| Granular computing,as a new method simulating human thinking and solving complex problems in the field of computational intelligence,is becoming popular.Information granulation is the premise and key of granular computing.It abstracts the information granules that can express the meaning of data according to a given granulation strategy in the problem-solving space,which has a wide range of application in the fields such as knowledge discovery,data mining and complex problem solving.In order to solve fuzzy and inseparable problems,Pedrycz divided information by granules and formed reasonable granules according to existent basis,and a two-phase information granulation framework of the principle of justifiable granularity was proposed.In the two-phase framework,an unsupervised learning method of clustering analysis was used in the first phase to form the prototype of the data set structure from original data.In a supervision mode,information granules were constructed in the second phase based on clusters which help to capture the core structure of the data set,so as to build a more comprehensive granularity structure.And the final granule prototype with better overall performance.Combining the unsupervised clustering algorithm with the supervised granulation,the two-phase information granulation framework of the principle of justifiable granularity can transform disorder data into representative information granules and has the advantages of high precision and simple.However,the principle still lacks a reasonable granulation function to describe the semantic structure of granules,the poor granule quality.Meanwhile,it is urgent to solve the problem of the crossed and unbalanced data granularity.Furthermore,the mixed data with various types is increasing every day.So,studying the two-phase information granulation of mixed data is of great theoretical value for enriching the theory and method of mixed data information granularity under the framework of the principle of justifiable granularity.Based on the theoretical research achievements,the application analysis of information granularity has an important practical meaning for the development of granular computing theory.Taking the two-phase information granulation of the principle of justifiable granularity as basic framework,this dissertation takes the following research as the mainline:the two-phase information granulation combined with the interval type-2 fuzzy rough K-mean and the mixed metric? the Information granulation of the mixed data based on the entropy fuzzy rough K-prototype under the two-phase framework?the application of two-phase information granulation in the production process of FCC adsorption desulfurization.The main research contents are as follows:(1)Two-phase information granulation combined with interval type-2 FRKM and mixed metricsAiming at the information granulation of the data with unbalanced distribution and the crossed cluster,a two-phase information granulation method combined interval type-2 FRKM and mixed metric is proposed.It could firstly get the basic information granules,based on the interval type-2 fuzzy rough K-means clustering algorithm,in which the weight index related to the cluster size and the spatial distribution information is introduced to calculate the membership weight in order to weaken the influence of the boundary area,and a severe deviation of cluster center to boundary area could be avoid.In the granulation stage,the common effects of density and interval are considered to describe the uniqueness of particle,which further enhances the ability of the function describe the structure and properties of the granules.Finally,a good granule boundary is obtained by optimizing the compound objective function,and the representative and standard information granules are obtained.The experimental analysis under artificial data sets and UCI standard data sets proved that the proposed algorithm has the advantages of self-adaptability and clustering accuracy for the imbalanced and crossed data sets.The information granular structure obtained is more compact and more representative.(2)The Information granulation of the mixed data based on the entropy fuzzy rough K-prototype under the two-phase frameworkFor the requirement of the mixed data granulation,the information granulation algorithm for the mixed data based on the entropy fuzzy rough K-prototype under the two-phase framework is proposed.The first stage of the algorithm considers the differences of the diversified data and their clustering contribution,the objective weights are assigned to different attributes by entropy weight method,and initial information granulates can be acquired by fuzzy rough K-prototype clustering.Because there is no effective way to measure the particle characteristics of the mixed data,in the second phase,taking the contribution of two kinds of attribute data into account,the uniqueness expression of the mixed data for describing the internal spatial structure of particles is redefined.Besides,the density of data distribution is also described reasonably.Comparing and analyzing the results of different experimental datasets,the algorithm further improved the stability of information granulation of the mixed data.(3)The application of two-phase information granulation in the production process of FCC adsorption desulfurizationAccording to the actual data from FCC gasoline production process,the two-phase information granulation algorithm combined with the interval type-2 fuzzy rough K-means and mixed metric is used to granulate the production data.Meanwhile,the granulation expressions of various condition parameters are extracted from different granularity levels.And then,the potential connections between reactor temperature,reaction system pressure,device flow,adsorbent level,etc.and the desulfurization rate of gasoline product,octane loss in the process are analyzed.Finally,the balance point of gasoline desulfurization rate and octane loss under the influence of different factors is explored,which could give a guidance to clean gasoline. |
PDF Full Text Request