基金项目:山西省软科学研究计划项目(2019041023-5)收稿日期:2021-11-19修回日期:2022-01-27第40卷第4期计算机仿真2023年4月文章编号:1006-9348(2023)04-0499-05基于雷达回波光流场的天气预报数据误差识别郭艳萍,高云,吕丙东,彭炜(山西大同大学计算机与网络工程学院,山西大同037009)摘要:天气预报数据具有相似度高、关联性强的特点,导致挖掘和分类的误差识别难度较大。为此提出新的天气预报数据误差识别方法。将雷达回波图划分为多个小空间区域,求解得到各空间的回波光流场。通过整合得到多普勒天气雷达测得的整个雷达回波光流场。利用解得的实际与预报回波相关系数,定量描述数据误差相关性,结合平均绝对误差与探测概率指标,定性补充预报数据误差分析结果。将预报数据误差的识别转换为数据挖掘与分类问题,以监督决策树理念为基础,建立误差识别模型。仿真以降水强度、位置以及形态展开误差识别检验。通过对错误、缺漏识别次数与识别时间等指标的验证,证明了所设计模型具有较高的误差识别精度,且短时预报的误差识别实时性较强。关键词:雷达回波外推技术;预报数据误差;误差识别;光流场;有监督决策树中图分类号:TP399文献标识码:BErrorIdentificationofWeatherForecastDataBasedonRadarEchoOpticalFlowFieldGUOYan-ping,GAOYun,LVBing-dong,PENGWei(SchoolofComputerandNetworkEngineering,ShanxiDatongUniversity,DatongShanxi037009,China)ABSTRACT:Weatherforecastdatahasthecharacteristicsofhighsimilarityandstrongcorrelation.However,itisdifficulttoidentifyminingerrorsandclassificationerrors.Therefore,thispaperputsforwardanewmethodtoidentifyweatherforecastdataerrors.Theradarechoimagewasdividedintoseveralsmallspaceregionsatfirst.Andthenechoopticalflowfieldofeachspacewassolved.Afterintegration,thewholeopticalflowfieldmeasuredbyDopplerweath-erradarwasobtained.Moreover,thecorrelationcoefficientsbetweenactualandpredictedechoeswereusedtoquanti-tativelydescribethecorrelationbetweendataerrors.Combinedwiththemeanabsoluteerroranddetectionprobabilityindex,theanalysisresultsofpredictiondataerrorweresupplementedqualitatively.Thentheidentificationofpredic-tiondataerrorwastransformedintotheproblemofdataminingandclassification.Basedontheconceptofsuperviseddecisiontree,amodeloferroridentifi...
