基于深度残差Bi-LSTM的风电功率预测

打开文本图片集

Abstract：Deep learning models typicall outperform traditional machine learning models in windpower forecasting.With theincreaseinthenumberofnetworklayers，performancegainsareoftenhinderedbytheproblemofnetworkdegradation.On thisbasis，awind powerforecastingtechnologycombining thedeepresidualstructure with bidirectionallongshort-termmemory （Bi-LSTM）networkisproposed.Inthismethod，thetrainingstabiltyofthedeeBi-LSTMnetwork isenhancedbyintroducing residualconnections，capturingthelong-termtemporaldependenciesinwindpowerdata.Additionally，theAdamoptiization algorithmisusedtooptimizethemodel’shyperparameters.Theempiricaltestingofthismethodwasconductedusingadataset fromawind powerenterpriseinQinghaiProvince.Theexperimentalresultsdemonstratethat，incomparison withsupportvector regresion （SVR），thestandardLSTMmodelandtheBi-LSTMmodel，thedeepresidualBi-LSTMmodelcan exhibitsignificant advantagesinwindpowerforecasting.Itsmeanabsoluteeror（MAE）predictionerrorisonly61.55，whichissignificantlylower than that of other three methods，and the coefficient of determination R2 valuereaches O.9377，indicating that this model has goodfittingandpredictionacuracy.ItfullydemonstratesthepotentialandvalueofthedeepresidualBi-LSTMmodelinthe field of wind power forecasting.

Keywords：windpowerprediction；deepresidual；Bi-LSTM；residual connection；Adamoptimizationalgorithm; hyperparameteroptimization

0 引言

模化潜力的可再生能源之一，在实现碳中和目标中扮演着关键角色[1。（剩余9196字）