Modelling and real-time optimisation of an industrial cooling-water network

Modelling and eal- ime op imisa ion o an indus ial cooling-wa e ne wo k Ma ´ ıa P. Ma cos Sys ems Enginee ing and Au oma ic Con ol Depa men Uni e sidad de Valladolid Valladolid, Spain [email p o ec ed]a.es C´ esa de P ada Ins i u e o Sus ainable P ocesses (IPS) Sys ems Enginee ing and Au oma ic Con ol Depa men Uni e sidad de Valladolid Valladolid, Spain [email p o ec ed]a.es Jos´ e Luis Pi a ch Sys ems Enginee ing and Au oma ic Con ol Depa men Uni e sidad de Valladolid Valladolid, Spain [email p o ec ed]a.es Ch is ian Jasch Lenzing Ak iengesellscha We ks aße 2, 4860 Lenzing, Aus ia [email p o ec ed] Abs ac —This wo k deals wi h he p oblem o dis ibu ion o cooling wa e in an e apo a ion p ocess. The aim is o de elop a Real-Time Op imisa ion (RTO) ool which imp o es he esou ce e ficiency by supplying he op imal wa e dis ibu ion wi hin a su ace-condense s ne wo k o a gi en p oduc ion demand. The app oach includes expe imen al models and he au oma ic upda e o ouling ac o . The p oblem is o mula ed and sol ed ia non- linea p og amming. P oduc ion cons ain s and conce ns abou he p ac ical implemen a ion a e also aken in o accoun in he design o he RTO ool. Index Te ms—modelling, op imisa ion, e apo a ion plan , op- imal dis ibu ion, expe imen al models, RTO I. INTRODUCTION In he p ocess indus y, he e is an inc easing consensus on he impo ance o how o manu ac u e he p oduc s in he bes possible way. To do his, we mus ake in o accoun he eal- ime p oduc ion si ua ion and he global ene gy and esou ce e ficiency. Fu he mo e, we mus add ha he egula ion o en i onmen al ma e s is inc easingly es ic i e. As a esul , i an indus y wan s o keep being compe i i e in a global ma ke , i will ha e o pe o m op imisa ion a di e en le els: con ol laye , Real-Time Op imisa ion (RTO), p oduc ion scheduling and economic planning [1]. Imp o emen s on all his le els can lead o huge sa ings in consump ion o ene gy and esou ces, and consequen ly o he educ ion o p oduc ion cos s [2]. In o de o do ha , i is necessa y o p o ide compu e - based ools which acili a e he decision-making p ocess o he ope a o and plan manage s. These ools a e no mally model- based so ha an impo an e o in adap ing heo e ical models o he eal sys em [3] has o be done. In pa icula , RTO ools will need eal- ime inpu s, so hey mus be in eg a ed wi h he in o ma ion echnology (IT) in as uc u e o he plan s, e.g., This esea ch is unded by he Eu opean Union’s Ho izon 2020 esea ch and inno a ion p og amme unde g an ag eemen No. 723575, and by he Spanish Go e nmen (MINECO/FEDER DPI2015-70975-P). ia a neu al deploymen pla o m ha connec s o di e en IT sys ems [4]. This wo k deals wi h RTO in he cooling sys em o he e apo a ion ne wo k in Lenzing A.G., one o he wo ld leading ac o ies o human-made iscose fib e p oduc ion, which is loca ed in Aus ia. Hence, in his pape he app oach and a p o o ypical ool o he op imisa ion o he cooling wa e dis ibu ion in he cooling sys em is desc ibed , wi h he goal o minimizing he ade-o de e mined by he cos o he s eam and he cooling wa e consump ion. The op imisa ion has been p og ammed in CasADi [5] using MATLAB, and hen linked o he PI Sys em in he plan . The pape o ganizes as ollow. Nex sec ion b iefly desc ibes he indus ial p ocess and he ne wo k which is going o be op imised. Sec ion 3 shows he models ha ha e been ob ained om expe imen al da a, and he conside a ions aken in o ac- coun o de elop hose models. In Sec ion 4 he ma hema ical o mula ion o he op imisa ion p oblem is exposed, i.e he cooling wa e dis ibu ion and he conce ns abou he p ac ical implemen a ion. Some esul s and a summa y o he wo k done a e gi en in Sec ion 5. Finally, in Sec ion 6 he u u e wo k is shown. II. DESCRIPTION OF THE PROCESS As ea lie was poin ed ou , his wo k is done in collab- o a ion wi h Lenzing AG, a ac o y which p oduces iscose fib es based on a enewable esou ce: wood. Once he wood is sh edded, he cellulose pulp ha con ained he wood is chemically ea ed and i becomes a iscose solu ion. The key s age o p oduc ion is he spinning, i.e he con e sion o his solu ion in o fib es by passing i h ough fine diame e sie es unde p essu e, and in oducing i in o an acid ba h (called spinba h he eina e ). In addi ion o he new solid fib es, sodium sulpha e (Na2SO4) and wa e a e also p oduced Copy igh © 2018 IEEE. Pe sonal use o his ma e ial is pe mi ed. Pe mission om IEEE mus be ob ained o all o he uses, in any cu en o u u e media, including ep in ing/ epublishing his ma e ial o ad e ising o p omo ional pu poses, c ea ing new collec i e wo ks, o esale o edis ibu ion o se e s o lis s, o euse o any copy igh ed componen o his wo k in o he wo ks. This is a p ep in e sion which may con ain e a a. The published e sion is a ailable in IEEE Xplo e. DOI:10.1109/ICSTCC.2018.8540655 as by-p oduc s. This causes he deg ada ion o he spinba h, and, consequen ly, he quali y o he fib es dec eases. The e o e, i is necessa y o egene a e he acidi y o he spinba h h ough he con inuous ex ac ion o wa e and Na2SO4. In o de o do ha , an e apo a ion ne wo k and a c ys allisa ion sec ion, a ached o he p incipal p ocess a e used. The e apo a ion ne wo k is composed o fi een e apo a ion plan s wi h di e en nominal capaci ies. This ne wo k should be able o egene a e he acid ba hs ha come om spinning, aking in o accoun ha he plan p oduces di e en fib es. The e ficiency o each plan depends on di e en ac o s: he e apo a ion load, he ope a ion condi ions (i.e spinba h empe a u e and flow), he pe o mance o he cooling sys em and he ouling s a e. Fo a u he explana ion o he e apo a ion ne wo k and he cha ac e is ics o each plan and i s con ol sys em, he eade is e e ed o [6], [7]. Fig. 1. Simplified scheme o an e apo a ion plan A. Cooling sys em ne wo k The e apo a ion plan s could ha e wo di e en cooling sys ems: one is based on Su ace Condense s (SC, Fig. 1) and he o he one on cooling owe s (ou o he scope o his wo k). The e a e fi een e apo a ion plan s which use SC as cooling sys em. This E=15e apo a o s a e g ouped in wo sub-ne wo ks depending on hei cooling wa e sou ce (see Fig.2). The aim o his SC is o condense he s eam apou ha comes om he e apo a ion plan s, o be used la e in o he pa s o he ac o y. Hence, he cooling wa e dis ibu ion p oblem is, he mo e cooling wa e is p o ided o he SC, he mo e e ficien he e apo a ion plan becomes, because less specific s eam apou is needed. Ne e heless, he o al a ailable cooling wa e om sou ces is limi ed and sha ed wi h o he depa men s o he ac o y, he e o e i s use implies a cos . Fig. 2. Cooling sys em ne wo k III. MODELLING Se e al expe imen s ha e been done in o de o s udy he beha iou o SC in s eady s a e1in di e en ope a ing condi ions. These expe imen s consis o unning he SCs in di e en condi ions co e ing hei usual ange o ope a ion and collec da a o he inle and ou le empe a u es o he cooling wa e . I has o be aken in o accoun ha o all he expe imen s, he e apo a ion capaci y (EC), i.e he flow o e apo a ed wa e om he spinba h, is kep cons an . F om collec ed da a, di e en s a ic models ha e been buil o ep esen he cooling sys em o each plan . A. Ou le empe a u e Fi s , models p o iding he ou le empe a u e o cooling wa e Tou wi h espec o he cooling wa e flow Fha e been ob ained. I has o be emphasised ha he inle cooling wa e empe a u e is assumed cons an in all he expe imen s. As a esul , he ela ion be ween ou le empe a u e and cooling flow can be fi ed o a polynomial cu e, di e en o each e apo a o . (See an example in Fig. 3.) Fig. 3. Expe imen al model empe a u e s. flow 1Dynamics o p essu e d op and hea ans e in he SCs a e neglec ed because o hem being qui e as in compa ison o he ime he plan is going o ope a e in he op imal s eady s a e. Ne e heless, hese models depend on he inle wa e em- pe a u e, so ha we p opose emo ing he e e ence inle wa e empe a u e Tin eco ded a he ime he es s we e ca ied ou . Hence, we ge an inc emen al model (1), so ha , gi en a eal- ime measu emen o he inle empe a u e, deno ed by ˆ Tin, he ou le empe a u e can be compu ed by (2). ΔT= (F)−Tin (1) ¯ Tou =ΔT+ˆ Tin (2) Whe e (·)is a non-linea unc ion ha ep esen s he expe - imen al model, e.g. he polynomial cu e depic ed in Fig. 3. In addi ion, he SC su e s om ypical ouling e ec s as in o he simila indus ies, p o oking ha he hea ans e dec eases wi h ime and, consequen ly, he ou le empe a u e o he cooling wa e will also dec ease. Thus, assuming he es s we e ca ied ou wi h he SC ully clean he eal ime ou le empe a u e will be lowe han he p edic ed as i is shown in Fig. 4. Fig. 4. Model adap a ion o cu en ouling s a e To o e come his issue, he idea is o add a bias pa ame e K o he abo e base model which adjus s he cu e o he eal- ime measu emen : Tou =ΔT+ˆ Tin −K (3) In his way, he cu en s a e o ouling in he SC is aken in o accoun o u he op imisa ion. The bias K can be easily upda ed wi h eal- ime measu emen s o he ou le empe a u e (ˆ Tou ) by: K =¯ Tou −ˆ Tou (4) No e ha his app oach allows o isola e he ouling e ec s in he SC sys em om he ones in he spinba h hea ing line (Fig. 1), which also a ec s he o e all specific s eam consump ion. Fouling in he hea ing line is ou o his wo k, o de ails on how o deal wi h i see [8]. B. Specific S eam Consump ion On he o he hand, om he collec ed da a o inle and ou le empe a u e and olume ic flow o he cooling wa e , he ac ual cooling capaci y in he SC can be compu ed by: Cpow =4.18 3600F(Tou −Tin)(5) Mo eo e , by eco ding he li e s eam consump ion o he e apo a ion plan in he es s, we can depic he specific s eam consump ion (SSC) e sus he a ailable cooling powe in he SC sys em and fi a model o i . Howe e , analogous o he empe a u e model, o emo e he dependency on he ope a ing poin (load) om he base model is needed. To do so, he simples idea is o compu e he bes specific s eam consump ion (BSSC), i.e he minimum SSC ob ained in he expe imen s, and o build an inc emen al model: ΔSSC =g(Cpow)−BSSC (6) Whe e g(·)is ano he non-linea unc ion ha ep esen s he expe imen al model, e.g. he polynomial cu e depic ed in Fig. 5. Fig. 5. Specific s eam consump ion s. cooling powe Fo ha o be ue, wo assump ions a e made: •The es s we e ca ied ou wi h clean SC. •The model o ΔSSC (i.e., he shape o he cu e in Fig. 5 o ins ance) does no a y significan ly om one ope a ion poin o ano he (plan e apo a ion loads). C. Modelling me hodology The polynomial unc ions (·)and g(·)ha e been ob ained fi ing he cu es o he expe imen al da a by leas squa es me hod independen ly o each e apo a o . Based on he da a, polynomials wi h deg ades no g ea e han h ee a e enough o ge a good fi . These models ep esen sa is ac o ily he sys em in he ope a ion ange and a e sui able o he op imisa ion. Ne e heless, his me hodology does no ake in o accoun possible ou -laye poin s in he expe imen al da a due o dis u bances o noise. Consequen ly, hese fi s migh no be he bes o ep esen he eal beha iou o he sys em. Thus, a mo e sophis ica ed modelling ou ine o ob ain hese cu es, as i is p oposed in [9], could be aken in o accoun . Once bo h models, (·)and g(·), ha e been fixed by iden ifica ion, hey can be used o p edic ion, ecei ing he wa e flows Fand he inle empe a u es ˆ Tin as inpu s and p o iding he wa e ou le empe a u es Tou and he inc ease o specific s eam consump ion ΔSSC. No e ha al hough he measu ed ou le empe a u e ˆ Tou is also used o upda e he pa ame e K , i is no conside ed as a model inpu in he p edic ion s a e. IV. NETWORK OPTIMISATION As men ioned abo e, he e apo a ion ne wo k ha uses SC as cooling sys em can be g ouped in wo sub-ne wo ks: he fi s one (SN1) is composed o ESN1=4plan s and he second one (SN2) includes he o he s ESN2=11. I should be aken in o accoun ha all he SCs a e connec ed in pa allel and ha SN2 can ecei e wa e om SN1 bu no he o he way a ound. Fig. 6. Simplified scheme o he cooling sys em ne wo k The op imal dis ibu ion s ongly depends on he e ficiency o each plan , i.e he nominal specific-s eam consump ion SSC, de e mined by he assigned e apo a ion capaci y EC and he ouling s a e K . Thus, he p oblem objec i e is minimizing he ade-o be ween he cos o li e s eam and wa e usage, which is gi en by he absolu e s eam consump ion (ASC) imes i s p ice (Ps eam) plus he wa e flow (F) imes i s p ice (Pwa e ). The Ps eam is gi en by he ene gy depa men , meanwhile, he Pwa e is calcula ed by nego ia ion be ween all in ol ed depa men s. A. Ma hema ical o mula ion Fo he wo se s o SC e apo a o s, he p oblem cons ain s a e as ollow: •The o al flow in each subne (SN1, SN2) has o be lowe han he maximum limi (FS1,FS2 espec i ely). •Exceeding wa e can go om SN1 o SN2 bu no backwa ds. •Uppe and lowe flow limi s defined o each SC (Fe,Fe), i.e sui able ope a ion ange in o de o a oid p oblems o en ainmen 2in o he SC. •The ou le wa e empe a u e pe plan has o be lowe han he maximum allowed (Tmax). This cons ain is because he ou le cooling wa e goes o he i e and i has o ulfil he cu en en i onmen al cons ain s. 2En ainmen he e is unde s ood as he p esence o he acid ba h in he SC pipes. This e ec is pa icula ly ha m ul o he equipmen , so i needs o be a oided. Hence, he op imisa ion p oblem is: min Fe,FN12∈RE+1 J= E  e=1 (ASCe·Ps eam +Fe·Pwa e )(7a) s. .: ESN1  e=1 Fe+FN12 ≤FS1 (7b) ESN2  e=1 Fe−FN12 ≤FS2 (7c) FN12 ≥0(7d) Fe≤Fe≤Fe∀e∈E (7e) Tou e≤Tmax ∀e∈E (7 ) Taking in o accoun ha FN12 s a es o he wa e om SN1 o SN2 and ASCecan be ob ained as o he specific s eam consump ion and he assigned load capaci y o each plan , as shown in: ASCe=ΔSSCe·ECe.(8) Whe e Tou has been calcula ed by he expe imen al model o each plan as s a ed in sec ion III, o mula (3), and ΔSSC can be ob ained combining (5) and (6) as ollows: ΔSSCe=g4.18 3600Fe(Tou e−Tine)−BSSCe(9) Finally, as he expe imen al models a e C1non-linea unc- ions, usually quad a ic polynomials, he op imisa ion p ob- lem (7) is easily handled ia non-linea p og amming (NLP). B. Implemen a ion Once he p oblem is o mula ed, we coded i in CasADi- Ma lab using an in e io poin op imise [10]. Ne e heless, some conside a ions ha e o be aken in o accoun . Fi s , as he expe imen al models ha e been buil o an specific flow ange, i he eal- ime cooling wa e flows ˆ Fe a e ou o ange, he op imisa ion should no be execu ed, as he es ima ion o he ouling pa ame e K in (4) may be w ong due o plan -model misma ch. The e o e, a wa ning message should appea o in o m o such si ua ion. Howe e , i his happens because he plan s a e in main enance, i.e hey a e no p ocessing any spinba h, he flow o his e apo a o mus no be op imised, bu he es o he ne wo k mus . To do ha , i he load o an e apo a o (EC) is less han 1, he flow o ha condense is se o he one ha is measu ed a ha momen , i.e we p opose eplacing cons ain (7e) by he ollowing exp essions: Fe≤Fe≤Fe∀e∈{e|ECe>1}(10) Fe=ˆ Fe∀e/∈{e|ECe>1}(11) Finally, i may be he case ha , due o he s a e o ouling and/o he wa e inle empe a u e o he condense s, he op imum cooling capaci y is ou o he ange whe e he expe - imen al models (6) we e buil . In his case, he op imisa ion should no un and a wa ning should be displayed, in o de o in o m he ope a o s o his si ua ion. V. RESULTS AND DISCUSSION The op imisa ion o mula ed in he sec ion abo e has been es ed o fline wi h eal sample da a, eco ded in a pa icula ime ins an . The alues o he pa ame e s3used o sol e he p oblem a e: •Maximum allowed empe a u e, Tmax =31 ◦C •Cooling wa e a ailable om sou ce 1, FS1 = 864 m3/h •Cooling wa e a ailable om sou ce 2, FS2 = 943 m3/h In Figs. 7-9 he ob ained esul s a e shown along wi h he eal- ime measu ed da a o he ime ha he pa ame e s we e eco ded. Fig. 7. Cooling wa e dis ibu ion Compa ing he op imised flows wi h he measu ed ones (Fig.7), i is obse ed ha mos e apo a o s mus inc ease hei cooling wa e consump ion. No e ha , his can look inconsis en a p io i as he cooling wa e has a cos . Howe e , as shown in he Fig.8, he consump ion o s eam has dec eased because i s p ice is 10 imes highe han he wa e one, so when adding bo h cos s in he objec i e unc ion he esul is ha benefi s ha e been ob ained. Fig. 8. Inc emen al Specific S eam Consump ion I should be no ed ha he ΔSSC o e apo a o s 18 and 33 is ze o. This is because he expe imen al models o ob ain he SSC a e no consis en , so we assumed ha he measu ed SSC 3The alues o he Tin, EC, Fe,Fe,Ps eam and Pwa e a e omi ed due o confiden iali y easons wi h Lenzing AG. is he bes SSC ha hese e apo a o s can achie e. Doing ha he flow o hese SCs will fi wi hin he ope a ion flow ange, wi h he unique cons ain o complying wi h he maximum ou le wa e empe a u e. In con as , o e apo a o s 35 and 36, he e is a eal specific s eam consump ion bu he op i- misa ion adjus s he flow in o de o make ha op imal SSC ma ches wi h he BSSC in bo h e apo a o s. Analysing Fig.9 we can obse e ha he eal- ime measu e- men o he ou le empe a u e goes beyond he maximum al- lowed in se e al cases. In con as o ha , he op imal solu ion chooses he flows in o de o fi he ou le empe a u e below he maximum, hus ulfilling he en i onmen al es ic ions. Fig. 9. Ou le cooling wa e empe a u e Fu he mo e, flow om he subne 1 o subne 2, FN12, is 19 m3/h, so we can sum up ha he o al cooling wa e a ailable in subne 2 is no enough o ope a e he SCs in hei op imal poin . Table I shows he cos s due o he cooling sys em be o e and a e he op imisa ion, as well as he alue o he sa ings. No e ha hese alues only ep esen a snapsho in a pa icula ime ins an , bu he annual po en ial benefi ha would be ob ained by applying his op imisa ion ool in daily ope a ion looks p omising. The e o e, he ool is now deployed on si e, cu en ly unde es ing pe iod. TABLE I RESULTS OF COSTS AND SAVINGS Cos be o e op imisa ion 71.39e/h 625376.40e/yea Cos a e op imisa ion 19.88e/h 174148.80e/yea Sa ings 51.51e/h 451227.60e/yea Summa y and conclusions In his wo k we add essed a p oblem on esou ce e ficiency in a cooling sys em o a eal indus ial e apo a ion ne wo k. The modelling, op imisa ion and isualiza ion concep s p e- sen ed in his pape suppo he ope a o s when i comes o aking be e decisions in eal ime o imp o e he ne wo k ope a ion. The models ob ained allow o execu es an au oma ic upda e o he SCs ouling s a e based on eal- ime measu e- men s. Ne e heless, he models de eloped a e e y sensi i e o dis u bances and noise e o o he da a used o ob ained he polynomial unc ions, so a modelling ou ine o calcula e hese cu es is ecommended. The esul ing models a e inco po a ed in he RTO scheme ha sol es an NLP p oblem acco ding o he cu en p oduc- ion cons ain s and he plan s ouling s a es. Ne e heless, he ins an aneous RTO does no ake in o accoun any p edic ion o he ouling e ec , so he p oposed con ol ac ions may be subop imal in he long e m. In despi e o ha , significan s eam consump ion sa ings ha e been ob ained in he es s. The de eloped RTO ools a e cu en ly unde e alua ion a Lenzing AG: he implemen a ion in exis ing sys ems and ope a ional policies is pe o med s ep by s ep o ge expe ience in li e es ing and o ensu e accep abili y om he plan pe sonnel. Abou a yea o no mal ope a ion is equi ed o assess he impac , bu p elimina y es s wi h his o ical da a p edic sa ings a ound 400000 e/yea . Howe e , he op imisa ion canno un au oma ically in he cu en o m due o exis en ou -o - ange si ua ions. In ad- di ion, he a e some e apo a o s which miss consis en da a se s. Fo his eason u he es s a e planned on si e in o de o imp o e he SCs models, he e o e he ool eliabili y. VI. FURTHER STEPS The comple e e apo a ion sys em in Lenzing A.G. can be mainly desc ibed by wo ne wo ks o equipmen . The fi s one conce ns he e apo a ion plan s whe e decisions on load alloca ion need o be aken [8]. The second one e e s o he cooling sys ems a ached o each plan , eed by a wa e dis ibu ion ne wo k, whe e decisions on wa e flows o plan s need o be aken, i.e he p oblem add essed in his wo k. Cu en ly, independen op imisa ion se ups a e al eady de- eloped o each ne wo k. Howe e , bo h sys ems a e coupled by he spinba h loads ECeand he ou le wa e empe a u es Tou ,e in he cooling sys ems, so ha he decisions on he load alloca ion influence he cooling-wa e dis ibu ion and ice e sa (see he o mula ion o he op imisa ion in [8] and compa e wi h he one in Sec ion IV-A). Fig. 10. Rela ion be ween ne wo ks I bo h p oblems a e sol ed independen ly, by ea ing hese sha ed a iables as “a p io i” fixed da a o each p oblem, he o e all op imisa ion becomes an i e a i e p ocedu e wi h no global op imali y gua an ees. On he con a y, i bo h o mu- la ions a e me ged in o a cen alized p oblem, he op imisa ion becomes an MINLP one, as he wa e dis ibu ion is an NLP p oblem and he load alloca ion in ol es disc e e decisions (alloca ion o plan s o p oduc s). The e o e, he complexi y o he hypo he ical cen alized p oblem is much highe han he one o he wo sepa a e p oblems abo e, guessed o be unsui able o eal- ime applica ion. To o e come he abo e issues, we p opose as u u e wo k, o add ess he p oblem in a dis ibu ed ashion ia Lag angean decomposi ion [11] and p ice-coo dina ion schemes [12], i.e., adding he sha ed cons ain s ( a iables in his case) as a penal y in he objec i e unc ion Jo each indi idual p oblem. The modified objec i es o each p oblem will be in he o m: JM=J+ e pe(Re−ˆ Re)(12) Whe e Rea e he sha ed a iables ha should be equal ( o ˆ Re) in bo h p oblems and pea e he associa ed Lag ange mul iplie , also e e ed as “ esou ce p ices” in he li e a u e. Then, a sma ule o upda ing ˆ Reand pein each i e a ion is equi ed. Resea ch on his will be conduc ed in o de o speed up he esolu ion. REFERENCES [1] C. S. Kho and D. Va a ezos, “Pe oleum efine y op imiza ion,” Op imiza ion and Enginee ing, ol. 18, no. 4, pp. 943–989, Dec 2017. [2] S. K ¨ ame and S. Engell, Resou ce E ficiency o P ocessing Plan s: Moni o ing and Imp o emen . John Wiley & Sons, 2017. [3] C. de P ada, D. Sa abia, G. Gu ie ez, E. Gomez, S. Ma mol, M. Sola, C. Pascual, and R. Gonzalez, “In eg a ion o o and mpc in he hyd ogen ne wo k o a pe ol efine y,” P ocesses, ol. 5, no. 1, 2017. [4] LeiKon, “D4.1 equi emen specifica ion o he in eg a ed deploymen pla o m,” Ou comes o he MORE P ojec , 2014. [5] J. Ande sson, J. ˚ Akesson, and M. Diehl, “Casadi: A symbolic package o au oma ic di e en ia ion and op imal con ol,” in Recen Ad ances in Algo i hmic Di e en ia ion, S. Fo h, P. Ho land, E. Phipps, J. U ke, and A. Wal he , Eds. Be lin, Heidelbe g: Sp inge Be lin Heidelbe g, 2012, pp. 297–307. [6] J. L. Pi a ch, C. G. Palac´ ın, A. Me ino, and C. de P ada, “Op imal ope a ion o an e apo a ion p ocess,” in Modeling, Simula ion and Op imiza ion o Complex P ocesses HPSC 2015, H. G. Bock, H. X. Phu, R. Rannache , and J. P. Schl¨ ode , Eds. Cham: Sp inge In e na ional Publishing, 2017, pp. 189–203. [7] C. G. Palac´ ın, J. L. Pi a ch, C. Jasch, C. A. M´ endez, and C. de P ada, “Robus in eg a ed p oduc ion-main enance scheduling o an e apo a- ion ne wo k,” Compu e s & Chemical Enginee ing, ol. 110, pp. 140 – 151, 2018. [8] M. Kalliski, J. L. Pi a ch, C. Jasch, and C. de P ada, “Suppo o decision-making in a ne wo k o indus ial e apo a o s,” Re is a Ibe oame icana de Au om´ a ica e In o m´ a ica indus ial, 2018, In P ess, DOI:10.4995/ iai.2018.9233. [9] M. P. Ma cos, J. L. Pi a ch, C. Jasch, and C. de P ada, “Modeling o op imal ope a ion o an indus ial cooling sys em,” in Ac as de las XXXIX Jo nadas de Au om´ a ica. Uni e sidad de Ex emadu a, 2018, Accep ed. [10] A. W¨ ach e and L. T. Biegle , “On he implemen a ion o an in e io - poin fil e line-sea ch algo i hm o la ge-scale nonlinea p og amming,” Ma hema ical P og amming, ol. 106, no. 1, pp. 25–57, Ma 2006. [11] M. L. Fishe , “The lag angian elaxa ion me hod o sol ing in ege p og amming p oblems,” Managemen science, ol. 27, no. 1, pp. 1–18, 1981. [12] R. Ma ´ ı, “P ice coo dina ion s a egies in la ge-scale p ocess con ol,” Ph.D. disse a ion, Escuela de Ingenie ´ ıas Indus iales, Uni e sidad de Valladolid, Spain, 2015.