李 駿,白莉君
蘭州理工大學(xué) 理學(xué)院,蘭州 730050
模糊推理SIS算法的統(tǒng)一形式及其還原性*
李駿+,白莉君
蘭州理工大學(xué) 理學(xué)院,蘭州 730050
模糊推理是模糊控制的核心問題,還原性則是評(píng)價(jià)模糊推理算法好壞的重要標(biāo)準(zhǔn)之一。在正則蘊(yùn)涵算子的統(tǒng)一框架下,給出了基于模糊推理SIS(subsethood infer subsethood)算法的模糊取式(fuzzy modus ponens,F(xiàn)MP)問題解的統(tǒng)一表達(dá)式;基于SIS算法為模糊拒取式(fuzzy modus tollens,F(xiàn)MT)問題提出了一種改進(jìn)的求解原則,并給出了FMT問題解的統(tǒng)一形式;證明了SIS FMP算法和SIS FMT算法均滿足無條件還原性,討論了FMP問題及FMT問題基于SIS算法的λ-水平解。該算法將為模糊控制領(lǐng)域提供更多可供選擇的模糊推理方法。
模糊控制;模糊推理;正則蘊(yùn)涵算子;SIS算法
模糊推理是模糊控制的核心問題,模糊推理最基本的兩種推理形式如下:
模糊取式(fuzzy modus ponens,F(xiàn)MP):
模糊拒取式(fuzzy modus tollens,F(xiàn)MT):
這里A,A*∈F(X),B,B*∈F(Y),F(xiàn)(X)、F(Y)分別表示非空論域X、Y上的全體模糊集。
1973年,美國控制論專家Zadeh提出了求解FMP問題的合成推理方法(compositional rule of inference,CRI)[1]。隨后模糊界圍繞CRI算法展開了比較深入的研究,在理論上和應(yīng)用上取得了豐富的成果[2-4]。然而盡管CRI算法在計(jì)算上是簡(jiǎn)便的,但是它卻缺乏嚴(yán)格的邏輯依據(jù)[5-6]。為了給模糊推理奠定嚴(yán)格的邏輯基礎(chǔ),王國俊教授提出了模糊推理的全蘊(yùn)涵三I算法[5]。三I算法相較于CRI算法具有更好的邏輯背景,因此吸引了不少學(xué)者對(duì)其進(jìn)行研究,并得到了大量研究成果。比如:文獻(xiàn)[7-8]分別給出了基于正則蘊(yùn)涵算子和剩余型蘊(yùn)涵算子的三I算法的統(tǒng)一形式;文獻(xiàn)[9]研究了基于反向支持度的三I算法;文獻(xiàn)[10-14]基于不同度量研究了三I算法的魯棒性。
另一方面,除了魯棒性,算法的還原性也是評(píng)價(jià)模糊推理方法好壞的重要標(biāo)準(zhǔn)之一,盡管三I算法在還原性方面具有比CRI算法更好的性質(zhì),但它并不滿足無條件還原性[5-8]。為此,文獻(xiàn)[15]提出了一種新的模糊推理算法——SIS(subsethood infer subsethood)算法。
SIS FMP原則[15]FMP問題(1)的SIS解B*是使得下式:
取最大值的F(Y)中的最大模糊集。
SIS FMT原則[15]FMT問題(2)的SIS解A*是使得下式:
取最大值的F(X)中的最大模糊集。這里B′、A′分別是模糊集B、A的補(bǔ)集。
文獻(xiàn)[15]在蘊(yùn)涵算子分別取R0算子和Lukasiewicz算子的情形下證明了滿足SIS FMP和SIS FMT原則的解均存在,并給出了求解算法,同時(shí)證明了對(duì)上述兩種蘊(yùn)涵算子給出的求解算法均滿足無條件還原性。但文獻(xiàn)[15]僅考慮了蘊(yùn)涵算子為R0和Lukasiewicz算子的情形,若蘊(yùn)涵算子取別的算子,特別是在邏輯推理中有重要應(yīng)用的正則蘊(yùn)涵算子類,情況會(huì)如何呢?另外,文獻(xiàn)[15]針對(duì)FMT問題(2)的SIS求解原則是在把大前提A→B等價(jià)地轉(zhuǎn)化為B′→A′時(shí)提出的,但這只適用于滿足換質(zhì)位對(duì)稱性的蘊(yùn)涵算子(即滿足等式 A(x)→B(y)=B′(y)→A′(x)的蘊(yùn)涵算子),比如R0算子和Lukasiewicz算子,對(duì)更一般的蘊(yùn)涵算子(特別是不具有換質(zhì)位對(duì)稱性的算子),文獻(xiàn)[15]中給出的SIS FMT求解原則和求解算法則不再適用。本文在正則蘊(yùn)涵算子的統(tǒng)一框架下,給出了基于SIS算法的FMP問題解的統(tǒng)一表達(dá)式,提出了一種改進(jìn)的SIS FMT求解原則,并給出了基于該原則的FMT問題解的統(tǒng)一算法,進(jìn)而證明了SIS FMP算法和SIS FMT算法均滿足無條件還原性,最后討論了FMP問題及FMT問題基于SIS算法的λ-水平解。
定義1[6,16]設(shè)?:[0,1]2→[0,1]是二元函數(shù),a,b,c∈[0,1],I為指標(biāo)集,若
(1)a?b=b?a
(2)(a?b)?c=a?(b?c)
(3)a?1=a
(4)若b≤c,則a?b≤a?c
則稱?為[0,1]上的三角模,簡(jiǎn)稱t-模,如果?還滿足
則稱?是左連續(xù)的三角模。
定義2[16]設(shè)R:[0,1]2→[0,1]是二元函數(shù),?是[0,1]上的三角模,若a?b≤c當(dāng)且僅當(dāng)a≤R(b,c),則稱R是與?相伴隨的蘊(yùn)涵算子,R(b,c)也常記為b→c。當(dāng)?是左連續(xù)的三角模時(shí),則稱與其相伴隨的蘊(yùn)涵算子為正則蘊(yùn)涵算子。
下面給出幾種常用的左連續(xù)的三角模和它們所對(duì)應(yīng)的正則蘊(yùn)涵算子。
注1當(dāng)蘊(yùn)涵算子為正則蘊(yùn)涵算子時(shí),由引理1(1)知式(3)的最大取值為1,從而FMP問題(1)的SIS解B*是使得下式成立的F(Y)中的最大模糊集:
本文在正則蘊(yùn)涵算子的統(tǒng)一框架下給出了SIS FMP以及SIS FMT求解算法的統(tǒng)一形式,證明了SIS FMP算法和SIS FMT算法都具有無條件的還原性,并給出了基于正則蘊(yùn)涵算子的SIS算法的λ-水平解,為模糊控制領(lǐng)域中模糊推理方法的選擇提供了更多的可能性。關(guān)于SIS FMP算法和SIS FMT算法的魯棒性,將另文討論。
[1]Zadeh L A.Outline of a new approach to the analysis of complex systems and decision processes[J].IEEE Transactions on Systems,Man and Cybernetics,1973,3(1):28-44.
[2]Cai Kaiyuan,Zhang Lei.Fuzzy reasoning as a control problem[J].IEEE Transactions on Fuzzy Systems,2006,14(6): 709-715.
[3]Ying Mingsheng.Perturbation of fuzzy reasoning[J].IEEE Transactions on Fuzzy Systems,1999,7(5):625-629.
[4]Perfilieva I,Lehmke S.Correct models of fuzzy IF-THEN rules are continuous[J].Fuzzy Sets and Systems,2006,157 (24):3188-3197.
[5]Wang Guojun.The full implication triple I method for fuzzy reasoning[J].Science in China:Series E,1999,29(1):43-53.
[6]Pei Daowu.Theory and applications of fuzzy logic based on triangular norms[M].Beijing:Science Press,2013.
[7]Wang Guojun,Fu Li.Unified forms of triple I method[J]. Computers and Mathematics with Applications,2005,49(5/6): 923-932.
[8]Pei Daowu.The unified algorithms of triple I methods for fuzzy reasoning[J].Information Sciences,2008,178(2): 520-530.
[9]Song Shiji,Feng Chunbo,Lee E S.Triple I method of fuzzy reasoning[J].Computers and Mathematics with Applications,2002,44(12):1567-1579.
[10]Wang Guojun,Duan Jingyao.On robustness of the full implication triple I inference method with respect to finer measurements[J].International Journal of Approximate Reasoning, 2014,55(3):787-796.
[11]Li Yongming,Li Dechao,Pedrycz W,et al.An approach to measure the robustness of fuzzy reasoning[J].International Journal of Intelligent Systems,2005,20(4):393-413.
[12]Li Yingfang,Qin Keyun,He Xingxing.Robustness of fuzzy connectives and fuzzy reasoning[J].Fuzzy Sets and Systems,2013,225(3):93-105.
[13]Dai Songsong,Pei Daowu,Wang Sanmin.Perturbation of fuzzy sets and fuzzy reasoning based on normalized Minkowski distances[J].Fuzzy Sets and systems,2012,189 (1):63-73.
[14]Dai Songsong,Pei Daowu.Robustness analysis of full implication inference method[J].International Journal of Approximate Reasoning,2013,54(5):653-666.
[15]Zou Xiangfu,Pei Daowu.SIS algorithms for fuzzy reasoning [J].Fuzzy Systems and Mathematics,2010,24(6):1-7.
[16]Wang Guojun,Zhou Hongjun.Introduction to Mathematical logic and Resolution Principle[M].Beijing:Science Press/ Oxford,UK:Alpha Science International Limited,2009.
附中文參考文獻(xiàn):
[5]王國俊.模糊推理的全蘊(yùn)涵三I算法[J].中國科學(xué):E輯, 1999,29(1):43-53.
[6]裴道武.基于三角模的模糊邏輯理論及其應(yīng)用[M].北京:科學(xué)出版社,2013.
[15]鄒祥福,裴道武.模糊推理的SIS算法[J].模糊系統(tǒng)與數(shù)學(xué),2010,24(6):1-7.
LI Jun was born in 1972.He received the Ph.D.degree in uncertainty reasoning from Shaanxi Normal University in 2008.Now he is an associate professor at Lanzhou University of Technology.His research interests include computational intelligence and uncertainty reasoning,etc.
李駿(1972—),男,甘肅白銀人,2008年于陜西師范大學(xué)獲得博士學(xué)位,現(xiàn)為蘭州理工大學(xué)理學(xué)院副教授,主要研究領(lǐng)域?yàn)橛?jì)算智能,不確定性推理等。
BAI Lijun was born in 1989.She is an M.S.candidate at Lanzhou University of Technology.Her research interests include computational intelligence and uncertainty reasoning,etc.
白莉君(1989—),女,陜西渭南人,蘭州理工大學(xué)理學(xué)院碩士研究生,主要研究領(lǐng)域?yàn)橛?jì)算智能,不確定性推理等。
Unified Form and Reversibility of SISAlgorithms for Fuzzy Reasoning?
LI Jun+,BAI Lijun
School of Science,Lanzhou University of Technology,Lanzhou 730050,China
E-mail:lj99120@163.com
Fuzzy reasoning is the core of fuzzy control,the reversibility is one of most important evaluation standards for fuzzy reasoning algorithms.Under the framework of regular implication operators,this paper firstly gives the unified expression for solving FMP(fuzzy modus ponens)problems based on SIS(subsethood infer subsethood) method.Secondly,this paper proposes an improved SIS FMT(fuzzy modus tollens)algorithm and the unified form of its solution.In the end,this paper proves that the SIS FMP algorithm and SIS FMT algorithm are both unconditionally reversible,and studies theλ-solution of SIS algorithm for FMP and FMT.This algorithm will provide more alternative methods of fuzzy reasoning for the area of fuzzy control.
fuzzy control;fuzzy reasoning;regular implication operators;SIS algorithm
2015-07,Accepted 2015-09.
10.3778/j.issn.1673-9418.1507039
A
TP181
*The National Natural Science Foundation of China under Grant No.11261032(國家自然科學(xué)基金).
CNKI網(wǎng)絡(luò)優(yōu)先出版:2015-09-28,http://www.cnki.net/kcms/detail/11.5602.TP.20150928.1652.006.html
LI Jun,BAI Lijun.Unified form and reversibility of SIS algorithms for fuzzy reasoning.Journal of Frontiers of Computer Science and Technology,2016,10(10):1469-1474.