2023年第37卷第4期测试技术学报Vol.37No.42023(总第160期)JOURNALOFTESTANDMEASUREMENTTECHNOLOGY(SumNo.160)文章编号:1671-7449(2023)04-0284-05①超短期风力发电量预测技术及其比较分析张利平,赵俊梅,刘丹,陈昌鑫(中北大学电气与控制工程学院,山西太原030051)摘要:风力发电作为清洁绿色的新能源,是实现“双碳”目标的主力军之一,但是其对新能源消纳系统提出了新要求,故对风力发电量的科学分析和精确预测研究具有现实意义。首先,对风电多维历史数据属性、特点和离群值、噪声平滑等进行分析与预处理,再通过2种回归树集成和4种回归神经网络及其超参数优化算法对不同机组数据进行回归分析,超参数优化运行时间代价较高。回归拟合效果通过5个评价指标进行对比与分析,经过大量仿真实验,证明了三层神经网络的回归模型拟合和预测效果均较好。关键词:回归树集成;回归神经网络;超参数优化;预测技术中图分类号:TM715文献标识码:Adoi:10.3969/j.issn.1671-7449.2023.04.002UltraShortTermWindPowerGenerationForecastingTechnologyandItsComparativeAnalysisZHANGLiping,ZHAOJunmei,LIUDan,CHENChangxin(SchoolofElectricalandControlEngineering,NorthUniversityofChina,Taiyuan030051,China)Abstract:Ascleanandgreennewenergy,windpowergenerationisoneofthemainforcestoachievethe“doublecarbon”goal.However,itputsforwardnewrequirementsforthenewenergyconsumptionsystem,soithaspracticalsignificanceforthescientificanalysisandaccuratepredictionofwindpowergeneration.Firstly,theattributes,characteristics,outliersandnoisesmoothingofmulti-dimensionalhistoricaldataofwindpowerareanalyzedandpreprocessed,andthenthedataofdifferentunitsarere-gressedandanalyzedthroughtworegressiontreeensemble,fourregressionneuralnetworksandtheirsuperparameteroptimizationalgorithms.Thecostofsuperparameteroptimizationishigh.Theregres-sionfittingeffectiscomparedandanalyzedthroughfiveevaluationindexes.Alargenumberofsimula-tionexperimentsprovedthattheregressionmodelfittingandpredictionverificationeffectofthethree-layerneuralnetworkaregood.Keywords:regressiontreeensemble;regressionneuralnetwork;hyperparametricoptimization;forcast-ingtechnology0引言国家深入贯彻新发展理念,亟待解决资源高效利用问题,实现绿色低碳...
