基于基数排序及属性约简的雷电预报因子提取方法  

Extraction method of lightning forecast factors based on radix sort and attribute reduction

作　　者：刘冲[1,2] 李向军[1] 沈华[3] 邱桃荣[1] 徐苏[1] 

机构地区：[1]南昌大学计算机科学与技术系,江西南昌330031 [2]江西警察学院,江西南昌330103 [3]南昌大学信息管理系,江西南昌330031

出　　处：《南昌大学学报（理科版）》2013年第4期344-349,共6页Journal of Nanchang University(Natural Science)

基　　金：国家自然科学基金资助项目(61070139);江西省青年科学家培养计划资助项目(20112BCB23004);江西省科技支撑计划重点资助项目(20111BBE50008;20112BBG70087);江西省自然科学基金资助项目(20114BAB201039;2009GZS0071)

摘　　要：雷电预报因子筛选是构建雷电预报模型的关键,也是提高雷电预报准确度的一个瓶颈问题。本文提出了一种基于粗糙集正域属性约简的雷电预报因子提取方法,该方法采用基数排序方法快速求解决策表的正域,引入可分辨度概念度量属性的重要性,提出了基于可分辨度的核属性与非核属性的提取方法,设计了相应的面向雷电预报因子提取的属性约简算法。真实气象数据集测试结果表明,该方法能有效提取最小集合的雷电预报因子,并较以往方法具有更好的性能。Screening of lightning forecast factors is an key step to establish lightning forecast model, at the same time a bottleneck for enhancing the accuracy of lightning forecast as well. We proposed a method to obtain lightning forecast factors through attribute the reduction based on rough set positive region. This method adopted efficient computation of positive region based on radix sort to calculate positive region of decision table. Meanwhile,we also introduced the concept of discernibility to measure the importance of attribute, and proposed the obtaining method for both core and non-core attribute based on recognizable resolution. The attribute reduction method for obtaining lightning forecast factors was designed. Tests on real and big meteorological data set showed that this method was effective. Such method not only obtainedmore effective lightning forecast factors, but also demonstrated better performance.

关 键 词：属性约简 属性重要度 基数排序 雷电预报因子 雷电预报模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]