichefectivelyenhances featureinterpretabilityandsuppresesnoiseinterference；thena dual-stagemeta-learningmechanismisdesigned-inthebasictrainingstage，LSTMandGRUnetworkarchitecturesareusedto conductmuti-taskpre-trainingonpigandporkpricestolearnsharedfeaturerepresentationsofpricefluctuationsacrossbreds; Intherapidadaptationstage，therapidmigrationofthemodelinsmallsampletargetscenariosisachievedthroughcolaborative optimizationofMAMLsinternalloopparameterfine-tuningandexterallopmeta-parameterupdate.Empiricalresultsshowthat thismethodhassignificantimprovementoverthebenchmarkmodelintermsofroot-mean-squareeror（RMSE）andmeanabsolute eror（MAE）indicators，providinganinterpretableandtransferabledecisionanalysistolfortheconstructionofagricultural economic early warning systems.

Keywords： pig price prediction; meta-learning; MAML; STL; LSTM和機(jī)器學(xué)習(xí)方法。（剩余8409字）