Model predictive control for ethanol steam reformers with membrane separation

Model P edic i e Con ol o E hanol S eam Re o me s wi h Memb ane Sepa a ion Ma ia Se a, Ca los Ocampo-Ma inez, Mingming Li and Jo di Llo ca Ins i u de Rob` o ica i In o m` a ica Indus ial (CSIC - UPC) Uni e si a Poli ` ecnica de Ca alunya Llo ens i A igas 4-6, 08028 Ba celona, Spain Ins i u de T` ecniques Ene g` e iques Uni e si a Poli ` ecnica de Ca alunya Diagonal 647, Ed. ETSEIB, 08028 Ba celona, Spain Abs ac This pape ocuses on he dynamic modelling and he p edic i e con ol o an e hanol s eam e o me (ESR) wi h Pd-Ag memb ane sepa a ion s age o he gen- e a ion o pu e hyd ogen. Hyd ogen pu i y necessa y o eed a p o on exchange memb ane uel cell (PEMFC) is equi ed. A non-linea dynamic model o he ESR is de eloped oge he wi h a p ocedu e o adjus ing he model pa ame e s in o de o i a bank o expe imen al da a o a eal ESR sys em. S a ic and dy- namic analysis o he non-linea ESR model is p esen ed. F om his non-linea model, a linea , educed o de and disc e ised model is de i ed and a model p e- dic i e con olle (LMPC) is designed o he ESR sys em. Con ol objec i es a e pu e hyd ogen low a e acking and e hanol inle minimiza ion. Compa isons be- ween he non-linea and linea models a e ca ied ou o de e mine he con ol cons ain s. Finally, simula ion esul s o he implemen ed LMPC con olle a e p esen ed and discussed. E hanol s eam e o me , s aged-sepa a ion memb ane eac o , con ol-o ien ed model, linea model p edic i e con ol 1 In oduc ion Nowadays hyd ogen has eme ged as an in e es ing ene gy ec o which may be used o s o e ene gy om enewable sou ces. Th ough uel cells his hyd ogen may be con- e ed in o elec ici y when necessa y a a high e iciency. Fo ins ance, uel cell e- hicles ha e e iciencies mo e han wo imes highe han hose o combus ion engines. The mos ma u e uel cell echnology, wi h a wide ange o applica ions, is PEMFC 1 [1]. Du abili y, eliabili y and e iciency o PEMFC a e signi ican ly imp o ed when pu e hyd ogen is supplied o hem. Due o i s physical p ope ies, hyd ogen is di icul o anspo and s o e, which makes he echniques o in si u hyd ogen p oduc ion inc easingly in e es ing o many applica ions. Hyd ogen can be ob ained om many di e en sou ces such as wa e , e hanol, na u al gas o o he ossil uels. Among hese sou ces, e hanol has been chosen o gene a ing hyd ogen in he las decades o se e al ad an ages: high con en o hyd ogen, easy po abili y and s o abili y, enewable na u e and low oxici y. Nowadays, he e a e h ee common echniques o hyd ogen p oduc ion om e hanol: s eam e o ming (SR), pa ial oxida ion (POX) and au o- he mal e o ming (ATR). The bes o hese op ions is SR because o i s lowe ope a ing empe a u e and highe hy- d ogen yield. Howe e , his echnique gene a es he highes emission o CO, which can poison he anode ca alys o PEMFCs i he quan i y is mo e han 100 pa s pe million (ppm). I high empe a u e PEMFC a e used, he ole ance o CO is highe , as explained in [2], whe e he in eg a ion o an e hanol s eam e o me o a high empe - a u e PEMFC is s udied. In any case, hyd ogen pu i ica ion is essen ial be o e eeding PEMFCs. Con en ional me hods o pu e hyd ogen gene a ion include wa e gas shi (WGS), CO p e e en ial oxida ion (COP Ox), and p essu e-swing adso p ion (PSA), among o he s. Addi ionally, memb ane sepa a ion echniques ha e al eady been applied o ob ain pu e hyd ogen om he gas mix u e [3, 4, 5, 6]. Especially, palladium-based memb anes ha e gained g ea in e es wi h mo e han 6000 scien i ic a icles [7] in- ol ed since Juenke e al. [8] analysed he use o palladium memb anes o hyd ogen pu i ica ion in 1955. The ema kable p og ess achie ed in he ield o palladium-based memb ane eac o s (MRs) is due o hei comple e hyd ogen pe m-selec i i y wi h e- spec o all o he gases [7]. Fu he mo e, only one hea ing uni is equi ed o bo h he e o me s age and sepa a ion s age. In [9], an exe ge ic s udy o an e hanol s eam e o me wi h a palladium-based memb ane has been p esen ed and he bes ope a ing condi ions ega ding he exe ge ic and he mal e iciency ha e been de e mined. The ca alys plays an impo an ole in he e hanol s eam e o ming (ESR) p o- cess, since i can accele a e he eac ion a e and imp o e he pe o mance o he ESR. Cobal -based ca alys s ha e been conside ed o be a sui able choice o hei low cos , high ac i i y and selec i i y [10, 11, 12], especially o he ESR o p oduce hyd ogen a mode a e empe a u e [11, 13, 14]. Ca alys s se up is also a ele an issue. In [15], ESR o hyd ogen gene a ion o e s uc u ed ca alys s is desc ibed based on di e en expe imen al es s. Dynamic modelling o ESRs wi h memb ane sepa a ion is e y limi ed. In [16], he way o ob ain a con ol-o ien ed model om a dynamic simula ion-o ien ed model o an ESR wi hou memb ane sepa a ion s age is p esen ed, while [5] p esen s he model o a s aged-sepa a ion memb ane eac o o s eam e o ming o me hane. Only [17] p esen s he modelling o he ESR wi h memb ane sepa a ion o he gene a ion o pu e hyd ogen. In he pape , he ESR model is ob ained acco ding o he phenomenological kine ic model wi h powe law [18] and expe imen al da a. On he o he hand, so a , he e ha e been e y ew wo ks in ol ed in o designing con olle s o ESR. In spi e o he non-linea ESR na u e, linea con olle s a e designed in wo wo ks [19, 20] since hey can yield a sa is ac o y pe o mance i he p ocess is ope a ed close o a nominal 2 ope a ing poin . Speci ically, in [20], a linea model p edic i e con olle (LMPC) is designed o he ESR wi hou memb ane sepa a ion. Model p edic i e con ol (MPC) is an ad anced s a egy o p ocess con ol ha has been widely used in indus y and chemical p ocesses since he 1980s[21, 22]. The MPC s a egy is a se o con ol me hodologies ha use a ma hema ical model o a conside ed sys em o ob ain con ol ac ions h ough minimizing a cos unc ion ela ed o selec ed con ol objec i es conside ing he desi ed sys em pe o mance. MPC has p esen ed ob ious ad an ages o e o he me hods [21]: mul i-objec i es easy o deal wi h he mul i- a iable sys em, eed- o wa d con ol been used in a na u- al way o compensa e measu able dis u bances and bene icial o acking u u e e - e ences. Howe e , he s a egy has also i s own d awbacks: i s uning, which is no a gene al heo y ha can be implemen ed in a gene ic way, and he high compu a- ional ime de i ed om he numbe o a iables and cons ain s, mainly in la ge-scale sys em. In [20], an LMPC scheme has been p esen ed o an ESR p ocess wi hou memb ane sepa a ion. The main con ibu ion o his pape is o ob ain a dynamic model o an ESR wi h memb ane sepa a ion and o use his model o design a LMPC able o ope a e he sys em p ope ly a ound a nominal ope a ing poin . The eminde o his pape is o ganized as ollows: Sec ion 2 gi es a b ie de- sc ip ion o he ESR sys em wi h memb ane sepa a ion s udied in his wo k. Sec ion 3 i s ly p esen s he non-linea ma hema ical model o he ESR and secondly p esen s he con ol-o ien ed model o he MPC con olle design. Sec ion 4 p esen s he MPC design o he ESR, including he o mula ion o he co esponding op imiza ion p ob- lem and he LMPC con olle con igu a ion wi h ou pu eedback. Sec ion 5 shows and discusses he main simula ion esul s. Finally, he main conclusions a e d awn in Sec ion 6. 2 Sys em desc ip ion 2.1 Expe imen al se up Fo he pu pose o simpli ying he sys em se up, he p ocess o gene a ing hyd ogen ia ESR and pu i ying hyd ogen using selec i e memb ane may be adop ed in a single eac ion and sepa a ion module, named S aged-Sepa a ion Memb ane Reac o (SSMR) [17]. In his pape , he ESR was conduc ed in o a SSMR using a cobal -based ca alys o e co die i e monoli hs ha we e implemen ed in se ies in o a s ainless-s eel Pd-Ag memb ane s age [23]. A scheme o he SSMR is shown in Figu e 1. The SSMR mod- elled in his wo k co esponds o a eal labo a o y sys em wi h he ollowing cha ac e - is ics: he eac ion-sepa a ion chambe measu es 230 mm high and 22 mm o ou side diame e and has a lowe head o allow he exi o he sepa a ed s eams, which a e e en a e and pe mea e s eams. A eed e apo a ion condui is used o e apo a e bo h e hanol and wa e be o e en e ing he eac o . Fi e ca aly ic honeycomb pieces o 2 cm leng h each a e disposed in se ies in o he eac o ollowed by a Pd-Ag memb ane ube. The ca aly ic honeycomb pieces a e loaded wi h a o al o 1.32 g o cobal -based ca alys . The pine-hole ee and dead-end memb ane ube measu es 76 mm high, 1/8 3 inch diame e and a o al a ea o 7.1 cm2. The Pd-Ag ac i e laye is 30 µm hick o e a po ous s ainless s eel suppo . The liquid mix u e o he e hanol and wa e a e ed di ec ly om he s o age ank by a high-pe o mance liquid ch oma og aphy (HPLC) pump. The e en a e p essu e is adjus ed and con olled by a manually-ope a ed back- p essu e egula o and he pe mea e p essu e is main ained a a mosphe ic p essu e. Figu e 1: S aged-sepa a ion memb ane eac o The expe imen al da a used in his wo k was ob ained a di e en p essu es in he ange 1 ba –14 ba , di e en empe a u es in he ange 500°C–600°Cand di e - en s eam o ca bon (S/C) a ios. The s eam o ca bon a io is de ined as he numbe o wa e molecules di ided by he numbe o ca bon a oms, and i is applied in his case o he inle low a e mix u e o bo h e hanol and wa e . Also, wo se s o expe imen s we e done in o de o cha ac e ise he beha iou o he sys em: expe imen s wi hou he sepa a ion s age (wi hou memb ane) and expe imen s wi h he sepa a ion s age (wi h memb ane). A ull-con e sion a e (100%) o e hanol and ace aldehyde we e measu ed in he e o me s age. 4 Figu e 2: Scheme o he ESR 2.2 Chemical eac ion The ESR eac ions o e cobal -based ca alys s a e exp essed as ollows [12, 24, 25]: C2H5OH −−→ CH3CHO + H2,(1a) C2H5OH −−→ CO + CH4+ H2,(1b) CO + H2O−−* )−− CO2+ H2,(1c) CH3CHO + 3 H2O−−→ 2 CO2+ 5 H2.(1d) These ou eac ions a e aking place in he same space and condi ions simul ane- ously. Fi s ly, e hanol dehyd ogena es in o hyd ogen and ace aldehyde (1a), which is u he e o med wi h wa e o ca bon dioxide (1d). In addi ion, cobal ca alys s a e ac i e o he Wa e Gas Shi (WGS) eac ion (1c) unde ypical ope a ing condi ions. The undesi ed eac ion is he e hanol decomposi ion o p oduce ca bon monoxide and me hane (1b). In he memb ane sepa a ion s age, he Pd-Ag memb ane pe mea es only he hyd ogen and lea es he was e gases in he e en a e side [26, 27]. Figu e 2 shows a scheme o he p ocess wi h he wo s ages in se ies. 2.3 Ene gy s udy An ene gy s udy has been done in o de o quan i y he global ene gy balance o he SSMR desc ibed in Sec ion 2.1 a s eady s a e and o some speci ic ope a ing condi- ions. In pa icula , 8 ba o p essu e, 813.15 K (540¨ ı¿½ C) o empe a u e, an inle wa- e low o 0.0108 mol/min and an inle e hanol low o 0.0018 mol/min a e conside ed. The ela ionship be ween hese wa e and e hanol inle lows is 6 o 1, o equi alen ly S/C =3. The ene gy supplied is calcula ed aking in o accoun he powe necessa y o apo ize and hea he eac an s o 813.15 K, he ene gy necessa y o p essu ize he gases and he hea consumed by he eac ion. The ene gy o he hyd ogen p essu isa- ion is no conside ed. The supplied ene gy esul s in 14.78 Wa s. On he o he hand, o he selec ed ope a ing poin , 3.4 mols o hyd ogen a e ob ained om each mole o e hanol in he e o ming s age and he e o e, he ou le hyd ogen low a e o m he e o ming s age is 6.2 ×10−3mol/min. This low a e could be ans o med in o 29.5 Wa s o hea i combus ed (High Hea ing Value used o he calcula ion). Conside ing he sepa a ion s age as well o he same ope a ing condi ions, he hyd ogen low a e in he e en a e side is 4.65 ×10−3mol/min and he pu e hyd ogen low a e in he pe me- a e side is 1.55 ×10−3mol/min. These low a es could be ans o med in o 22.15 Wa s 5 o hea and 3.57 Wa s o elec ici y, espec i ely, conside ing ha he pu e hyd ogen eeds a PEM uel cell wi h a s oichiome y o 1.1 and 0.8 Vol s. This numbe s, co e- sponding o he expe imen al se up desc ibed in Sec ion 2.1, could easily be imp o ed i imp o ed designs we e conside ed. 3 Ma hema ical modelling 3.1 Main assump ions In his pape , wo plug- low kind eac o s a e assumed o ep esen he dynamic beha - io o he ESR wi h memb ane sepa a ion. The i s eac o is he e o me s age, and he second is he memb ane sepa a ion s age. The main modelling assump ions a e: • Iso he mal ope a ion condi ions wi hin each s age • Isoba ic ope a ion condi ions assumed due o he high oid ac ion o monoli hic s uc u es • Comple ely mixed luid inside he plug- low eac o (PFR) in any c oss-sec ion a any posi ion • Neglec ed di usion e ms due o he p edominance o con ec ion ac ion o e di usion ac ion • Ideal gases due o he low ope a ing p essu e • Fluid eloci y depending only on he axial posi ion • 1D model due o he small diame e o he eac o 3.2 Mass balance o he e o me s age The mole balance equa ion oge he wi h he ini ial and bounda y condi ions o he plug low eac o model a e exp essed as [16] ∂Cj ∂ + Cj ∂υ ∂z +υ∂Cj ∂z =X i νj,i i,(2a) Cj(0, z)=Cj,0(z),∀z∈[in, ou ](2b) Cj( , 0) = Cj,in ( ),∀ > 0(2c) i= 1,...,4,(2d) j= 1,...,7,(2e) whe e jdeno es he componen , which could be C2H5OH, H2O, CH4, CO, CO2, CH3CHO o H2and ideno es he eac ion acco ding o (1a)-(1d). Mo o e , Cjis he concen a ion o he j- h componen , iis he eac ion a e o eac ion i,υis he linea eloci y o he gases, in is he se o inle condi ions while ou is he se o ou le 6 condi ions, νj,i is he s oichiome ic coe icien o componen jin eac ion iand zis he axial posi ion a iable. In o de o sol e he se o pa ial di e en ial equa ions (PDEs) p esen ed in (2a), υand a e exp essed as unc ions o Cj. In he PFR wi h isoba ic and iso he mal con- di ions, i is assumed ha υ a ies wi h ime and posi ion wi hin he eac o acco ding o he ollowing exp essions [16]: υ=υin (1 + εX),(3a) X=1−Cj Cj,in 1 + Cj Cj,in ,(3b) whe e Xis he con e sion a e o he e hanol while εis he mola ela ion o he conside ed eac ion. Mo eo e , he olume ic low a e (Q) and mola low a e o he di e en componen s (Fi), which a e assumed measu ed a iables o he con ol p ob- lem, a e compu ed as Q=A1υ, (4a) Fj=QCj,(4b) whe e A1is he sec ion a ea o he ubula eac o . Reac ion a es depend on concen- a ion, empe a u e and p essu e. Acco ding o [18], he ollowing phenomenological kine ics ha e been conside ed o he ou modeled eac ions: 1=k1(PC2H5OH)m(P),(5a) 2=k2PC2H5OH,(5b) 3=k3PCOPH2O−PCO2PH2 kW GS ,(5c) 4=k4PCH3CHOPH2O3,(5d) ki=k∞,iexp −Ea,i 1 RT −1 RT e ,(5e) kW GS =exp 4577.8 T−4.33,(5 ) m(P)=1.2+0.23(P−4),(5g) whe e k∞,i a e he p e-exponen ial ac o s, ki he kine ic cons an s and Ea,i he ac i- a ion ene gies o each eac ion. The e e ence empe a u e T e has been selec ed o be 873.15 K, which is he highes empe a u e o he expe imen . Besides, m(P)is an exponen ial ac o depending on p essu e and PC2H5OH,PCO,PH2O,PCO2,PCH3CHO a e he pa ial p essu es o each componen . 3.3 Mass ans e o he memb ane sepa a ion s age In his s age, he p ocess o he hyd ogen selec i e sepa a ion wi hou chemical eac- ions akes place. A Pd-Ag me allic memb ane is used o pe mea e only he hyd ogen 7 lea ing he es o he gas on he e en a e side. The mass ans e mechanism can be exp essed using he Sie e s’ law as [4] JH2=Pe δA2pPH2 e en a e −pPH2pe mea e,(6) wi h Pe=Pe0·exp −Ea RT ,(7) whe e Peco esponds wi h he gas pe meabili y, Pe0is he p e-exponen ial ac o , Ris he ideal gas cons an , Tis empe a u e, PH2 e en a e is he hyd ogen pa ial p essu e in he e en a e side, PH2pe mea e is he hyd ogen pa ial p essu e in he pe mea e side, Eais he appa en ac i a ion ene gy, A2is he su ace a ea o he memb ane and JH2 is he pe mea ing hyd ogen lux. 3.4 Spa ial disc e iza ion Each s age is di ided in o 20 slices o smalle size and each slice is conside ed as a con inuous s i ed ank eac o wi h homogeneous condi ions. Backwa d ini e di e - ences a e applied o ans o m he PDEs in o o dina y di e en ial equa ions (ODEs), which a e exp essed as ∂Cj ∂z ∼ =Cj(z)−Cj(z−1) 4z.(8) 3.5 Pa ame e s adjus men The model pa ame e s ha e been adjus ed o sa is y a se o s a ic expe imen al da a. Speci ically, he mola low a es o he di e en componen s Fiwe e collec ed. Pa- ame e s o he e o me s age ha e been adjus ed using expe imen al da a ob ained wi hou he sepa a ion s age. Once hese pa ame e s ha e been ixed, he pa ame e s o he sepa a ion s age ha e been adjus ed using expe imen al da a ob ained wi h he whole ( e o me s age plus sepa a ion s age) sys em. The p e-exponen ial ac o s and ac i a ion ene gies in (5) we e adjus ed in o de ha all he e hanol and ace aldehyde a e comple ely ans o med and he ou le mola low a es o he es o componen s a e close o he expe imen al da a. Speci ically, he ollowing s eps we e ollowed o i he eal da a: • Fi s o all, adjus k∞,2and Ea,2 o i he quan i y o CH4, which is only p o- duced in eac ion (1b) • Then, adjus k∞,1and Ea,1 o ensu e ha he whole con e sion o e hanol is 100% • Then, adjus k∞,4and Ea,4 o ensu e ha he con e sion o ace aldehyde is 100% • Finally, adjus k∞,3and Ea,3acco ding o he quan i y o H2 om he expe i- men al da a. 8 Fo each k∞,i-Ea,i pai , k∞,i was i s adjus ed a T=T e and hen Ea,i was adjus ed using expe imen al da a a di e en empe a u es. The alues o all he ob ained pa- ame e s a e shown in he Appendix. On he o he hand, i m(P)= 1, he kine ic model in (5) only i s o p essu e a 1 ba . In o de ha i is sui able o p essu es be ween 1 ba o 14 ba , a powe -law exp ession is equi ed. The e o e, an exponen ial numbe m(P)in (5a) was added. Equa ion (5g) shows he ela ionship be ween he cons an m(P)and he p essu e. The accu acy o he model wi h espec o he expe imen al da a is assessed h ough he e o s shown in Figu e 3. I can be seen ha he ag eemen be ween he expe imen al and simula ed da a is sa is ac o y. Finally, pa ame e s Pe0 and Eain (7) we e also adjus ed acco ding o he eal da a o pu e hyd ogen ob ained in he expe imen s. Exac alues o bo h pa ame e s a e also gi en in he Appendix. Figu e 3: Model da a e sus expe imen al da a. Red line shows a 1:1 co espondence 3.6 S a ic analysis o he non-linea model The s a ic and dynamic beha iou o he non-linea ESR model has been analysed in open loop. This in o ma ion plays a signi ican ole in he MPC con olle design ac ing as a guidance o he manual uning o he MPC con olle . The s a ic esul s a e ob ained by sol ing he di e en ial algeb aic equa ions in (2) un il all he de i a i es wi h espec o ime a e equal o ze o. S a ic low a e p o iles o each componen wi h espec o he posi ion in he axial di ec ion o he eac o a e shown in Figu es 4(a) and 4(b) o he i s and second s age o he eac o , espec i ely. 9 Figu e 6: LMPC-based closed-loop con igu a ion 5 Simula ion esul s Some pa ame e s in ol ed in he MPC con olle can be uned o mee he con ol ob- jec i es, which a e he p edic ion ho izon (HP), he con ol ho izon (HC), and he weigh s o each con ol objec i e (Mand N). In his wo k, HChas been gi en he same alue as HP. E en ually, HP= 50 (25 minu es), which ensu es an adequa e ho izon o he closed-loop p edic ion. Rega ding Mand N, wo op ions a e imple- men ed in o de ha bo h con ol objec i es can be weigh ed di e en ly. The i s op ion is o assign he same weighs o Mand Nin o de ha bo h con ol objec i es ha e he same p io i y. Figu e 7 shows he esul s o his con olle uning case wi h se poin changes o ±10%. In o de o sa is y bo h con ol objec i es a he same ime, he con olle sea ches he op imal inpu s o wa e and p essu e. Howe e , i can be seen ha , since e hanol minimiza ion is gi en he same weigh as hyd ogen acking, he hyd ogen is always less han i s e e ence wi hin he simula ion ime (see Figu e 7(b)). This shows ha he capaci y o mee bo h con ol objec i es h ough he manipula ion o wa e and p essu e is limi ed. The e o e, i we conside ha he p incipal con ol objec i e is o ack he e e ence o hyd ogen, he weigh ing op ion should be changed. Conside ing he hyd ogen low a e se poin acking as he main objec i e, i is nec- essa y o assign much mo e weigh on M han N. Figu e 8 shows he simula ion esul s ela ed o his uning case. In pa icula , Figu e 8(a) shows he con ol ac ions applied o he eal ESR sys em. The empe a u e is always main ained a i s nominal alue because i is no conside ed a manipula ed inpu . Figu e 8(b) shows he pu e hyd ogen low a e and i s e e ence. The se poin is changed ±10% a ime 10 minu es, 25 minu es and 40 minu es. The pu e hyd ogen p oduced p ope ly ollows i s e e ence. Fu he mo e, con ol ac ions change qui e smoo hly du ing he se poin acking in 16 o de o ind he new s eady s a e. The e o e, conside ing di e en dynamic and s a ic aspec s, i can be concluded ha he con ol objec i e has been success ully achie ed. 6 Conclusions This pape p esen s an LMPC con olle o he con ol o an e hanol s eam e o me wi h memb ane sepa a ion. The wo k is based on a non-linea dis ibu ed model whose pa ame e s ha e been adjus ed using expe imen al da a. Howe e , he LMPC con- olle is designed based on a linea , o de educed and disc e ised model ha has been de i ed om he non-linea model. A compa ison be ween he non-linea and linea models shows ha he linea model is capable o ep esen ing he impo an dynamics o he ESR in a neighbou hood o he selec ed nominal ope a ing poin . Finally, he designed LMPC con olle has been shown o be app op ia e o se poin acking o hyd ogen p oduc ion o posi i e and nega i e changes up o 10%. Also, he LMPC has shown i s capabili y o op ionally ake in o accoun he minimiza ion o e hanol in- low h ough he uning o some weigh ing pa ame e s, al hough o physical easons, he minimiza ion o he e hanol in low implies limi a ions in he achie able hyd ogen p oduc ion. Fu u e wo k will add ess he analysis o he sys em a di e en ope a ing condi ions and he conside a ion o non linea MPC con olle s. Also, i will add ess he mal model inclusion, unce ain y conside a ion as well as se up design imp o e- men . Acknowledgemen This pape has pa ially suppo ed by he Spanish esea ch p ojec MICAPEM ( e . DPI2015-69286-C3-2-R, MINECO/FEDER). Mingming Li has been suppo ed by he China Schola ship Council (CSC). Jo di Llo ca is Se a H´ un e Fellow and is g a e ul o ICREA Academia p og am. Ac onyms ATR Au o- he mal Re o ming ESR E hanol S eam Re o me HPLC High-pe o mance Liquid Ch oma og aphy LMPC Linea Model P edic i e Con ol LRDsys Linea Reduced Disc e e Model LRsys Linea Model wi h O de Reduc ion Lsys Linea Full-o de Con inuous Model MI Manipula ed Inpu s MO Measu ed Ou pu s MPC Model P edic i e Con ol MRs Memb ane Reac o s NLsys Non-linea Con inuous Model 17 NMPC Non-linea Model P edic i e Con ol ODEs O dina y Di e en ial Equa ions PDEs Pa ial Di e en ial Equa ions PEMFC Polyme Elec oly e Memb ane Fuel Cell PFR Plug Flow Reac o PID P opo ion In eg a ion Di e en ia ion POX Pa ial Oxida ion ppm Pa s pe million PSA P essu e-swing Adso p ion SR S eam Re o ming WGS Wa e Gas Shi Nomencla u e V1Volume, m3 A1Sec ion a ea o eac o , m2 εMola ela ion, dimensionless V2Volume, m3 δThickness o Pd-Ag memb ane, m A2Su ace a ea o Pd-Ag memb ane, m2 DDiame e o Pd-Ag memb ane, m RIdeal gas cons an , J mol−1K−1 TTempe a u e inside he eac o , K T e Tempe a u e e e ence, K υLinea eloci y o gases, m min−1 Reac ion a es, mol m−3min−1 νS oichiome ic coe icien , dimensionless CConcen a ion, mol/m−3 QVolume ic low a e, m3min−1 FMola low a e, mol min−1 PH2 e en a e Hyd ogen p essu e in e en a e side, Pa PH2pe mea e Hyd ogen p essu e in pe mea e side, Pa PeGas pe meabili y, mol m−1min−1Pa−0.5 JH2Flux o pe mea ing hyd ogen, mol min−1 PP essu e in he e o me s age, ba PC2H5OH Pa ial p essu e o C2H5OH, Pa PH2OPa ial p essu e o H2O, Pa PCO Pa ial p essu e o CO, Pa PCO2Pa ial p essu e o CO2, Pa PCH3CHO Pa ial p essu e o CH3CHO, Pa in Reac o inle ou Reac o ou le TsSampling ime, minu es 18 HPP edic ion ho izon MWeigh ing ma ix (con ol uning) NWeigh ing ma ix (con ol uning) umin Minimum alues o manipula ed inpu s umax Maximum alues o manipula ed inpu s ymin Minimum alues o measu ed ou pu s ymax Maximum alues o measu ed ou pu s Appendix: Adjus ed pa ame e s Re o me s age Ac i a ion ene gies o each eac ion: Ea1= 7.0×103 J mol−1 Ea2=1.3×104J mol−1 Ea3=7.0×103J mol−1 Ea4=8.9×103J mol−1 P e-exponen ial ac o s o each eac ion: k∞1=5.025 ×103 mol m−3min−1ba −1 k∞2=4.788 ×104 mol m−3min−1ba −1 k∞3=4.548 ×103 mol m−3min−1ba −2 k∞4=4.788 ×104 mol m−3min−1ba −4 Memb ane sepa a ion s age Ea=8.8 ×103J mol−1 Pe0=9.0 ×10−7mol m−3min−1Pa−0.5 Re e ences Re e ences [1] O. Z. Sha a , M. F. O han, An o e iew o uel cell echnology: Fundamen als and applica ions, Renewable and Sus ainable Ene gy Re iews 32 (0) (2014) 810– 853. 19 [2] D. Geo ge, P. Su esh, De ailed analysis o in eg a ed s eam e hanol e o me and high empe a u e polyme elec oly e memb ane uel cell, In e na ional Jou nal o Hyd ogen Ene gy 41 (2) (2016) 1248–1258. [3] E. McLea y, J. Jansen, F. Kap eijn, Zeoli e based ilms, memb anes and mem- b ane eac o s: P og ess and p ospec s, Mic opo ous and Mesopo ous Ma e ials 90 (1-3) (2006) 198–220. [4] R. Koch, E. L´ opez, N. J. Di ins, M. Allu´ e, A. Jossen, J. Rie a, J. Llo ca, E hanol ca aly ic memb ane e o me o di ec PEM FC eeding, In e na ional Jou nal o Hyd ogen Ene gy 38 (14) (2013) 5605–5615. [5] A. Li, C. J. Lim, J. R. G ace, S aged-sepa a ion memb ane eac o o s eam me hane e o ming, Chemical Enginee ing Jou nal 138 (1-3) (2008) 452–459. [6] S. Tos i, M. Fabb icino, A. Mo iani, G. Aga iello, C. Scudie i, F. Bo gognoni, A. San ucci, P essu e e ec in e hanol s eam e o ming ia dense pd-based mem- b anes, Jou nal o Memb ane Science (2011) 65–74. [7] A. Basile, A. Iulianelli, T. Longo, S. Liguo i, M. De Falco, Pd-based selec i e memb ane s a e-o - he-a , in: M. De Falco, L. Ma elli, G. Iaquaniello (Eds.), Memb ane Reac o s o Hyd ogen P oduc ion P ocesses, Sp inge London, 2011, pp. 21–55. [8] D. W. Juenke , M. an Swaay, C. E. Bi chenall, On he Use o Palladium Di - usion Memb anes o he Pu i ica ion o Hyd ogen, Re iew o Scien i ic Ins u- men s 26 (1955) 888. [9] A. Hedaya i, O. Le Co e, B. Laca ie e, J. Llo ca, Exe ge ic s udy o ca aly ic s eam e o ming o bio-e hanol o e Pd-Rh/CeO2 wi h hyd ogen pu i ica ion in a memb ane eac o , In e na ional Jou nal o Hyd ogen Ene gy 40 (8) (2015) 3574– 3581. [10] M. Dom´ ınguez, G. C is iano, E. L´ opez, J. Llo ca, E hanol s eam e o ming o e cobal alc in a pla e mic o eac o , Chemical Enginee ing Jou nal 176-177 (0) (2011) 280–285. [11] M. Dom´ ınguez, E. Taboada, E. Molins, J. Llo ca, E hanol s eam e o ming a e y low empe a u e o e cobal alc in a memb ane eac o , Ca alysis Today 193 (1) (2012) 101–106. [12] S. Losse, J. G. Vos, S. Rau, Ca aly ic hyd ogen p oduc ion a cobal cen es, Co- o dina ion Chemis y Re iews 254 (21) (2010) 2492–2504. [13] J. D. Holladay, J. Hu, D. L. King, Y. Wang, An o e iew o hyd ogen p oduc ion echnologies, Ca alysis Today 139 (4) (2009) 244–260. [14] M. Ni, D. Y. C. Leung, M. K. H. Leung, A e iew on e o ming bio-e hanol o hyd ogen p oduc ion, In e na ional Jou nal o Hyd ogen Ene gy 32 (15) (2007) 3238–3247. 20 [15] E. Lopez, N. J. Di ins, A. Anzola, S. Schbib, D. Bo io, J. Llo ca, E hanol s eam e o ming o hyd ogen gene a ion o e s uc u ed ca alys s, In e na ional Jou nal o Hyd ogen Ene gy 38 (11) (2013) 4418–4428. [16] V. M. Ga c´ ıa, E. L´ opez, M. Se a, J. Llo ca, Dynamic modeling o a h ee- s age low- empe a u e e hanol e o me o uel cell applica ion, Jou nal o Powe Sou ces 192 (1) (2009) 208 – 215. [17] E. L´ opez, N. J. Di ins, J. Llo ca, Hyd ogen p oduc ion om e hanol o e Pd Rh/CeO2wi h a me allic memb ane eac o , Ca alysis Today 193 (1) (2012) 145– 150. [18] I. U iz, G. A zamendi, E. L´ opez, J. Llo ca, L. M. Gand´ ıa, Compu a ional luid dynamics simula ion o e hanol s eam e o ming in ca aly ic wall mic ochannels, Chemical Enginee ing Jou nal 167 (3) (2011) 603–609. [19] V. M. Ga cia, M. Se a, J. Llo ca, J. Rie a, Design o linea con olle s applied o an e hanol s eam e o me o pem uel cell applica ions, In e na ional Jou nal o Hyd ogen Ene gy 38 (18) (2013) 7640–7646. [20] D. Recio-Ga ido, C. Ocampo-Ma ´ ınez, M. Se a-P a , Design o op imiza ion- based con olle s applied o an e hanol s eam e o me o hyd ogen p oduc ion, In e na ional Jou nal o Hyd ogen Ene gy 37 (15) (2012) 11141–11156. [21] J. M. Maciejowski, P edic i e con ol wi h cons ain s, P en ice Hall, 2002. [22] P. Rullo, L. N. Degliuomini, M. Ga c´ ıa, M. Basualdo, Model p edic i e con ol o ensu e high quali y hyd ogen p oduc ion o uel cells, In e na ional Jou nal o Hyd ogen Ene gy 39 (16) (2014) 8635–8649. [23] R. Espinal, A. Anzola, E. Ad o e , M. Roig, R. Chimen ao, F. Medina, E. L´ opez, D. Bo io, J. Llo ca, Du able e hanol s eam e o ming in a ca aly ic memb ane eac o a mode a e empe a u e o e cobal hyd o alci e, In e na ional Jou nal o Hyd ogen Ene gy 39 (21) (2014) 10902–10910. [24] R. Espinal, E. Taboada, E. Molins, R. J. Chimen ao, F. Medina, , J. Llo ca, Cobal hyd o alci e o he s eam e o ming o e hanol wi h sca ce ca bon p oduc ion, RSC Ad ances 7 (2) (2012) 2946–2956. [25] R. Espinal, E. Taboada, E. Molins, R. J. Chimen ao, F. Medina, J. Llo ca, Cobal hyd o alci es as ca alys s o bioe hanol s eam e o ming: The p omo ing e ec o po assium on ca alys ac i i y and long- e m s abili y, Applied Ca alysis B: En i onmen al 127 (2012) 59–67. [26] H. Li, A. Goldbach, W. Li, H. Xu, PdC o ma ion in ul a- hin pd memb anes du ing sepa a ion o H2/CO mix u es, Jou nal o Memb ane Science 299 (1-2) (2007) 130–137. [27] D. Wang, T. B. Flanagan, K. L. Shanahan, Pe mea ion o hyd ogen h ough p e- oxidized Pd memb anes in he p esence and absence o CO, Jou nal o Alloys and Compounds 372 (1-2) (2004) 158–164. 21 [28] A. A. S oo ogel, A. Sabe i, B. M. Chen, Full and educed-o de obse e -based con olle design o H2op imiza ion, In e na ional Jou nal o Con ol 58 (4) (1993) 803–834. 22 (a) Con ol ac ions o he closed-loop sys em wi h a posi i e e e ence (b) Hyd ogen ou pu o he closed-loop sys em wi h a posi i e e e ence Figu e 7: Simula ion esul s o he i s case o con olle uning (M=N) 23 (a) Con ol ac ions o he closed-loop sys em wi h a posi i e e e ence (b) Hyd ogen ou pu o he closed-loop sys em wi h a posi i e e e ence Figu e 8: Simula ion esul s o he second case o con olle uning (M > N) 24