Multi-granularity Rough Set Model And Its Application In The Correlation Analysis Of Haze And Meteorological Factors

There exists correlation mechanisms between hazes and meteorological factors.The paper is devoted to exploring and mining what meteorological factors are related to haze and how robust those correlations are.A multi-granularity rough set with variable decision attribute significance is proposed.Furthermore,based on this rough set model,a correlation analysis model for haze and meteorological factors is constructed.The main work and innovations of this paper are as follows:(1)Aiming at the problem of discretization of continuous attributes on large volume of meteorological observation data related to haze,a discretization method based on information entropy and inconsistency rate is proposed.First,regarding the discretization of continuous values,the paper gives the calculation formula of inconsistency rate,and optimizes the interval merger condition of an existing information entropy-based discretization method based on the inconsistency rate to eliminate subjective effects of setting inconsistency rate threshold manually and reduces the discretization points.Next,for the discretization method's efficiency problem under large scale of data,parallelize the improved discretization method with the map Reduce frame on the Hadoop platform.(2)The meteorological data set related to haze has the following features: on the one hand,the number of meteorological factors is large and the data set is incomplete.On the other hand,the meteorological factors lead to haze are more complex than those don't,so that the amount of noise data between the two situations are different.So in the practical haze analysis,the two situations need to be treated distinctively.That is to see inconsistently the significance of the decision attribute values.Moreover,the existing multi-granularity rough set model has the advantage of high generalization ability,and variable precision rough set model has the ability of dealing with noise data.In this paper,a multi-granularity rough set with variable decision attribute significance is proposed by combining the two rough set models above,and a correlation analysis model of haze and meteorological factors is constructed based on the proposed rough set model.Firstly,the paper defines thevariable decision attribute significance multi-granularity rough set.Secondly,a heuristic granularity reduction algorithm is proposed based on the rough set model.The heuristic granularity reduction algorithm allows to set different variable precision thresholds for different decision classes according to the decision attribute significance so as to eliminate the granularity not relevant to the target in decision system to reduce redundancy of granulation space.Finally,the paper put forward a rule abstracting algorithm based on confidence to extract classification rules that satisfies the setting confidences.The abstracted rules will analyze what meteorological factors relate to haze under setting confidence,and provide decision support for meteorological worker.(3)Test and analyze the results of the improved discretization method and the analysis model based on variable decision attribute significance multi-granularity rough set on real meteorological data.The test results show the improved discretization method and analysis model proposed in the paper are effective and the analysis model will provider a new method for the correlation analysis of haze and meteorological factors.