資源描述:
《轉(zhuǎn)爐煉鋼終點(diǎn)靜態(tài)控制預(yù)測模型》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫。
1、中文摘要終點(diǎn)控制是轉(zhuǎn)爐吹煉后期的一項(xiàng)重要操作。天鐵冶金集團(tuán)煉鋼廠采用的是人工經(jīng)驗(yàn)控制,終點(diǎn)命中率很低,延長了吹煉時(shí)間,增加金屬料消耗,影響鋼的品質(zhì)。當(dāng)前,主要采用靜態(tài)和動(dòng)態(tài)相結(jié)合的控制方法。機(jī)理模型、統(tǒng)計(jì)模型及人工神經(jīng)網(wǎng)絡(luò)模型是目前常用的模型。機(jī)理模型在實(shí)際應(yīng)用中涉及的參數(shù)較多且不易測取與控制,準(zhǔn)確建模還有相當(dāng)?shù)睦щy。如果生產(chǎn)條件和工藝操作方法比較穩(wěn)定,轉(zhuǎn)爐煉鋼冶煉特性具有較好的再現(xiàn)性,這時(shí)可以采用反映終點(diǎn)鋼水碳含量與溫度的統(tǒng)計(jì)模型。由于轉(zhuǎn)爐煉鋼過程高度復(fù)雜、隨機(jī)性強(qiáng),采用具有很強(qiáng)的自學(xué)習(xí)能力、容錯(cuò)能力和推理能力的人工神經(jīng)網(wǎng)絡(luò)技術(shù)對轉(zhuǎn)爐煉鋼終點(diǎn)控制提供了一條新
2、的實(shí)現(xiàn)途徑。基于天鐵冶金集團(tuán)30噸轉(zhuǎn)爐煉鋼實(shí)際生產(chǎn)工況數(shù)據(jù),首先建立了轉(zhuǎn)爐煉鋼終點(diǎn)靜態(tài)控制的吹氧量及礦石用量統(tǒng)計(jì)模型,其預(yù)測100個(gè)爐次的吹氧量及礦石用量平均相對誤差分別為O.58%及10.4%??紤]到影響終點(diǎn)鋼水溫度及碳含量的因素比較復(fù)雜,本文設(shè)計(jì)了預(yù)測鋼水終點(diǎn)溫度及碳含量的人工神經(jīng)網(wǎng)絡(luò)模型,并利用Levenberg.Marquardt算法對257個(gè)爐次實(shí)際生產(chǎn)數(shù)據(jù)進(jìn)行了模型訓(xùn)練,并對另外100個(gè)爐次的終點(diǎn)鋼水溫度及碳含量進(jìn)行了預(yù)測,在終點(diǎn)鋼水溫度1646℃~1698℃及終點(diǎn)碳含量0.033%~0.128%范圍內(nèi),得到的終點(diǎn)碳溫雙命中率為55%。應(yīng)用顯示,明
3、顯地提高了終點(diǎn)鋼水溫度及碳含量的預(yù)測精度。關(guān)鍵詞:轉(zhuǎn)爐煉鋼,終點(diǎn)控制,統(tǒng)計(jì)模型,人工神經(jīng)網(wǎng)絡(luò)模型ABSTRACTTheend—pointcontrolisanimportantoperationinthelaterperiodofLDconvertersteelmaking.ArtificialexperiencecontrolmethodisstillusedintheTrMGsteelplant,theend—pointhittingrateisverylow.Asaresult,itprolongsthesmeltingtime,increasesthe
4、metalconsumeandinfluencesthesteelquality.Atpresent,theadvancedcontrolmethodisthecombinationofstaticanddynamiccontr01.Currentlythemechanismmodel,statisticalmodelandartificialneuralnetworkmodelareincommonuse.Inpracticalapplications,themechanismmodelisinvolvedinmanyparametersthataredif
5、ficulttomeasureandcontrol,SOtheaccuratemodelingisstillverydifficult.Iftheconvertersmeltingcharacteristichasthegoodreproducibilityandtheproductionconditionandthecraftoperationmethodarestable,wecanusethestatisticalmodeltopredictemoltensteelfinaltemperatureandcarboncontent.Becausetheco
6、nvertersteelmakingprocessishighlycomplexandrandom,andANNhastheverystrongself-learning,faulttolearnandinferencecapability,itprovidesanewimplementwaytOthestaticcontroloftheLDsteelmakingprocess.Basedonthe30tproductiondatainconvertersteelmakingprocesswiththeTTMGfirstlytheblowingoxygenan
7、dorequantitiesmodelsoffinalstaticcontrolinconvertersteelmakingprocesswereestablishedandthepredictionaccuracywas0.58%and10.4%respectively.Duetothecomplicationseffectonthefinaltemperatureandcarboncontentofmoltensteel,weusedtheartificialneuralnetworkmodelwiththeLevenberg-Marquardtalgor
8、ithm,the257producti