Radix Sort and Attribute Reduction-based Lightning Forecast Factor Extraction

摘要

Screening of lightning forecast factors is an important step to establish lightning forecast model and a bottleneck for enhancing the accuracy of lightning forecast. This paper proposes a method to obtain lightning forecast factors through attribute reduction based on rough set positive region. This method adopts efficient computation of positive region based on radix sort to compute positive region of decision table, introduces the concept of discernibility to measure the importance of attribute, proposes the obtaining method of core attribute and non-core attribute based on recognizable resolution and designs attribute reduction method for obtaining lightning forecast factors. Tests on real and big meteorological data set show that this method is effective. Such method not only obtains fewer and effective lightning forecast factors, but also demonstrates better performance.