基金项目:河北省自然科学基金高端钢铁冶金联合研究基金专项项目(F2017209120);河北省自然基金面上项目(F2019209553)收稿日期:2021-09-15修回日期:2021-10-03第40卷第4期计算机仿真2023年4月文章编号:1006-9348(2023)04-0089-05基于VMD-RL-LSTM的短期风功率预测谷学静1,2,陈洪磊1,2,孙泽贤1,张怡1(1.华北理工大学电气工程学院,河北唐山063210;2.唐山市数字媒体工程技术研究中心,河北唐山063000)摘要:由于风电具有较高的随机性和较低的波动性,为了提高风电的非平稳性对于电力系统运行稳定性的影响,提出一种基于变分模态分解(Variationalmodedecomposition,VMD)、强化学习(Reinforcementlearning,RL)参数寻优和长短时记忆网络(Longshorttermmemory,LSTM)的短期风功率预测。为了降低数据的复杂度,首先通过变分模态分解将风功率原始数据分解为若干子模态。其次,通过强化学习对LSTM模型进行参数寻优,再对每个子模态建立LSTM模型预测。最终把各子模型预测的数据进行叠加,得到结果。对比分析结果显示,上述模型的预测精度较LSTM神经网络和EMD-LSTM预测模型均有不同程度的提高。关键词:风功率预测;变分模态分解;参数寻优;长短期记忆网络;深度学习中图分类号:TM914文献标识码:BShort-TermWindPowerPredictionBasedonVMD-RL-LSTMGUXue-jing1,2,CHENHong-lei1,2,SUNZexian1,ZHANGYi1(1.CollegeofElectricalEngineering,NorthChinaUniversityofScienceandTechnology,Tangshan,Hebei063210,China;2.TangshanDigitalMediaEngineeringTechnologyResearchCenter,Tangshan,Hebei063000,China)ABSTRACT:Sincewindpowerhashighrandomnessandlowvolatility,inordertoimprovetheinfluenceofwindpowernon-stationarityontheoperationstabilityofthepowersystem,anewmethodbasedonVMD,Reinforcementlearning(RL)andLongShorttermmemory(LSTM)networkisproposedforshort-termwindpowerprediction.Inordertoreducethecomplexityofthedata,theoriginalwindpowerdatawasdecomposedintoseveralsub-modesbyvariationalmodaldecomposition.Secondly,theparametersoftheLSTMmodelwereoptimizedbyreinforcementlearn-ing,andthenLSTMmodelpredictionwasestablishedforeachsub-mode.Finally,thedatapredictedbyeachsub-modelweresuperimposedtoobtaintheresults.ComparedwithLSTMneuralnetworkandEMD-LSTMpredictionmodel,thepredictionaccuracyofthismodelisimprovedtodifferentdegree...
