摘要
强对流天气对社会公众的危害极大,强对流风暴内部的中气旋通常与灾害天气相关.为了预测强对流天气的发生时间,研究了中气旋参数与强对流风暴参数的相关关系.针对强对流风暴参数与中气旋参数之间的变化规律,引入了时间序列关联规则的挖掘方法.收集了天津塘沽雷达站的24个包含中气旋的强对流风暴案例,使用强对流风暴参数和中气旋参数的时间序列来描述案例集合,并使用典型变量分析算法将归一化的时间序列数据降维,考虑到序列间的变化趋势和局部关键点的影响,提取单调性特征和局部极值点特征对时间序列符号化表示,将同一时间段内的符号组成事务集,使用Eclat算法发现频繁项集;并且类比于学习器的性能度量,提出评估两类数据变化规律的正比率与反比率概念.实验结果显示,强对流风暴参数与中气旋参数之间存在显著的相关性,其第1对典型变量的相关系数大于0.5;在支持度为0.05~0.30下,Eclat算法的运行时间小于0.7 ms,内存占用小于121 KB,均优于Apriori算法和FP-growth算法;强对流风暴参数与中气旋参数之间的正比率大于0.7,直观地反映出强对流风暴参数与中气旋参数之间变化趋势的相似程度,且再次验证了频繁项集的正确性.该研究为利用中气旋数据预测强对流天气的变化趋势与发生时间提供了理论依据.
Severe convective weathers are extremely harmful to the public. The mesocyclone inside a severe convective storm is usually related to disaster weathers. To predict the occurrence time of severe convective weathers,the correlations between the parameters of mesocyclones and those of severe convective storms were studied. Moreover,a method for mining time series association rules was introduced in order to evaluate the variation law between severe convective storm parameters and mesocyclone parameters. Twenty-four severe convective storm cases involving mesocyclones in the Tianjin Tanggu radar station were collected. The case set was described by a time series of severe convective storm parameters and mesocyclone parameters,and a variable analysis algorithm was used to reduce the normalized time series data. Considering the trend of the variation between sequences and the influence of local key points,the monotonic features and local extremum features were extracted to symbolize the time series. The symbols in the same period were composed into a transaction set,and the frequent itemsets were found using the Eclat algorithm. Furthermore,a performance metrics method that reflects the similarity of the trends between the two sets of time series data according to positive ratio and inverse ratio is proposed to evaluate the variation law of two sets of data. Experimental results show that there is a significant correlation between the severe convective storm parameters and the mesocyclone parameters. The correlation coefficient of the first pair of typical variables is higher than 0.5.Under the support degree of 0.05—0.30,the running time of the Eclat algorithm is less than 0.7 ms,and the memory occupancy is less than 121 KB,which are better than those of both Apriori algorithm and FP-growth algorithm;the positive ratio between the severe convective storm parameters and the mesocyclone parameters is greater than 0.7,which reflects the similarity degree between the severe convective storm and the mesocyclone parameters and verifies the correctness of the frequent itemsets. This study provides a theoretical basis for predicting the trend and occurrence time of severe convective weathers using mesocyclone data.
作者
张军
闫丽飞
侯谨毅
Zhang Jun;Yan Lifei;Hou Jinyi(School of Electrical and Information Engineering,Tianjin University,Tianjin 300072,China)
出处
《天津大学学报（自然科学与工程技术版）》
EI
CSCD
北大核心
2019年第3期277-284,共8页
Journal of Tianjin University：Science and Technology
基金
天津市青年科学基金资助项目(2016120024002432)~~
关键词
强对流风暴
中气旋
关联规则挖掘
时间序列
severe convective storm
mesocyclone
association rule mining
time series
