【摘要】：迅猛發(fā)展的信息技術,特別是現(xiàn)代網(wǎng)絡、云計算等技術的廣泛應用,數(shù)據(jù)量呈爆炸式增長態(tài)勢,同時進入系統(tǒng)的信息來源越來越廣,相關層次越來越復雜.具備多源多模態(tài)等特征的復雜大數(shù)據(jù)已經(jīng)成為現(xiàn)代社會中數(shù)據(jù)資源和知識發(fā)現(xiàn)的主體.人們迫切需要去分析處理這些復雜數(shù)據(jù),從中找到有價值的信息.然而面對這些復雜數(shù)據(jù),傳統(tǒng)的數(shù)據(jù)處理技術遇到了極大挑戰(zhàn).因此,如何有效、快速地的處理這些復雜數(shù)據(jù),并提取出隱含其中的、潛在有用的知識,一直是智能信息處理領域的一個研究重點.作為知識獲取和問題求解的重要工具,粒計算方法是在問題求解過程中通過將復雜數(shù)據(jù)進行信息�；眯畔⒘４鏄颖咀鳛橛嬎愕幕締卧�,從多個角度、多個層次出發(fā)對現(xiàn)實問題進行描述、推理與求解,可大大提高計算效率并獲得問題更加合理、更加滿意的求解.本文將人類解決問題的多粒度思想引入到粗糙數(shù)據(jù)分析中,系統(tǒng)的開展了基于多粒度粗糙計算的方法研究.這極大地豐富了粗糙數(shù)據(jù)建模理論研究與應用范疇,有望為多源信息系統(tǒng)下的多粒度信息融合提供一個新途徑.獲得主要研究成果和創(chuàng)新如下：(1)發(fā)展了多源符號型數(shù)據(jù)和多源模糊型數(shù)據(jù)信息粒度的結構表示與融合模型.為了拓展多粒度粗糙集的建模能力和應用范圍,分別建立了多粒度覆蓋粗糙集模型和模糊多粒度決策粗糙集模型,深入探討了模型的性質(zhì),并揭示了這些模型之間的本質(zhì)差異,為多源粗糙數(shù)據(jù)分析中的模型選擇提供了理論基礎和可行的依據(jù).(2)從拓撲學理論的角度探討了多粒度粗糙集模型的相關理論.定義了多粒度拓撲粗糙空間并討論了該拓撲空間的重要性質(zhì),揭示了多粒度拓撲空間的內(nèi)部結構,通過定義粒度的重要性度量,并根據(jù)保持目標概念的內(nèi)部和閉包不變原則,提出了一個粒度空間的選擇算法,從而進一步完善了多粒度粗糙計算理論.(3)從不同的角度提出了多粒度近似空間的不確定性度量.借鑒廣義知識距離的思想,構造了多粒度近似空間的融合信息熵,融合粗糙熵和融合知識粒度,給出了多粒度拓撲粗糙空間的拓撲粒度和拓撲熵；提出了多粒度覆蓋粗糙集的粗糙度和粗糙熵,并討論這些度量的有關重要性質(zhì).這些結果將有助于理解多粒度粗糙計算理論作為不確定性問題求解理論的本質(zhì).(4)結合證據(jù)理論,提出了一類基于證據(jù)理論的多粒度融合算法.討論了樂觀／悲觀多粒度粗糙近似和經(jīng)典／模糊證據(jù)理論的信任函數(shù)之間的關系,給出了多粒度粗糙近似空間證據(jù)的基本概率指派獲取等問題.借鑒K-Modes聚類的思想完成多個粒結構的聚類,提出了一類基于證據(jù)理論的多粒度融合算法.在一定程度上解決了多源不確定信息的定量和定性融合問題,也增強了處理多源信息系統(tǒng)不確定問題求解的能力.(5)提出了三類整體決策性能評價指標.通過分析近似精度和近似質(zhì)量在度量決策性能的不足基礎上,利用最大最小合成方式提出了整體確定度,整體協(xié)調(diào)性,整體支持度.理論分析和實例驗證結果表明,提出的決策規(guī)則集的評價方法對未來的預測更合理可靠.通過以上系統(tǒng)研究,本文在多粒度粗糙計算理論與方法研究方面取得了系統(tǒng)的研究結果,發(fā)展了多粒度覆蓋信息粒度的結構表示和定性的融合方法；從多粒度拓撲理論和多粒度近似空間的不確定性這兩個側面完善了多粒度粗糙計算基本理論；建立了多粒度定性融合算子和定量的證據(jù)理論的信任函數(shù)之間的關系,發(fā)展了一類基于證據(jù)理論的多粒度融合算法,提出了整體融合決策性能評價方法,這些成果豐富和發(fā)展了多粒度粗糙計算理論和方法,為多粒度粒計算方法能夠更好的處理多源復雜數(shù)據(jù)提供了理論指導和技術支持.
[Abstract]:With the rapid development of information technology, especially the wide application of modern network and cloud computing technology, the amount of data is increasing and the information source of the system is getting wider and more complex. Complex large data with multi-source multi-modal characteristics has become the main body of data resources and knowledge discovery in modern society. There is an urgent need to analyse and process these complex data from which valuable information can be found. However, in the face of these complex data, conventional data processing techniques have encountered great challenges. Therefore, how to efficiently and quickly process these complex data and extract hidden and potentially useful knowledge has been a research focus in the field of intelligent information processing. As an important tool for knowledge acquisition and problem solving, the calculation method is to describe, infer and solve the real problem from multiple perspectives and multiple levels by replacing the complex data with information particles instead of the sample as the basic unit of calculation in the problem solving process. the calculation efficiency can be greatly improved and the problem is more reasonable and the solution is more satisfied. In this paper, the multi-granularity idea of human problem solving is introduced into rough data analysis, and a method based on multi-granularity rough calculation is carried out. This greatly enriches the research and application of rough data modeling theory, and is expected to provide a new way for multi-granularity information fusion in multi-source information system. The main research results and innovation are as follows: (1) The structure representation and fusion model of multi-source symbol data and multi-source fuzzy data information granularity are developed. In order to expand the modeling capability and application range of multi-granularity rough set, a rough set model and fuzzy multi-granularity decision rough set model with multiple granularity are established, the nature of the model is discussed in detail, and the essential difference between these models is revealed. It provides theoretical basis and feasible basis for model selection in multi-source rough data analysis. (2) The correlation theory of multi-granularity rough set model is discussed from the point of view of theory. A multi-granularity topological rough space is defined and the important properties of the topological space are discussed, the internal structure of the multi-granularity topological space is revealed, the importance measure of granularity is defined, and according to the principle of keeping the internal and closed packets of the target concept unchanged, A selection algorithm of granularity space is presented to further improve the theory of multi-granularity coarse computation. (3) The uncertainty measure of multi-granularity approximation space is presented from different angles. Based on the idea of the generalized knowledge distance, the fusion information entropy, the fusion rough entropy and the fusion knowledge granularity of the multi-granularity approximation space are constructed, the topological granularity and the topological entropy of the multi-granularity topological rough space are given, and the roughness and the rough entropy of the multi-granularity coverage rough set are proposed. Some important properties of these measures are discussed and discussed. These results will help to understand the nature of multi-particle size rough calculation theory as the solution of uncertainty problem. (4) Based on the theory of evidence, a kind of multi-granularity fusion algorithm based on evidence theory is proposed. The relationship between optimistic/ pessimistic multi-granularity coarse approximation and trust function of classical/ fuzzy evidence theory is discussed, and the basic probability assignment acquisition of multi-granularity coarse approximation space evidence is given. Based on the idea of K-Modes clustering, a class of multi-granularity fusion algorithm based on evidence theory is proposed. The quantitative and qualitative fusion problem of multi-source uncertain information is solved to a certain extent, and the ability to deal with uncertain problem of multi-source information system is also enhanced. (5) Three types of overall decision-making performance evaluation indexes are proposed. Based on the analysis of the deficiency of the approximate precision and the approximate quality in the metric decision-making performance, the whole determination degree, the overall coordination and the overall support degree are put forward by means of the maximum minimum synthesis method. The results of theoretical analysis and case verification show that the proposed method of decision rule set is more reasonable and reliable for future prediction. Through the above system research, this paper has obtained systematic research results in multi-particle size rough calculation theory and method research, and developed the structure representation and qualitative fusion method of multi-granularity coverage information granularity. The basic theory of multi-granularity coarse computation is improved from the two sides of multi-granularity topological theory and the uncertainty of multi-granularity approximation space, and the relation between the multi-granularity qualitative fusion operator and the trust function of the quantitative evidence theory is established. A kind of multi-granularity fusion algorithm based on evidence theory is developed, and a method for evaluating the whole fusion decision-making performance is proposed, which enriches and develops the theory and method of multi-granularity coarse computation. The method can better deal with multi-source complex data and provide theoretical guidance and technical support for multi-granularity calculation method.
【學位授予單位】：山西大學
【學位級別】：博士
【學位授予年份】：2016
【分類號】：TP18

【參考文獻】

相關期刊論文 前5條

1 梁吉業(yè);錢宇華;李德玉;胡清華;;大數(shù)據(jù)挖掘的粒計算理論與方法[J];中國科學:信息科學;2015年11期

2 錢學森;;一個科學新領域——開放的復雜巨系統(tǒng)及其方法論[J];上海理工大學學報;2011年06期

3 張楠;苗奪謙;岳曉冬;;區(qū)間值信息系統(tǒng)的知識約簡[J];計算機研究與發(fā)展;2010年08期

4 陳德剛,張文修;粗糙集和拓撲空間[J];西安交通大學學報;2001年12期

5 劉清,黃兆華,劉少輝,姚力文;帶Rough算子的決策規(guī)則及數(shù)據(jù)挖掘中的軟計算[J];計算機研究與發(fā)展;1999年07期

，

本文編號：2256021