This algorithm recovers the absence of the empiric in the case of the fixed - topology network and generates an optimal topology automatically . we end this chapter with some problems in the future . in chapter 2 , we present an evolution strategy to infer fuzzy finite - state automaton , the fitness function of a generated automaton with respect to the set of examples of a fuzzy language , the representation of the transition and the output of the automaton and the simple mutation operators that work on these representations are given 目前,國(guó)內(nèi)外對(duì)神經(jīng)網(wǎng)絡(luò)與自動(dòng)機(jī)的結(jié)合的研究己取得了一系列成果;在第一章,我們首先將對(duì)這些結(jié)果以及這個(gè)領(lǐng)域的研究思想與方法做一個(gè)概要的介紹;然后提出一種推導(dǎo)模糊有限狀態(tài)自動(dòng)機(jī)的構(gòu)造性算法,解決了仿真實(shí)驗(yàn)中所給出的具體網(wǎng)絡(luò)的隱藏層神經(jīng)元個(gè)數(shù)的確定問(wèn)題;在實(shí)驗(yàn)中,我們首先將樣本輸入帶1個(gè)隱藏層神經(jīng)元的反饋網(wǎng)絡(luò)訓(xùn)練, 150個(gè)紀(jì)元以后增加神經(jīng)元,此時(shí)的新網(wǎng)絡(luò)在124紀(jì)元時(shí)收斂;而blanco [ 3 ]的固定性網(wǎng)絡(luò)學(xué)習(xí)好相同的樣本需要432個(gè)紀(jì)元。