基于动态集成LSSVR的超短期风电功率预测.doc

《基于动态集成LSSVR的超短期风电功率预测.doc》由会员上传分享，免费在线阅读，更多相关内容在应用文档-天天文库。

1、基于动态集成LSSVR的超短期风电功率预测　　摘要：�对最小二乘支持向量回归建模风电功率时变特性的局限性，提出了一种基于动态集成LSSVR的超短期风电功率预测模型.首先利用风电场监测控制与数据采集与数值天气预报系统的历史数据建立离线单体LSSVR模型库，然后根据预测时段与训练时段NWP序列的相似度从单体LSSVR模型库中动态选择候选集成成员，再后综合考虑正确性与多样性确定集成成员.最后由预测时段与训练时段NWP序列间的相似度分配集成LSSVR成员的权重.通过对湖南省某风电场输出功率进行预测，验证了动态集成LSSVR预测模型的有效性，与持续法、自回归求和移动平均法、单体LSSV

2、R模型、常权重LSSVR组合模型及BPNN动态集成模型相比，动态集成LSSVR模型具有更高的精度，在天气非平稳变化阶段更加明显.　　关键词：超短期风电功率预测；最小二乘支持向量回归；动态集成；动态时间弯曲距离；数值天气预报　　中图分类号：TM614文献标志码：A　　Ultra-short-termWindPowerPredictionBasedonDynamicalEnsembleLeast　　SquareSupportVectorRegressionLIURongsheng1，PENGMinfang1，ZHANGHaiyan1，WANXun2，SHENMeie3　　　　Ab

3、stract：Forthelimitationofleastsquaresupportvectorregressioninmodelingthetimevaryingfeatureofwindpower，anultra-short-termwindpowerpredictionmodelbasedondynamicalensembleLSSVRwasproposed.Firstly，theoff-lineLSSVRmodellibrarywascreatedbymakinguseofthehistoricaldatawhichwereobtainedfromNumerical

4、WeatherPredictionandsupervisorycontrolanddataacquisitionsystemofwindfarm.Then，thecandidatemembersofensembleLSSVRwereselectedfromoff-lineLSSVRmodellibrarydynamicallyaccordingtothesimilaritybetweentheNWPofforecastingperiodandtheNWPoftrainingperiod.Theensemblemembersweredecidedbyconsideringthe

5、accuracyanddiversity.Finally，theweightsofensembleLSSVRmemberswereassignedaccordingtothesimilaritybetweentheNWPoftrainingandNWPofpredictionperiod.ThevalidityofthedynamicalensembleLSSVRbasedpredictorwasverifiedbypredictingthewindpowerofawindfarminHunanProvince.Comparedwithpersistencemethod，au

6、toregressiveintegratedmovingaverage，LSSVR，constantweightensembleLSSVR，andensembleartificialneuralnetworks，thedynamicalensembleLSSVRismoreaccurate，especiallywhentheweatherchangesseverely.　　Keywords：ultra-short-termwindpowerprediction；leastsquaresupportvectorregression；dynamicalEnsemble；dynam

7、icaltimewarp；numericalweatherprediction