經濟部中夹標準局R工消費合作社印製 2194ϋ0 Α6 _BG_ 五、發明説明（） 本發明偽為一種可缩短處理時間之模糊推論法則記億 區分配方法，主要偽将模期推論法則記億區之各位址區分 為輸出入辨識元、输出入變數、前導/後導筠靨函數旗標 ,並令儲存於記億區之各法則位址匾間之起始位址存放與 下一法則之起始位址差資料（以供後缠跳躍至下一法則之 用），位址區間之中段位置存放輪入受數及前導结屬函數 、而位址區間末段位置存放輸出變數及後導嫌羼函數.據 此等法刖分配方法，於讀取起始位址所存放之位址差資料 時，可逕將現在位址與該位址差資料相加而儲存於下一法 則之位址暂存器内，當後缅依次讀取輪入受數及前導歸屬 函數旗標並做.相關計算取出歸1函數程度值時，若該程度 值為零，即表示為無法披输入變數激發，故直接跳躍至下 一法則位址暫存器之值所表示之下一法則起始位址予以進 行處理，故前述法則區段之配置方式即可獲致跳過不激發 之法則，達到提昇處理速度之效益者。 按現今模瑚（FUZZY)理論E用的階層已進入漸為普及之 階段，如工業上機械的蓮用以至於家電産品均可見蓮用模 湖理論之物品，而隨箸精密控制及複雜的控制上，邸使用 大置的模糊推論法則及相願蓮算，從而對於蓮算速度上的 快慢卽為其效能表現的主要因素，故為此提昇蓮算速度即 為業界主要的工作目標，而本發明主要即為達成該項提昇 運算速度之理念下，邸鑒於傳统模糊蓮算方式上均為直接 蓮算，對於一些可判定為無效或導致輸出為無效之蓮算亦 本纸張尺及適用中园S家標準（CNS) V 4现格（210 X 297公货） 81.9.20.000 ---------L---------------装------，玎------'^ <請先閱讀背面之注悉事項存塡寫本頁) 五、發明説明（） A6 B6 娌濟部中夬標準扃員工消費合作社印製 推行至間 樓之輸為址偽受歸始存則導值暫，跳 0 撕進 《時 之址該標位址入導起暫法前度址算可 待 棋算跳算 間位 ，旗慊位输後之址 一及程位蓮邸 ， 過運接蓮 時慊成數記始放及含位下數該則述 ， 法 透於直之 理記組函則起存數内之至變斷法前法 方 為可可則 處則所靨法之置變令則躍入判一覆方。之 要到邸法 短法標筠各段位出邸法跳輸若下重配間明 主達 ，效 缩各旗及 ，上段輸 -一接之 ，述而分時發 人 ，時無 可将數數造其中故時下直址值前址之算本 明式態該 種偽函變構 ,,存址入以位度照位容蓮解 發方狀對 一要屬續區份料置位存值他程依始内期瞭 本置之短 。供主歸後僂部資位始而之其數故起段模步 ，配發縮者提 ，及識記豳差段起加器對函 ，則匾値一 此及激而果於法數辨成三址後則相存次屬效法慊整進 因變不 -效在方爱為構下位 ，法值暫依歸無一記短能 , 改或箄之的配 、做以 、始標各址該再得為下則缩員 間的效蓮度目分元元據中起旗取位用後取則至法到委 時容無行速要 S 游位， 、之數讀在取其算法躍等達査 算内為進算主慊辨一態.±則函當現缠 ，運此跳此 ，審 蓮區入置蓮之記入以型 +法靥 -與後 } 標示逕明則貴 的僚輸位昇明則出元出割一箱標料供上旗表而發法 當記斷則提發法输議输分下導旗資以址數即值本之使 相則判法致本論以辨或更與前數差丨位函 之 ，效為 費法若一獲 推容入入段放及函址内始靥零器此無 耗論中下, 期内出输區存數屬位器起歸為存據過 ------„----.---------------裝------訂------線— (請先《讀背面之注悉事項再填寫本頁) 本纸張尺Λ適用中國國家標準（CNS) r 说格（210 X 297公釐） 81.9.20.000 _____ 烴濟部中央標準局g工消¢·合作社印s A6 _BG _ 五、發明説明（） 及其他目的，玆 附以圖式詳細說明如后： (一）.圖式部份： 第一圖：偽本發明之辨識元及输出入變數定義表 第二圖：偽本發明之記憶區配置實施例。 藺於本發明之所以將無效或不受輸入變數激發之法則 據以跳過之理由説明如下： 現今各式模湖運算過程中，無論是採用最大/最小合 成法或最大.乘積合成法，若在運算求得之多數輪入歸屬 函数程度值中有任一為零，則其在最小或乘積運算後，即 令對應之輸出歸屬函數程度值必然為零（A Ο邏輯運箄）， 如此，若其後缠之其他輪入歸屬函數程度值卽使有爱化， 亦無法造成輸出結果之差異，故而傳統樓瑚運算均為直接 蓮算至結束，此舉即造成運算時間無謂浪費及缺乏效益， 故本發明即偵測各法則之其一輸入歸靥函數程度值為零時 ,邸不再行蓮算而直接跳過至次一法則予以重覆前述步思 ,如此即可免除蓮算時間之浪費現象，並據此逹到提高蓮 算效率者。 而達成前述跳躍（J U Μ P )至下一法則，即需於前一法則 進行蓮算時，需一併記錄下一法則之起始位置，以待前述 偵测若為無效（不激發狀態）時，邸可依據該下一法則之 起始位置而逕跳至下一法則位置上予以進行蓮算，故本發 明即提供一種不同於傳统模糊理論法則記億區分配方法， 以達成前述快速跳躍之作用及縮短運算推論時間者，以下 本纸張又度適用中國园家標準（CNS)甲4現格（210 X 297公釐） -----:---.-----------------装------ΤΓ------象 (請先閲熗背面之;±念事項再塡寫本頁) 81.9.20,000 319400 A6 _B6 __ 五、發明説明（） (请先閲洁背面之注念事項再壜寫本頁) 邸以一具有八輸入（各輸入具有十六餹前導结靥函數）及 六値输出（各输出具有十六値前導歸屬函數）之模期推論 法則記億匾為例說明之，其中，各記慊位址上即含有一位 元之输出入辨識元（ίο)、三位元之輸出入變數（οι V)及四位元之歸屬函數旗檫（MEM)所組成（如第二 圔），而該輸出入辨識元（1〇)及输出入受數（〇 IV )之组合關傜及相對意義邸如第一圔之表格所示，該輪出 入辨識元（IO)為0或1邸分別表示输出或输入，而輪 出入變數（OIV)則代表不同受數號碼，若該輸出人變 數（OIV)為000,且输出入辨雄元（IO)亦為0 時，即頭示為浪下一法則之r法則位址差』，而於输出入 變數（OIV)為1 1 1時，印表示所有法則敍述结束（ END)，而在输出入變數（OIV)為其他值時，邸對 應於输出入辨戡元（IO)之〇或1狀態，分別為不同之 输出變數或不同之输入爱數。 經濟部中喪標準局〃«工消費合作社印贫 依照上述定義，實際配置方式上，即如第二圖所示， 依據位在圖面最上方之法則1之公式中，即有四個输入條 件及一値輸出結果，而法則2之公式中，則有二画輸入條 件及一値输出结果，據此，首先將各法則之起始位址（法 則1之第0位址）中之歸屬函數旗標（MEM)之值設為 此位址與下一法則（法則2 )之位址差（法則1位址〇之 MEM内容為0 1 0 1之二進位值，即表示其與下一法則 相差5,而位址1、 2、3之輸出入變數（OIV),則 -6 - 81.9.20.000 本纸張尺度適用中國國家標準（CNS)甲見格（210 X 297公釐） 213400 A6 _B6_ 五、發明説明（） (諳先閲汸背由之注念事,β再填寫本頁) 分別代表法則1之三健输入變數（w、x ' y )，而相應 之前導歸屬函數旗摞（MEM)之值即如公式中之A 1、 Bl、CO)，位址4之輸出入變數（OIV)及歸靥函 數旗標（MEM)則代表輸出受數e及後導歸屬函數旗欏 K0,而同理，法則2之起始位址（位址5)即設定其與 下—法則之位址差，而再依序為二涵输入受數、前導嫌靥 函數旗樣及一値輪出變數與後導歸屬函數旗標，而於整個 記億區之位址1 2 1則以輸出入辨斑元（1〇)及輸出入 變數（OIV)分別為〇及1 1 1，邸代表所有法則结束 ，據以停止推論蓮算作業。 經濟部中央標準总員工消費合作社印製 故依據上·述説明可知，本發明各法則.區域之配置方式 卽以各法則之起始位址存放『位址差J資料，而其後依序 為輸入變數、前導録屬函数旗標、最後才為輸出受數及後 導歸屬函數旗標，以形成本發明之記億配置方式，而於® 際運算上，於起始一蓮算法刖時（亦卽測知1 0 = 0，ο I V = 0 0 0 ),即將其歸屬函數旗標（MEM)之值與 現在位址相加蓮算，再予以存入至下一法刖之位址暫存器 内（以利後缅跳躍至下一法則），而於輸出入辨識元（工 〇)為1之各籲输入狀態下，邸根據输出入變數（OIV )與前導歸屬函數旗標（MEM)之值至前導歸屬記憧匾 中找出一前導函數四黏值，此四點值與輸入受數值一併送 入歸靨函數電路中，以處理取得一歸屬函數程度值。 於取出該歸屬函數程度值時，卽偵測此值是否為零， 81.9.20.000 本纸張度適用中aa家標準（CNS)甲4規格（210 X 297公货） ^ί^4〇 Ο Α6 Β6 則則至 變邸 址址暫則成區且標數配專 法法 a 入， 位位址法完® ，旗凾分符 此一跳 出束 始始位令 ，記間數屬區應 示下並 輪結 起起之即上則時函歸慊 ， 表出 ， 且述 則則則 ，址法箄屬導記者 即取丨。，敘 法法法時位等蓮嫌後之性 ，即址業 ο 則 各此 一零之此之及 \ 明步 零器位作為法 遇取下為指-需數導發進 為數始視}有 透讀入值所置所變前本及 若計起檢 ο 所 邸於存度器位則入及故化 ,則則述 I 示 要而加程存刖法出目 ，率 述法法前 {表 主，相數暫法效輸數求效 。 敘之一覆元邸 ，差址函址 | 無該數裔苻請 一部下重議 , 知址位屬位下短變變算更申 下内表以辨時 可位在歸則至縮改入蓮箄出 視 ，代 ，入 1 明之現得法躍邸要出的蓮提 檢發 {址出 1 說址與求一跳 ，需輸大論法 缰激值位输 1 施位容缅下速式際變較推依 繼數之始至為 實一内後至快方實改合糊爱 } 則變器起行 } 述下差也.跳而置應可符模 -< ,入存之進 V C 上與址以接則配因即以使件 明 零輸暫則當 I 理以入位 -直法及可數，可要 ^ 為披址法而 ο 處故存將内器發配亦元量即請 #a不法位一 ί 止 間，器數激分述·©數法申 q 若無的下 數停 空時存計未之上 t 之方利 五 ------------------'------装------tr------線一 (請先閲访背面之注念事項再塡寫本頁) 經濟部中與標準局員工消費合作社印¾ 本纸張尺度通用中國國家櫺準（CNS)甲4現格（210 X 297公ίί ) 81.9.20,000Printed 2194ϋ0 Α6 _BG_ by the R-Consumer Cooperative of the China Bureau of Standards, Ministry of Economic Affairs. 5. Description of the invention () The present invention is a method of allocating billions of regions for fuzzy inference rules that can shorten the processing time. Each address is divided into input and output identifiers, input and output variables, leading and trailing function flags, and the starting address between the plaques of each rule stored in the memory area is stored and the next rule starts. The starting address difference data (for the back-winding jump to the next rule), the middle position of the address interval stores the round acceptance number and the leading knot membership function, and the last position of the address interval stores the output variable and the trailing suspect羼 function. According to these methods, when reading the address difference data stored in the starting address, the current address can be added to the address difference data and stored in the address of the next rule In the temporary memory, when Burma reads the round acceptance number and the leading attribution function flag in turn and does it. The relevant calculation takes out the degree value of the return 1 function. If the degree value is zero, it means that the input variable cannot be excited. So jump directly to the next rule The value of the address register indicates the start address of the next rule to be processed. Therefore, the configuration method of the aforementioned rule section can be obtained by skipping the rule of non-excitation to achieve the benefit of improving the processing speed. According to the current FUZZY theory E, the class used has entered a stage of increasing popularity. For example, industrial machinery is used for lotus products. As for household electrical appliances, the lotus model lake theory items can be seen, and with the precision control and complex control In the above, Di uses Dazhi ’s fuzzy inference rules and the wishing lotus calculation, so that the speed of the lotus calculation is the main factor of its performance. Therefore, increasing the lotus calculation speed is the main work goal of the industry. The invention is mainly to achieve the concept of increasing the calculation speed. Since the traditional fuzzy lotus calculation methods are direct lotus calculations, for some lotus calculations that can be judged to be invalid or cause the output to be invalid, this paper ruler is also applicable. Park S home standard (CNS) V 4 is now available (210 X 297 public goods) 81.9.20.000 --------- L --------------- installed --- ---, 玎 ------ '^ < Please read the notes on the back to save this page and write this page) V. Description of the invention () A6 B6 Printed and implemented by the Ministry of Economic Affairs, the Ministry of Economy, China and the People ’s Consumer Cooperative If the input to the Jianlou is false, the initial value will be temporarily stored. If you jump to 0, you can tear it into "The address of the time. Waiting for the game to calculate the jump, the first place after the flag is lost, and the Chengdi lotus mansion. When passing the lotus, the lottery number will be released and the number containing the digit will be described. Before the change to the law before the change method is cocoa, the place where the law is set changes to jump to the judgment. The reason is that the Di method is short and the standard method is different. If the Di method is lost in the next step, if the next match is between the master and the leader, the flags will be reduced and the upper segment will lose-one after the other. Calculate the original version of the direct address value before and after the current expression. This kind of pseudo function is destructurized, and the storage address is stored in the position according to the position. It's only a short period of time for him to start his course. After the prince returns to the position of the bureaucracy, it starts with a number of stages, and the person who distributes the contraction mentions, and recognizes the difference between the adder and the adder to the letter. After the three sites are formed, the coexistence sub-method will be integrated and will not be changed-the effect will be subordinate to Fang Ai. The normal value is temporarily based on the lack of a short energy. In the next case, the effectiveness of the retrenchment between the retrenchment points and the meta-data is used to take the flag and then take the rule until the law is up to the time when there is no travel speed. S, the number of readings, and the number of readings is taken by the algorithm. In the calculation, it is the master of the calculation. ± The letter should be wrapped up now. If you jump here, you can enter the lotus in the lotus area. The type + law-and the back. In the Ming Dynasty, a box of standard materials will be cut out for the flag table, and the law will be issued. When it is remembered, the law will be issued, the input will be divided, and the guide flag will be used. The difference between the previous numbers and the position function is effective if the fee method is introduced into the input section and the zero device in the address. This is a no-loss theory. During the period, the number stored in the output area will be classified as a deposit. According to ------ „----.--------------- installed ------ ordered ------ Line — (please read the notes on the back and then fill out this page) This paper size Λ is applicable to the Chinese National Standard (CNS) r said grid (210 X 297 mm) 81.9.20.000 _____ Central Standards Bureau of the Ministry of Hydrocarbon Economy g Work Consumers ¢ · Cooperatives s A6 _BG _ V. Description of the invention () and other purposes, attached with drawings to explain in detail as follows: (1). Part of the drawings: the first picture: the identification element of the false invention and The second table of the definition table of the input and output variables: the embodiment of the memory area configuration of the pseudo-invention. The reason why the invention invalidates or is not stimulated by the input variables is explained as follows based on the reasons for skipping: In the process, no matter whether the maximum / minimum combination method or the maximum. Product combination method is used, if any of the majority of the degree values of the attribution function obtained in the operation is zero, then after the minimum or product operation, the corresponding The degree value of the output attribution function must be zero (A Ο logical operation). Thus, if the degree value of the other rounded attribution function wrapped after it is so loved, it will not cause the difference in the output results, so the traditional building operations are all For direct lotus to knot This will result in unnecessary waste of computing time and lack of benefit. Therefore, the present invention detects when one of the input rules of each rule is zero, Di skips the next rule and skips to the next rule. Repeat the above steps, so as to avoid the waste of lotus calculation time, and accordingly improve the efficiency of lotus calculation. To achieve the aforementioned jump (JU M P) to the next rule, it is necessary to perform lotus in the previous rule When calculating, it is necessary to record the starting position of the next rule together. When the aforementioned detection is invalid (not excited), Di can jump to the next rule position according to the starting position of the next rule The lotus calculation is performed, so the present invention provides a method of assigning billions of regions that is different from the traditional fuzzy theory rule to achieve the aforementioned rapid jump effect and shorten the calculation inference time. The following paper is again applicable to the Chinese Gardener Standard (CNS ) A 4 present grid (210 X 297 mm) -----: ---.----------------- installed ------ ΤΓ --- --- Elephant (please read the back of Kang first; read the matter before writing this page) 81.9.20,000 319400 A6 _B6 __ V. Description of invention () (please read first Note the matters on the back of the page and write this page again) Di Yi has a rule of inference for modules with eight inputs (each input has a sixteen-lead leading junction function) and six outputs (each output has a sixteen-lead leading attribution function) Taking the Yiyi plaque as an example, each address contains a one-bit input / output identification unit (ίο), a three-bit input / output variable (οι V), and a four-bit attribution function flag. (MEM) (such as the second 圔), and the combination of the input-output identification element (1〇) and the input-output acceptance number (〇IV) is as shown in the table of the first 圔. The input and output identifier (IO) is 0 or 1 D, respectively means output or input, and the round access variable (OIV) represents different acceptor numbers, if the output variable (OIV) is 000, and the output input and output (IO) When it is also 0, that is, the r-law address difference shown as the first law of the wave is shown below, and when the input-output variable (OIV) is 1 1 1, the seal indicates the end of all the rules (END), and the output variable (OIV) is other value, Di corresponds to the 0 or 1 state of the input and output unit (IO), which are different Enter the number of different variables or love. According to the above definition, the actual bureaucracy of the Ministry of Economic Affairs Standards Bureau 〃 «Poverty Printing in Industrial and Consumer Cooperatives, as shown in the second figure, is based on the formula of rule 1 at the top of the map, that is, there are four input conditions And a value output result, and in the formula of rule 2, there are two input conditions and a value output result. According to this, the attribution function in the starting address of each rule (the 0th address of rule 1) is first The value of the flag (MEM) is set to the difference between the address of this address and the next rule (Law 2) (Law 1 Address 0 The MEM content is 0 1 0 1 binary value, which means it is the next rule There is a difference of 5, and the input and output variables (OIV) of addresses 1, 2, and 3 are -6-81.9.20.000. The paper size is applicable to the Chinese National Standard (CNS) Jiajiange (210 X 297 mm) 213400 A6 _B6_ Fifth, the invention description () (know first read the memorable things, and then fill in this page) respectively represent the three key input variables (w, x 'y) of rule 1 and the corresponding leading attribution function stack ( The value of MEM) is the same as A1, Bl, CO in the formula, and the input / output variable (OIV) and return function flag (MEM) of address 4 are substituted. The table output is subject to the number e and the trailing attribution function flag K0. Similarly, the starting address of rule 2 (address 5) is to set the difference between the address of the next rule and the second order input. Receiving number, leading suspected function flag and a round of variable and trailing attribution function flag, and the address 1 2 1 in the entire billion-counting area is identified by the input and output of the spot element (10) and the input and output variable (OIV) are 0 and 11 respectively, Di stands for the end of all the laws, and accordingly stops the inference of lotus calculation. Printed by the Central Standards General Staff Consumer Cooperative of the Ministry of Economic Affairs. Based on the above description, it can be seen that the rules of the present invention are arranged in the region. The "address difference J data" is stored at the starting address of each rule, and then the order is as follows The input variable, the leading record function flag, and finally the output acceptor number and the trailing belonging function flag form the billion configuration method of the present invention, and on the international calculation, when the first lotus algorithm starts ( It is also known that 1 0 = 0, IV = 0 0 0), that is, add the value of its attribution function flag (MEM) to the current address, and then save it to the address of the next law. In the memory (Eliman jumps to the next rule), and in the input state where the input and output identification element (work 〇) is 1, Di according to the input and output variables (OIV) and the leading attribution function flag (MEM ) To the leading attribution record to find a four-viscosity value of the leading function. This four-point value is sent to the return function circuit together with the input receiving value to process and obtain a degree of attribution function. When the degree value of the attribution function is taken out, it is detected whether the value is zero. 81.9.20.000 of the paper size is applicable to the CNA standard A4 specification (210 X 297 public goods) ^ ί ^ 4〇Ο Α6 Β6 will be changed to the address of the residence, and the standard number is equipped with the special method a, the bit address method is completed®, the flag symbol breaks out of the beginning of the beam order, and the number of zones belongs to the area should be shown below When the time comes, the letter will be returned to the previous letter, and it will be shown. ， Syntactic method, law position, etc., the later nature of the lotus, that is, the address industry ο Then each one of this and the zero \ Mingbu zero device position is taken as a method to be taken as a finger-the number of leads is sent to the number of sights} If there is a copy of the value before the change of the read-in value and if it is checked, it will be entered and degraded if the residence is in the memory position, then it will be described as I and the extra-stored storage method will appear. The main, the phase number temporarily law effect loses the number to seek the effect. One of the categories covers the Yuandi, the address of the difference address | If there is no such number, please review the next one. If you know that the location is short, you should apply for the internal table to distinguish it. Lianshu looked out, replaced it, entered 1 Ming's present Fayuedi's lotus was issued and checked out {address out 1 said address and seeking a jump, you need to lose the big argument, the law value, lose 1 and place the tolerance to the speed. Change comparison pushes from the beginning of the number to the real one and then to the fast side of the real love. Then the transformer starts to run} The difference is also described. Jumping and setting should be able to match the pattern-<, into the VC The connection is based on the location and the allocation factor is to make the data clear and the input is temporary. The reason is to use the in-direct method and the countable. It can be ^ for the location method and the existing storage will be allocated to the internal device. Please #a 不法 位 一 ί Stop, the number of devices is violently described. © 数 法 申 q If there is no countdown, the number saved is not above t's Fang Liwu ----------- -------'------ installed ------ tr ------ line one (please read the notes on the back first and then write this page) Printed by the Staff Consumer Cooperative of the Bureau of Standards ¾ This paper standard is common to China National Standards (CNS) Grade A 4 (210 X 297 公 ί) 81.9.20,000