(東北大學(xué)自動(dòng)化研究中心， 沈陽(yáng) 110006)
摘 要: 板形檢測(cè)信息的模式分解是板形控制過程中的技術(shù)難點(diǎn)，該文提出的一種新的神經(jīng)網(wǎng)絡(luò)模式識(shí)別方法卻可以解決這個(gè)難題。 該識(shí)別方法的優(yōu)點(diǎn)是 ： 在ART網(wǎng)絡(luò)的特征表示場(chǎng)中采用了具有正反饋和非線性變換的結(jié)構(gòu)，能夠有效地抑制板形檢測(cè)數(shù)據(jù)中的干擾影響， 提高了模式識(shí)別系統(tǒng)的抗干擾能力； 在類別場(chǎng)中拋棄了傳統(tǒng)的競(jìng)爭(zhēng)學(xué)習(xí)機(jī)制， 新的學(xué)習(xí)機(jī)制可以迅速分解板形模式； 按照軋機(jī)執(zhí)行機(jī)構(gòu)板形控制的能力設(shè)置標(biāo)準(zhǔn)板形模式，可以對(duì)任意復(fù)雜形式的板形缺陷進(jìn)行控制。 用這種識(shí)別方法對(duì)實(shí)測(cè)板形進(jìn)行了模式分解， 識(shí)別結(jié)果完全正確， 充分說(shuō)明ART神經(jīng)網(wǎng)絡(luò)識(shí)別方法是一種理想的板形模式識(shí)別方法。
關(guān)鍵字: 關(guān)鍵詞: 板形檢測(cè) 模式識(shí)別 ART神經(jīng)網(wǎng)絡(luò)
（Research Center of Automation, Northeastern University, Shenyang 110006, P. R. China）
Abstract:Pattern decomposition of shape measurement is one of the difficult techniques in shape control system. A novel pattern recognition method based on neural network was presented in detail, its superiorities lie in the following points: positive feedback and nonlinear transforming structure are introduced in the representing field of ART neural network, so that the influences of disturbances exiting in shape measurements are rejected and in turn the system is improved in disturbance rejection. Traditional competitive learning mechanism in the type field of ART neural network was abandoned and replaced by a new learning method, which is very quick at completing shape pattern decomposition; standard shape patterns are set up according to the shape control capability of the actuating units in the rolling mill, so that shape defects of any type can be control led. When the new pattern recognition method is used in decomposing real shape measurements, completely correct results are obtained, this means that the recognition method based on ART neural network is an ideal shape recognition method.
Key words: shape measure pattern recognition ART neural network
