CS算法优化VMD-BiLSTM-AM的光伏功率预测①俞敏,王晓霞(华北电力大学计算机系,保定071003)通信作者:俞敏,E-mail:whale1028@126.com摘要:针对光伏发电功率的波动性与随机性对调度部门的负荷预测以及电网安全运行带来的严峻挑战,提出了一种基于变分模态分解(VMD)和布谷鸟搜索(CS)算法优化的双向长短期记忆网络(BiLSTM)光伏发电功率预测方法.首先使用VMD将光伏功率序列分解成不同频率的子模态,通过皮尔逊相关性分析确定影响各模态的关键气象因子.其次分别构建注意力机制(AM)和BiLSTM混合的光伏发电功率预测模型,利用CS算法获取网络最优的权重和阈值.最后,将不同模态的预测结果相叠加,得到最终的预测结果.通过对亚利桑那州地区光伏电站输出功率进行预测,验证了所提模型的有效性.关键词:双向长短期记忆网络;变分模态分解;布谷鸟搜索;注意力机制;光伏功率预测引用格式:俞敏,王晓霞.CS算法优化VMD-BiLSTM-AM的光伏功率预测.计算机系统应用,2023,32(2):347–355.http://www.c-s-a.org.cn/1003-3254/8921.htmlPhotovoltaicPowerPredictionBasedonVMD-BiLSTM-AMOptimizedbyCSAlgorithmYUMin,WANGXiao-Xia(DepartmentofComputer,NorthChinaElectricPowerUniversity,Baoding071003,China)Abstract:Fortheseverechallengesbroughtbythefluctuationandrandomnessofphotovoltaicpowergenerationtotheloadpredictionofthedispatchingdepartmentandthesafeoperationofthepowergrid,thisstudyproposesaphotovoltaicpowerpredictionmethodofbidirectionallongshort-termmemory(BiLSTM)optimizedbyvariationalmodaldecomposition(VMD)andcuckoosearch(CS)algorithm.Firstly,VMDisemployedtodecomposethephotovoltaicpowersequenceintosub-modeswithdifferentfrequencies,andPearsoncorrelationanalysisisadoptedtodeterminethekeymeteorologicalfactorsaffectingeachmode.Secondly,thehybridphotovoltaicpowerpredictionmodelsofattentionmechanism(AM)andBiLSTMareconstructed,andtheCSalgorithmisutilizedtoobtaintheoptimalweightandthresholdofthenetwork.Finally,thepredictionresultsofdifferentmodesaresuperimposedtoobtainthefinalpredictionresults.TheeffectivenessoftheproposedmodelisverifiedbypredictingtheoutputpowerofphotovoltaicpowerstationsinArizona.Keywords:bidirectionallongshort-termmemory(BiLSTM);variationalmodaldecomposition(VMD);cuc...
