空分裝置多股流換熱器的模擬與性能分析

ABSTRACT-1g）目標(biāo)函數(shù)（4-1a）是由兩個(gè)部分構(gòu)成，第一部分由設(shè)備兩端出口溫度的計(jì)算值與測(cè)量值的誤差總和構(gòu)成，第二部分由設(shè)備中間層溫度的計(jì)算值和同一層有測(cè)量點(diǎn)的測(cè)量值誤差總和構(gòu)成。由于MHEX復(fù)雜的結(jié)構(gòu)特點(diǎn)決定了溫度分布形式，在同一個(gè)橫截面上（圖4-1）各熱流股之間溫度接近，各冷流股之間溫度接近，熱流股與冷流股之間的溫差接近最小溫差，為了約束這些溫度，基于現(xiàn)場(chǎng)數(shù)據(jù)觀察，本文假定：熱流股之間溫差為0℃；冷流股之間溫差為0℃；最小溫差為1℃。目標(biāo)函數(shù)第二部分就是由這些約束構(gòu)成。公式（4-1b～g）構(gòu)成約束方程，（4-1b）表示任一段b（Bundle）的回歸模型，其中β表示被估參數(shù)。傳遞方程（4-1c）和（4-1d）表示上一層模型的出口溫度傳遞給下一層模型的入口溫度。方程（4-1e）和（4-1f）保證同一截面上最小溫差1℃。最后，方程（4-1g）表示被估參數(shù)的結(jié)構(gòu)特征，例如當(dāng)熱流股流量增加其出口溫度隨之增大，冷流股流量增加其出口溫度減小等。復(fù)雜MHEX模型由多個(gè)Bundle的回歸模型聯(lián)立構(gòu)成，每個(gè)Bundle由一個(gè)簡(jiǎn)單MHEX模型控制（模型推導(dǎo)見2.2），單個(gè)Bundle的回歸模型只需將冷熱流股數(shù)帶入公式（2-5）擴(kuò)展得到，三個(gè)Bundle的具體回歸模型見公式（4-2），（4-3），（4-4），其中,表示熱流股i在第b個(gè)Bundle的流量，表示冷流股j在第b個(gè)Bundle的流量，表示熱流股i在第b個(gè)Bundle的進(jìn)口溫度，表示冷流股j在第b個(gè)Bundle的進(jìn)口溫度，β表示被估參數(shù)的集。表示任意流股i/j在第b個(gè)Bundle的流量，表示任意流股i/j在第b個(gè)Bundle的出口溫度。Bundle1的回歸模型由五股流股28個(gè)結(jié)構(gòu)參數(shù)組成變量組，回歸結(jié)構(gòu)如下， （4-2）Bundle2的回歸模型由六股流股39個(gè)結(jié)構(gòu)參數(shù)組成變量組，回歸結(jié)構(gòu)如下， （4-3）Bundle3的回歸模型由五股流股28個(gè)結(jié)構(gòu)參數(shù)組成變量組，回歸結(jié)構(gòu)如下， （4-4）4.2多層出口的復(fù)雜多股流換熱器模型求解與結(jié)果4.2.1多層出口的復(fù)雜多股流換熱器模型求解按上述方法對(duì)空分裝置主換熱器和過冷器建模，建立一個(gè)多Bundle聯(lián)立回歸模型來模擬冷量回收的過程。復(fù)雜MHEX在GeneralAlgebraicModelingSystem(GAMS)中建立模型，采用聯(lián)立方程法求解。回歸問題被再構(gòu)成一個(gè)二次優(yōu)化問題，搜索一組最優(yōu)的結(jié)構(gòu)參數(shù)，使得出口溫度的計(jì)算與測(cè)量值誤差最小。采用CONOPT求解器求解非線性問題，具體的求解程序見第四章末尾的程序附錄。我們需要一組結(jié)構(gòu)參數(shù)的初值去搜索聯(lián)立方程的最優(yōu)解。初值很大程度上影響計(jì)算時(shí)間，而且差的初值很可能導(dǎo)致局部最優(yōu)解而非全局最優(yōu)解。因此，初值的求解是模型求解的第一步工作，這是一個(gè)迭代求解的過程見圖4-2，全部在Matlab@R2011b進(jìn)行。圖4-SEQ圖4-\*ARABIC2復(fù)雜模型初值計(jì)算過程首先將每一個(gè)Bundle的訓(xùn)練數(shù)據(jù)做簡(jiǎn)化處理，假定中間層熱流股溫度相同，冷流股溫度相同，且熱流股與冷流股溫差為1℃，利用Matlab數(shù)據(jù)庫(kù)中的Regress函數(shù)先對(duì)Bundle1進(jìn)行最小二乘回歸得到一組結(jié)構(gòu)參數(shù)，用結(jié)構(gòu)參數(shù)和入口溫度重新計(jì)算Bundle1出口溫度，將熱流股溫度傳遞給Bundle2的入口溫度，同樣的方法回歸Bundle2結(jié)構(gòu)，計(jì)算Bundle2熱流股出口溫度傳遞熱流給Bundle3，回歸Bundle3結(jié)構(gòu)參數(shù)計(jì)算冷流股出口溫度傳遞回Bundle2，依次按照?qǐng)D4-2的順序迭代。值得一提的是與其讓迭代達(dá)到局部最優(yōu)解，不如僅僅只迭代最主要的5次，得到三個(gè)Bundle模型的結(jié)構(gòu)參數(shù)作為聯(lián)立方程求解的初值。在求解初值時(shí)，對(duì)于沒有測(cè)量的數(shù)據(jù)，根據(jù)MHEX傳熱結(jié)構(gòu)的特征，假定在MHEX同一截面上沒有測(cè)量點(diǎn)的熱流股溫度近似等于有測(cè)量點(diǎn)的熱流股溫度，同理冷流股亦是，熱流股的溫度近似比冷流股高1℃，流股流量由一個(gè)Bundle傳遞到另一個(gè)Bundle,只要流股沒有變化，流量也等于前一個(gè)Bundle的流量。初值求解的結(jié)果包括每一個(gè)Bundle中每一股流股的結(jié)構(gòu)參數(shù)，進(jìn)口溫度，出口溫度和流量。4.2.2多層出口的復(fù)雜多股流換熱器模型求解結(jié)果與討論將初值傳遞給GAMS，開始聯(lián)立模型的求解，主換熱器的回歸模型計(jì)算如表4-1所示，復(fù)雜模型包含了39,423個(gè)變量，100,501個(gè)約束（其中等式約束為37,500個(gè)），用時(shí)253分鐘在1500訓(xùn)練樣本中搜索到最優(yōu)的結(jié)構(gòu)參數(shù)，Bundle1由兩股熱流股和三股冷流股組成，回歸得到5組，每組28結(jié)構(gòu)參數(shù)。Bundle2由三股熱流股和三股冷流股組成，回歸得到6組，每組39結(jié)構(gòu)參數(shù)。Bundle1由兩股熱流股和三股冷流股組成，回歸得到5組，每組28結(jié)構(gòu)參數(shù)。表4-SEQ表4-\*ARABIC1模型計(jì)算統(tǒng)計(jì)變量約束回歸時(shí)間計(jì)算時(shí)間計(jì)算精度多層出口復(fù)雜MHEX模型39,423100,501253分鐘17秒2%簡(jiǎn)單MHEX模型5070.12秒0.00138秒1%然后再用結(jié)構(gòu)參數(shù)求解1000組測(cè)試樣本的出口溫度，此過程亦是一個(gè)聯(lián)立計(jì)算的過程，用時(shí)17秒。將出口溫度的計(jì)算值與測(cè)試值比較來檢驗(yàn)?zāi)Ｐ凸浪愕臏?zhǔn)確性。對(duì)于訓(xùn)練樣本，最優(yōu)結(jié)構(gòu)參數(shù)在多Bundle模型中被固定去測(cè)試出口溫度的計(jì)算值和測(cè)量值。當(dāng)結(jié)構(gòu)參數(shù)在測(cè)試模型中被固定，則模型自由度為0，變成一個(gè)線性模擬問題。雖然多層出口的復(fù)雜模型在計(jì)算上確實(shí)比第二章的簡(jiǎn)單模型要費(fèi)時(shí)的多，在計(jì)算進(jìn)度上也不如簡(jiǎn)單模型，但是多層出口的復(fù)雜模型更合乎流股換熱的內(nèi)在關(guān)系，更能反映實(shí)際換熱過程，且能夠處理某些流股缺乏儀表測(cè)量的情況。用CPLEX求解器只需要幾秒鐘就能求解出線性模型的解，訓(xùn)練樣本和測(cè)試樣本的誤差見圖4-3。圖4-3畫出了七個(gè)流股訓(xùn)練數(shù)據(jù)出口溫度的回歸誤差與測(cè)試數(shù)據(jù)出口溫度的模擬誤差，圖中上層實(shí)線表示出口溫度的誤差，對(duì)應(yīng)左邊Y軸坐標(biāo)，下層實(shí)線表示主流空氣H3的進(jìn)口溫度（記TMinH3），對(duì)應(yīng)右邊Y軸坐標(biāo)，圖中虛線左邊是訓(xùn)練樣本，右邊是測(cè)試樣本。從圖中可以看出，訓(xùn)練樣本的誤差基本分布在-2℃到2℃之間，個(gè)別誤差大于2℃，最大最對(duì)誤差小于3℃。誤差較為均勻的分布在0軸的兩側(cè)，表明回歸模型得到了很好的訓(xùn)練。其次我們發(fā)現(xiàn)訓(xùn)練樣本的大誤差大多發(fā)生在主流空氣H3的進(jìn)口溫度劇烈波動(dòng)時(shí)，其原因同樣是分子篩周期性切換引起。當(dāng)空分換熱系統(tǒng)偏離正常穩(wěn)態(tài)操作的一段時(shí)間內(nèi)或者說面臨工況切換時(shí)，會(huì)導(dǎo)致了回歸模型的不準(zhǔn)確性。如果只考慮穩(wěn)態(tài)模擬，就不應(yīng)該將這些變負(fù)荷工況動(dòng)態(tài)切換過程的操作數(shù)據(jù)納入回歸范圍，這將明顯提高多層出口的復(fù)雜MHEX模型的回歸精度，若是考慮變負(fù)荷操作動(dòng)態(tài)切換過程，則出于對(duì)測(cè)量數(shù)據(jù)不準(zhǔn)確性的考慮，應(yīng)該對(duì)測(cè)量數(shù)據(jù)做標(biāo)定預(yù)處理。圖4-SEQ圖4-\*ARABIC3回歸結(jié)果誤差分析復(fù)雜MHEX模型回歸和測(cè)試的平均相對(duì)誤差和最大絕對(duì)誤差見表4-2。從圖4-3和表4-2中可以發(fā)現(xiàn)，雖然考慮多層出口的復(fù)雜MHEX模型訓(xùn)練結(jié)果和測(cè)試結(jié)果的最大誤差和平均誤差都比第二章考慮單層出口的MHEX模型要大一些，一方面是由于目前模型的約束條件做了簡(jiǎn)化處理還不夠完善，缺乏多股流傳熱結(jié)構(gòu)的約束，另一方面是模型缺乏一個(gè)更高級(jí)的算法。但是考慮多層出口的復(fù)雜MHEX模型應(yīng)用性更普遍，是對(duì)只考慮單層出口模型的簡(jiǎn)單MHEX模型的擴(kuò)展，若模型只有單層出口對(duì)應(yīng)復(fù)雜模型只有一個(gè)Bundle的情況，若模型流股數(shù)眾多，有多層出口，且某些流股缺乏測(cè)量數(shù)據(jù)時(shí)，考慮單層出口模型的簡(jiǎn)單MHEX模型將不再適用，但是由于考慮多層出口的復(fù)雜MHEX模型是一個(gè)優(yōu)化命題，可以通過條件約束來解決。表4-SEQ表4-\*ARABIC2訓(xùn)練與測(cè)試誤差分析H1H2H3C1C2C3C4TrainingdataMean*0.2910.0900.1330.1350.1470.3680.135Max*1.9220.5280.6330.9460.9182.9721.199CheckingdataMean0.4990.1550.2180.1450.1410.5870.172Max2.5250.5480.7231.1470.9362.9851.502*Intable1,“Mean”referstomeanabsoluteerror,“Max”referstomaxabsoluteerror.4.3小結(jié)本章考慮流股常常從MEHX中間進(jìn)入或離開，改進(jìn)建立了考慮多層出口的復(fù)雜MHEX聯(lián)立捷算模型，這個(gè)模型不再用傳統(tǒng)處理黑箱的方法將整個(gè)MHEX看成一個(gè)黑箱，而是將MHEX在中間進(jìn)出的位置分成多段，每一段記一個(gè)Bundle，每個(gè)Bundle是一個(gè)黑箱由一個(gè)模型控制。復(fù)雜MHEX模型由多個(gè)Bundle的回歸模型聯(lián)立構(gòu)成，由于有中間變量的存在，不再是輸入變量的顯函數(shù)。模型的預(yù)測(cè)平均誤差不大于2%，平均絕對(duì)誤差小于0.6℃。此模型更符合設(shè)備結(jié)構(gòu)的傳熱模式且能解決某些流股缺乏測(cè)量數(shù)據(jù)的問題。程序附錄*MHEXregressionSETSIflow/17,18,19,21,23,25,32/BBundle/1*3/INOUT1(I)in(out)flowofeachBundle/17*19,21,23/INOUT2(I)in(out)flowofeachBundle/17*19,21,25,32/INOUT3(I)in(out)flowofeachBundle/18,19,21,25,32/HOThotflow/17,18,25/COLDcoldflow/19,21,23,32/Ndata/1*120/M1numberregressionstructure1/1*29/M2numberregressionstructure2/1*40/M3numberregressionstructure3/1*29/;PARAMETERSTMin(I,B,N)inlettemperaturemeasurementTMout(I,B,N)outlettemperaturemeasurementFM(I,N)flowratemeasurementVARIABLESTCin(I,B,N)calculatedtemperatureTCout(I,B,N)calculatedtemperatureoutK1(INOUT1,M1)regressioncoefficientsforeachBundleK2(INOUT2,M2)regressioncoefficientsforeachBundleK3(INOUT3,M3)regressioncoefficientsforeachBundleZdummyobjectiveEQUATIONSREGRESSION1(INOUT1,N)regressionfunctionsREGRESSION2(INOUT2,N)regressionfunctionsREGRESSION3(INOUT3,N)regressionfunctionsCONNECTIONS1(N)connectionequationsCONNECTIONS2(N)connectionequationsCONNECTIONS3(N)connectionequationsCONNECTIONS4(N)connectionequationsCONNECTIONS5(N)connectionequationsCONNECTIONS6(N)connectionequationsCONNECTIONS7(N)connectionequationsCONNECTIONS8(N)connectionequationsCONNECTIONS9(N)connectionequationsOBJdummyobjectivefunciton;OBJ..z=e=sum(N,sqr(TCout("19","1",N)/TMout("19","1",N)-1)+sqr(TCout("21","1",N)/TMout("21","1",N)-1)+sqr(TCout("23","1",N)/TMout("23","1",N)-1)+sqr(TCout("17","2",N)/TMout("17","2",N)-1)+sqr(TCout("32","2",N)/TMout("32","2",N)-1)+sqr(TCout("18","3",N)/TMout("18","3",N)-1)+sqr(TCout("25","3",N)/TMout("25","3",N)-1));REGRESSION1(INOUT1,N)..Fm(INOUT1,N)*TCout(INOUT1,"1",N)=E=K1(INOUT1,"1")+K1(INOUT1,"2")*FM("17",N)*FM("17",N)+K1(INOUT1,"3")*FM("18",N)*FM("18",N)+K1(INOUT1,"4")*FM("19",N)*FM("19",N)+K1(INOUT1,"5")*FM("21",N)*FM("21",N)+K1(INOUT1,"6")*FM("23",N)*FM("23",N)+K1(INOUT1,"7")*FM("17",N)*FM("19",N)+K1(INOUT1,"8")*FM("17",N)*FM("21",N)+K1(INOUT1,"9")*FM("17",N)*FM("23",N)+K1(INOUT1,"10")*FM("18",N)*FM("19",N)+K1(INOUT1,"11")*FM("18",N)*FM("21",N)+K1(INOUT1,"12")*FM("18",N)*FM("23",N)+K1(INOUT1,"13")*FM("17",N)*TMin("17","1",N)+K1(INOUT1,"14")*FM("18",N)*TMin("18","1",N)+K1(INOUT1,"15")*FM("19",N)*TCin("19","1",N)+K1(INOUT1,"16")*FM("21",N)*TCin("21","1",N)+K1(INOUT1,"17")*FM("23",N)*TMin("23","1",N)+K1(INOUT1,"18")*FM("17",N)*TCin("19","1",N)+K1(INOUT1,"19")*FM("17",N)*TCin("21","1",N)+K1(INOUT1,"20")*FM("17",N)*TMin("23","1",N)+K1(INOUT1,"21")*FM("18",N)*TCin("19","1",N)+K1(INOUT1,"22")*FM("18",N)*TCin("21","1",N)+K1(INOUT1,"23")*FM("18",N)*TMin("23","1",N)+K1(INOUT1,"24")*FM("19",N)*TMin("17","1",N)+K1(INOUT1,"25")*FM("19",N)*TMin("18","1",N)+K1(INOUT1,"26")*FM("21",N)*TMin("17","1",N)+K1(INOUT1,"27")*FM("21",N)*TMin("18","1",N)+K1(INOUT1,"28")*FM("23",N)*TMin("17","1",N)+K1(INOUT1,"29")*FM("23",N)*TMin("18","1",N);REGRESSION2(INOUT2,N)..Fm(INOUT2,N)*TCout(INOUT2,"2",N)=E=K2(INOUT2,"1")+K2(INOUT2,"2")*FM("17",N)*FM("17",N)+K2(INOUT2,"3")*FM("18",N)*FM("18",N)+K2(INOUT2,"4")*FM("25",N)*FM("25",N)+K2(INOUT2,"5")*FM("19",N)*FM("19",N)+K2(INOUT2,"6")*FM("21",N)*FM("21",N)+K2(INOUT2,"7")*FM("32",N)*FM("32",N)+K2(INOUT2,"8")*FM("17",N)*FM("19",N)+K2(INOUT2,"9")*FM("17",N)*FM("21",N)+K2(INOUT2,"10")*FM("17",N)*FM("32",N)+K2(INOUT2,"11")*FM("18",N)*FM("19",N)+K2(INOUT2,"12")*FM("18",N)*FM("21",N)+K2(INOUT2,"13")*FM("18",N)*FM("32",N)+K2(INOUT2,"14")*FM("25",N)*FM("19",N)+K2(INOUT2,"15")*FM("25",N)*FM("21",N)+K2(INOUT2,"16")*FM("25",N)*FM("32",N)+K2(INOUT2,"17")*FM("17",N)*TCin("17","2",N)+K2(INOUT2,"18")*FM("18",N)*TCin("18","2",N)+K2(INOUT2,"19")*FM("25",N)*TMin("25","2",N)+K2(INOUT2,"20")*FM("19",N)*TCin("19","2",N)+K2(INOUT2,"21")*FM("21",N)*TCin("21","2",N)+K2(INOUT2,"22")*FM("32",N)*TCin("32","2",N)+K2(INOUT2,"23")*FM("17",N)*TCin("19","2",N)+K2(INOUT2,"24")*FM("17",N)*TCin("21","2",N)+K2(INOUT2,"25")*FM("17",N)*TCin("32","2",N)+K2(INOUT2,"26")*FM("18",N)*TCin("19","2",N)+K2(INOUT2,"27")*FM("18",N)*TCin("21","2",N)+K2(INOUT2,"28")*FM("18",N)*TCin("32","2",N)+K2(INOUT2,"29")*FM("25",N)*TCin("19","2",N)+K2(INOUT2,"30")*FM("25",N)*TCin("21","2",N)+K2(INOUT2,"31")*FM("25",N)*TCin("32","2",N)+K2(INOUT2,"32")*FM("19",N)*TCin("17","2",N)+K2(INOUT2,"33")*FM("19",N)*TCin("18","2",N)+K2(INOUT2,"34")*FM("19",N)*TMin("25","2",N)+K2(INOUT2,"35")*FM("21",N)*TCin("17","2",N)+K2(INOUT2,"36")*FM("21",N)*TCin("18","2",N)+K2(INOUT2,"37")*FM("21",N)*TMin("25","2",N)+K2(INOUT2,"38")*FM("32",N)*TCin("17","2",N)+K2(INOUT2,"39")*FM("32",N)*TCin("18","2",N)+K2(INOUT2,"40")*FM("32",N)*TMin("25","2",N);REGRESSION3(INOUT3,N)..Fm(INOUT3,N)*TCout(INOUT3,"3",N)=E=K3(INOUT3,"1")+K3(INOUT3,"2")*FM("18",N)*FM("18",N)+K3(INOUT3,"3")*FM("25",N)*FM("25",N)+K3(INOUT3,"4")*FM("19",N)*FM("19",N)+K3(INOUT3,"5")*FM("21",N)*FM("21",N)+K3(INOUT3,"6")*FM("32",N)*FM("32",N)+K3(INOUT3,"7")*FM("18",N)*FM("19",N)+K3(INOUT3,"8")*FM("18",N)*FM("21",N)+K3(INOUT3,"9")*FM("18",N)*FM("32",N)+K3(INOUT3,"10")*FM("25",N)*FM("19",N)+K3(INOUT3,"11")*FM("25",N)*FM("21",N)+K3(INOUT3,"12")*FM("25",N)*FM("32",N)+K3(INOUT3,"13")*FM("18",N)*TCin("18","3",N)+K3(INOUT3,"14")*FM("25",N)*TCin("25","3",N)+K3(INOUT3,"15")*FM("19",N)*TMin("19","3",N)+K3(INOUT3,"16")*FM("21",N)*TMin("21","3",N)+K3(INOUT3,"17")*FM("32",N)*TMin("32","3",N)+K3(INOUT3,"18")*FM("18",N)*TMin("19","3",N)+K3(INOUT3,"19")*FM("18",N)*TMin("21","3",N)+K3(INOUT3,"20")*FM("18",N)*TMin("32","3",N)+K3(INOUT3,"21")*FM("25",N)*TMin("19","3",N)+K3(INOUT3,"22")*FM("25",N)*TMin("21","3",N)+K3(INOUT3,"23")*FM("25",N)*TMin("32","3",N)+K3(INOUT3,"24")*FM("19",N)*TCin("18","3",N)+K3(INOUT3,"25")*FM("19",N)*TCin("25","3",N)+K3(INOUT3,"26")*FM("21",N)*TCin("18","3",N)+K3(INOUT3,"27")*FM("21",N)*TCin("25","3",N)+K3(INOUT3,"28")*FM("32",N)*TCin("18","3",N)+K3(INOUT3,"29")*FM("32",N)*TCin("25","3",N);CONNECTIONS1(N)..TCout("17","1",N)=E=TCin("17","2",N);CONNECTIONS2(N)..TCout("18","1",N)=E=TCin("18","2",N);CONNECTIONS3(N)..TCin("19","1",N)=E=TCout("19","2",N);CONNECTIONS4(N)..TCin("21","1",N)=E=TCout("21","2",N);CONNECTIONS5(N)..TCin("19","2",N)=E=TCout("19","3",N);CONNECTIONS6(N)..TCin("21","2",N)=E=TCout("21","3",N);CONNECTIONS7(N)..TCin("32","2",N)=E=TCout("32","3",N);CONNECTIONS8(N)..TCout("18","2",N)=E=TCin("18","3",N);CONNECTIONS9(N)..TCout("25","2",N)=E=TCin("25","3",N);ModelMHEX/all/;*===ImportdatafromExcel*TMin$CALLGDXXRW.EXEktest.xlspar=TMin1rng=TMin!B3:I123ParameterTMin1(N,I);$GDXINktest.gdx$LOADTMin1$GDXIN$CALLGDXXRW.EXEktest.xlspar=TMin2rng=TMin!J3:Q123ParameterTMin2(N,I);$GDXINktest.gdx$LOADTMin2$GDXIN$CALLGDXXRW.EXEktest.xlspar=TMin3rng=TMin!R3:Y123ParameterTMin3(N,I);$GDXINktest.gdx$LOADTMin3$GDXIN*===SetTMin(I,B,N)asTMin1,2,3TMin(I,"1",N)=TMin1(N,I);TMin(I,"2",N)=TMin2(N,I);TMin(I,"3",N)=TMin3(N,I);*TMout$CALLGDXXRW.EXEktest.xlspar=TMout1rng=TMout!B2:I122ParameterTMout1(N,I);$GDXINktest.gdx$LOADTMout1$GDXIN$CALLGDXXRW.EXEktest.xlspar=TMout2rng=TMout!J2:Q122ParameterTMout2(N,I);$GDXINktest.gdx$LOADTMout2$GDXIN$CALLGDXXRW.EXEktest.xlspar=TMout3rng=TMout!R2:Y122ParameterTMout3(N,I);$GDXINktest.gdx$LOADTMout3$GDXIN*===SetTMout(I,B,N)asTMout1,2,3TMout(I,"1",N)=TMout1(N,I);TMout(I,"2",N)=TMout2(N,I);TMout(I,"3",N)=TMout3(N,I);*FM(I,B,N)$CALLGDXXRW.EXEktest.xlspar=FM1rng=FM!B1:I121ParameterFM1(N,I);$GDXINktest.gdx$LOADFM1$GDXINFM(I,N)=FM1(N,I);*K1,K2,K3$CALLGDXXRW.EXEktest.xlspar=K10rng=K!B1:AE6ParameterK10(INOUT1,M1);$GDXINktest.gdx$LOADK10$GDXINK1.l(INOUT1,M1)=K10(INOUT1,M1);$CALLGDXXRW.EXEktest.xlspar=K20rng=K!B7:AP13ParameterK20(INOUT2,M2);$GDXINktest.gdx$LOADK20$GDXINK2.l(INOUT2,M2)=K20(INOUT2,M2);$CALLGDXXRW.EXEktest.xlspar=K30rng=K!B14:AE19ParameterK30(INOUT3,M3);$GDXINktest.gdx$LOADK30$GDXINK3.l(INOUT3,M3)=K30(INOUT3,M3);*optionsNLP=IPOPT;optionsreslim=3600;solveMHEXusingNLPminimizingz;displayK1.l,K2.l,K3.l*savesolutionexecute_unload"k.gdx"K1.lK2.lK3.lexecute'gdxxrw.exek.gdxvar=K1.lrng=B1:AE6'execute'gdxxrw.exek.gdxvar=K2.lrng=B7:AP13'execute'gdxxrw.exek.gdxvar=K3.lrng=B14:AE19'第五章總結(jié)與展望5.1總結(jié)在實(shí)際操作中尋找多股流換熱器的最佳工作點(diǎn)是空分裝置節(jié)能的重要手段，也是過程實(shí)時(shí)優(yōu)化的研究方向。本文以20000Nm3/h內(nèi)壓縮空分裝置的MHEX為代表，對(duì)結(jié)構(gòu)復(fù)雜，換熱溫差小，耦合程度高的MHEX進(jìn)行了建模、求解和傳熱性能研究，并對(duì)AspenPlus平臺(tái)上的空分流程模擬進(jìn)行了補(bǔ)充完善。本文的研究工作和成果包括：1）本文結(jié)合傳熱機(jī)理和數(shù)據(jù)回歸模型給出模型自變量的選取規(guī)則，基于數(shù)據(jù)回歸方法，建立MHEX的操作數(shù)據(jù)模型。模型利用進(jìn)出口溫度、和流量回歸出口溫度和進(jìn)口溫度、流量的關(guān)系，得到一組回歸參數(shù)。用于模擬的模型只有進(jìn)出變量，沒有中間變量，是輸入變量的顯函數(shù)。這模型求解出口溫度時(shí)計(jì)算速度開，收斂性好，計(jì)算得到的出口溫度與實(shí)際出口溫度相對(duì)誤差小于1%，主換熱器平均絕對(duì)誤差小于0.35℃，過冷器平均絕對(duì)誤差小于0.05℃。2）本文采用的機(jī)理分析從幾種評(píng)價(jià)方式中選取復(fù)熱不足冷損做MHEX傳熱性能的表征方式，從數(shù)據(jù)角度驗(yàn)證了選擇的正確性。關(guān)于原料空氣量做一維標(biāo)度化分析，復(fù)熱不足冷損比冷量回收率關(guān)于空氣總流量有更為明顯的靈敏度關(guān)系。采用MHEX模型和PCA做高維變量的數(shù)據(jù)處理，PCA通過特征提取，將多股流換熱器的高維變量轉(zhuǎn)化成相互獨(dú)立的低維變量，實(shí)現(xiàn)變量降維的目的。PCA結(jié)果表明主元一和主元二包含了原變量94%的信息。空分裝置現(xiàn)場(chǎng)控制時(shí)，復(fù)熱不足冷損在工程上直觀表現(xiàn)為熱端溫差，MHEX的相關(guān)變量應(yīng)盡量控制在主元一較小的區(qū)域。3）本文采用Fortran語(yǔ)言編譯了MHEX用戶自定義模型（Custommodular），實(shí)現(xiàn)了AspenPlus平臺(tái)的MHEX單元操作模擬。然后替換內(nèi)壓縮空分流程模擬中的MHeatX，通過靈敏度分析調(diào)和誤差大的測(cè)量數(shù)據(jù)，實(shí)現(xiàn)全流程的操作模擬，主要參數(shù)的模擬誤差基本小于5%，為空分變負(fù)荷優(yōu)化奠定了先決條件。4）