multivariable control and real-time optimization - an industrial practical view (2005)

8
PROCESS AND PLANT OPTIMIZATION SPECI LREPORT Multivariable control  a n d real-time optimization—an industrial practical view Here s what can  b e  achieved with each technology and how they  a r e  integrated  at  this plant O. ROTAVA  and  A C ZANIN Petrobras Brazil P rocess optimization, once mainly  a  subject  o f  academic inter est, was made possi ble when tools such  a s  multivari- able control  MVC)  an d  real- time optimization  RTO) became available. These tools  are a  consequence  of  the move from analog  to  digital instrumen tation  an d  development  of  t he distributed control system (DCS). Among people  no t  familiar with control theory,  and  even among process engineers,  a  perception exists that implementing advanced control  in the  form  of  multivariabl e controller  on a process unit solves once  and fo r all the  optimization problem of that unit.  To  t he  disappointment  an d  frustration  of  control engineers that implement such systems  a n d  managers responsible for automation investments, evaluating multivariable controller payback after  it has  operated  for a  while shows that profits  a re small er than those p lanned. MVC generates benefits  in  three ways: stabilizing the process (decoupling  the  manipulated variables), protecting  the  process from violating o perating c onstraints and using ava il able degr ees of freedom. This allows some constraints  t o  become active, optimiz- ing the ptocess by maximizing yield oFthe most valuable products and minimizing expenditure  o r a w  materials  a nd  utilities. The first  and  second points  a re  usually true: MVC stabilizes the process  a nd  protects  i t  f rom violating constraints. H ence,  the frustration i s largely due  t o  no satisfaction Fr om point three. Such dissatisfaction  i s  based  on  a  real  a nd  a False assumption. The real assumption: By lack oFconstant tollow-up by  a  process engineer, MV C o peration tends to degrade with time. Eventual changes  i n plant operation  an d  production objectives require changes  i n  th e controller structure, sometimes immediately after commissioning. These changes are  n o t  done because experts ar e  not  available. The most common reasons  f o r  degradation are: i nadequate limit s  o f the manipulated variables, conflicting specifications  o f  the  con- trolled variables  a n d  inadequat e tuning. The false assumption  is  that MVC gives  a  comp lete solution to  th e  unit optimization problem. Actually, MVC is required  to perform functions that  ar e  not in its  designed scope. Examples are distributing  the  heat load between reflux  an d  pumparounds in  a n  atmospheric distillation unit  an d  defining  the  reaction temperature  of  maximum conversion  in an FCC  unit. These features  ar e  no t  covered  by  MVC. Only  an  RTO  system, based on  a  rigorous process model, takes care  o f  both of them . This unfair view oFwhat  can be  achieved with  MVC  in terms  of  ptocess optimization allows continuous questioning of the heavy investments applied  in  upgrading instrumenta- tion.  In an  MVC  and RTO  implementation project, invest- ments  in  instrumentation  and a  DCS,  if  not already  i n  place, represent  the  largest cost.  On the  other hand,  it is  important that such investments should  be  made anyway, even without implementing MVC  and  TO  because  o f  analog instrumenta- tion obsolescence. Nevert hel ess, upgrading instrumentation and DCS implemen- tation  do not  define a competitive advantage because competitors are obliged  to do the  same.  The  convenience  oF  implementing M VC  and RTO  tools provides  the  rea l advantages. The  objective  of  this  articl e is to show clear ly throu gh examples what can be achieved in ter ms of optimizati on ft om MVC and what can only be obtained through RTO vi a a rigorous process model. Suppliers of automation  an d  optimization technology usually split total benefits related to petroleum refining operation optimi- zation equally between MVC  a nd  RTO  ap plications. According to Cutler  and  Perty,' available benefits  f o r  online optimization plus advanced control  c a n  amount  to  6—10 of  the  added value of  given process. Multivariable control: steady state quadratic opti mization and dynamic control.  Iwo distinct algorithms {Fig.  1 )  executed  at the  same frequency perform MVC. These algorithms  ar e  associated with solving  the  pr ocess control static and dynamic problems. The algorithm that corresponds to solving the st atic problem, executed  First,  searches  a  set of optimum val ues of t he manipulated variables, maximizing  a  quadratic objective function, namely operating proFit. Solving this problem must satisfy constraints established  by  maximum  and  minimum values  o th e  controlled and manipulated variables. Controlled variables  a r e  evaluated  at the steady state produced  b y  the  linear dy namic process model, with manipulated variables and disturbances as inputs. The result- ing optimum operating point  is  then sent  t o  the objective ftm ctio n of  the  dynamic problem. The second algor ithm solves the dynamic problem and accom- plishes  tw o  tasks.  The  first consists  o f  keeping  th e  process inside the operating region, compensating  f or  the frequent disturbances HYDROCARBON PROCESSING JUNE 2005  61

Upload: yang-gul-lee

Post on 03-Jun-2018

220 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Multivariable Control and Real-time Optimization - An Industrial Practical View (2005)

8/12/2019 Multivariable Control and Real-time Optimization - An Industrial Practical View (2005)

http://slidepdf.com/reader/full/multivariable-control-and-real-time-optimization-an-industrial-practical 1/8

PROCESS A N D PLANT O PT IMIZA TION SPECI LREPORT

Multivariable control and

real-time optimization—anindustrial practical viewHere s w hat can be achieved w ith each technology

and how they are integrated at this plant

O. ROTAVA and A C ZANIN Petrobras Brazi l

Process optimization, once mainly a subject of academicinterest, was made possible when tools such as m ultivari-able control  MVC) and real-time optimization  RTO)

became available. These tools are a consequence of the movefrom analog to digital instrumen tation  and  development of the

distributed control system (DCS).

Among people not  familiar with control theory, and evenamon g process engineers, a perception exists that implem entingadvanced control in the form of  multivariable controller on a

process unit solves once and for all the optimization problemof that unit. To the disappointment and frustration  of controlengineers that implement such systems and managers responsible

for automation investments, evaluating multivariable controllerpayback after  it has  operated for a while shows that profits aresmaller than those p lanned.

MVC generates benefits in three ways: stabilizing the process(decoupling the manipulated variables), protecting the processfrom violating operating constraints and using available degrees offreedom. This allows some constraints to become active, optimiz-ing the ptocess by maximizing yield oFthe most valuable productsand minimizing expenditure o raw materials and utilities.

The first and second points are usually true: MVC stabilizesthe process and protects it from violating constraints. H ence, the

frustration is largely due to no satisfaction From po int three. Suchdissatisfaction   is based on a real and  a False assum ption. T he real

assumption: By lack oFconstant tollow-up by a process engineer,MV C o peration tends to degrade with time. Eventual changes inplant operation and production objectives require changes in the

controller structure, sometimes immediately after commissioning.These changes are not done because experts are not available. Themost common reasons for degradation are: inadequate limits of

the manipulated variables, conflicting specifications of the con-

trolled variables and inadequate tuning.

The false assumption is that MVC gives a comp lete solutionto the unit optimization problem. Actually, MVC is required toperform functions that are not in its designed scope. Examplesare distributing the heat load between reflux and pumparoundsin an  atmospheric distillation unit and defining  the reactiontemperature of  maximum conversion  in an FCC unit. Thesefeatures are not covered by MVC. Only an RTO  system, based

This unfair view oFwhat  can be  achieved with  MVC  in

terms of ptocess optimiza tion allows contin uou s q uestion ingof the heavy investments applied  in  upgrading instrumentation. In an MVC and RTO implem entation project, investments in instrumentation  and a DCS, if not already in placerepresent the largest cost. On the other h and, it is  importanthat such investments should be made anyway, even withouimplementing MV C and RTO because of analog instrum entation obsolescence.

Nevertheless, upgrading instrumentation and D CS implemen-tation do not define a competitive advantage because competitorare obliged to do the same. The convenience  oF  implementin

MVC and RTO tools provides the real advan tages.The objective of this article is to show clearly throu gh example

what can be achieved in terms of optimization ftom MVC and whatcan only be obtained through RTO via a rigorous process model.

Suppliers of autom ation and optimization technology usuallysplit total benefits related to petroleum refining operation optimi-zation equally between MVC and RTO ap plications. Accordinto Cutler and Perty,' available benefits  for  online optimizationplus advanced control can amount to 6—10 of the added valuof  given process.

Multivariable control: steady state quadratic opti

mization and dynamic control.  Iwo distinct algorithms

{Fig. 1) executed  at the  same frequency perform MVC. Thesalgorithms are associated with solving the process control statiand dynamic problems.

Th e algorithm that corresponds to solving the static problem,executed First, searches a set of optimum values of the manipulatevariables, maximizing a quadratic objective function, nameloperating proFit. Solving this problem must satisfy constraintsestablished by maximum and minimu m values o the controlleand manipulated variables. Controlled variables are evaluated atthe steady state produced by the linear dy namic process m odelwith manipulated variables and disturbances as inputs. The result-ing optimum operating point is then sent to the objective ftm ctioof the  dynamic problem.

The second algorithm solves the dynamic problem and accom-plishes two tasks. The first consists of keeping the process insid

Page 2: Multivariable Control and Real-time Optimization - An Industrial Practical View (2005)

8/12/2019 Multivariable Control and Real-time Optimization - An Industrial Practical View (2005)

http://slidepdf.com/reader/full/multivariable-control-and-real-time-optimization-an-industrial-practical 2/8

SPECIALREPORT P RO C ES S N D P L N T O P T I M I Z T I O N

subject to:

Y <Y <Y

that enter the plant and would lead the unit to violate its con-

straints. The second consists oi  implementing the optimum val-

ues infotmed by the static solution ofthe first algorithm. The

mathematical procedure that satisfies the requirements solves a

least-squares ptoblem.

The quadra tic optimization (first task) solves the following

problem:

(1)

(2)

(3)

(4)

/here:

w,,, = vector of manipulated variables values at present time

Us - vector of manipulated variables values at steady state

W\  = diagonal matrix of economic coefficients ofthe manipu-

lated variables (partial derivatives ofthe economic func-

tion in relation to the manipulated variables)

W2 -  diagonal matrix of suppression factors oft he manipu-

lated variables

W i,  - diagonal matrix of slack variable weights

Ys = vector of predictions of the controlled variables on the

steady state (those that must strictly satisfy the chosen

boundaries)

Y^' = vector of controlled variable predictions at steady state

(those whose boundaries do not need to be strictly satis-fied)

SCV=  vector of slack variables (represents how much each

variable in Y^' surpasses its boundary; they are added

to guarantee existence of a solution to the optimization

problem).

The dynamic control (second task) is formulated as:

(6)

submitted to the following constraints:

-A« -(y7')<A«(;T)<A«™(;T); y - 1  nl  (7)

u ' [jT <u^+ £ A « ( / T )  < H- ( ; T ) ;  ;• = \,...,nl  (8)

where:

H /=   control horizon

nr =  prediction horizon

7 = sampling or algorithm period execution

A« = vector of control action size

u* = vector of optimal manipulated variable values calculated

by the linear optimizer

^Linear optimizer   1 min)-  steady-state linear model-  QP  or  LP  algorithm

Optimum  values: 0* ,   y

^ Controller  1 min)-dynamic   linear model- predictive multivariable algorithm

• * —

Setpoints manipulated), u  Variables,

DCS  - regulatory control

Two   distinct algorithms (linear optimizer and controller)

perform multivariable  control.

variables

tt^^ = diagonal matrix of manipulated variable supression f

tors

W , =   matrix of weights to lead the manipulated variables

their respective optimum values

yi=  vector of upper or lower boundaries of the dynamica

controlled variables

yj, =  vector oi linear prediction ofthe dynamically controll

variables.

Operating region Performance of MPC: optimumon the constraints The term operating region designate

the polyhedron ot dimension equal to the number of manipulate

variables. This region encloses all feasible unit operating cond

tions. The surface that limits the region is defined by the operati

constraints, made up mainly by the controlled variables. Neverth

less, the operator can also make active constraints related to upp

and lower boundaries ofthe manipulated variables, making th

operating region even more restricted. Obviously, this practice

not recommended, because controller freedom is reduced and th

solution eventually achieved is not the best.

Fig. 2 illustrates an operating region for an FCC unit convert

where two variables are considered: con\ersion (controlled variabl

and teed flowrate (manipulated variable). A solution o fthe MV

linear or quadratic optimizxr corresponds to a constraint vertex

Fig. 2. For instance, if the optimal operating points are represente

by point A (maximum conversion) or B (maximum feed load) th

MVC will fmd the best solution, pro\'ided the convenient coefficien

ofthe manipulated variables (feed flowrate and reactor temperatur

are given to the linear or quadratic optimization objective function

If the largest coefficient corresponds to reaction temperatur

the optimum solution will correspond to point A, otherwise th

operating optimum will be point B. If the operating economic

optimum alternates between points A and B, the relative values

the aforesaid coefficients need to be changed tor the right solutio

to be found by the optimizer. Overall, the optimizer has tuninparameters that are the coefficients  oi the manipulated variables

the linear or quadratic programming  {Wi m  Eq. 1) that must

Page 3: Multivariable Control and Real-time Optimization - An Industrial Practical View (2005)

8/12/2019 Multivariable Control and Real-time Optimization - An Industrial Practical View (2005)

http://slidepdf.com/reader/full/multivariable-control-and-real-time-optimization-an-industrial-practical 3/8

SPECIALREPORT PROCESS AN D PLANT OPTIM IZ ATION

Gas compressorconstraint Max catalyst

circulation rate

Max feedtemperature

Feed pum pconstraint

Min conversion

Feed rate

Operating region of  the FCC converter with two variables

If the operating optimum is  unconstrained (for instance, aninterior poin t of the shadowed area on Fig. 2), ir cannot be foundby the MV C quadratic optimizer.

  ultivariable controller optimization algorithm

l i m i t a t i o n . The hmitation of the  MVC consists of its  inabilityto find an optimum sokition when the optim um is inside theoperating region, since the controller quadratic programmingalgorithm only fmds solutions on the boundaries defined by theprocess constraints.

As an example, consider the diesel production of an atmo-spheric fractionator (Figs. 3 and 4) .

Fig.  shows that an economical balance exists between theincrease of diesel production through its heavy fraction and theenergy consumption in the furnace. For the same feed flowrateand diesel ASTM 85% distillation specification, an increase ofthe coil outlet temp erature (CO T) corresponds to an augment ofthe o\erflash that then improves fractionarion, allowing a largerdiesel draw. A limit to this procedure  is when a diese productionincrease does not compensate economically for  the additionalenergy consumption.

Again in Fig. 3, we see the op timum operating point is a func-tion of an economical balance between the energy cost and dieselprice. For instance, tor a larger energy cost, the optimum operat-ing point consists of  a  smaller COT and consequently smaller

amount of ovetflash.Fig. 4  illustrates the diesel production increase as a function

of the fractionation constrained by the maximum ASTM 85%specification. S tarting from an operating poin t with low overflashflowrate, SV ,  a COT increase allows the process to reach an

optimum overflash flowrate, 5V'"', which is the best economicalbalance between energy consumption and diesel production.Operating with excess overflash.  SV*,  the economical gain withthe additional diesel production does not compensate for the extraenergy consumption cost.

If the  optimization task was attributed to the MVC, it wouldlead the proce.ss to the maximum COT, provided no constraints

were violated, even if maximum CO T was  not the most economi-cally advantageous operating cond ition.

In diesel optimization by MVC, the process is operated with

Energy consumption/feed rate

F I G An economical balance exists between increasing diesel

production and energy consumption

cess analysis team), in sp ite of ch nges in petroleum quality afractionator operating cond itions.

R e a l t i m e o p t i m i z a t i o n .  Optimization through MVChindered by the simplicity of  the process model, which is a of linear equations. The objective function  is related to mamum or minimum values of  the  MVC manipulated variablesquadratic or linear programming algorithm achieves the optimsolution. According to Marlin and Forbes,^ due to the reliabiliand relative simplicity of M VC technology, the preferred locatitor economic optimization would be the controller, providedperformed well.

On the other hand, the R TO model consists of a ser of non lear equations that represent as close as possible th e system steastate.  The objective function consists of  the system econommodel that translates into its profitability. A nonlinear programming algorithm achieves the optimal solution.

Optimization makes the connection between production planing and scheduling tasks and those evaluated by MVC. Fig.illustrates the traditional RTO structure of a system w ith m ultipMVCs, with the interrelationship of its components.^ Its macomponents are:

• Process steady-state model. The m athematical model murepresent the system over a wide range of operating conditiowith a high degree of accuracy in such a way that the m aximuprofitabilit)' predicted by the objective function be effectively tmaximum potential profit of the real process. Also, constrainof the real process must not be violated when the optimizsolution is implemented. According to Friedman,'^ effectioptimiza tion is still limited by availability of good models.

• Data reconciliation and model parameter updating. Riorous models require a  large amount of measured informatithat con tains uncertainties. This subject is dealt with by the dareconciliation procedure. The first data reconciliation proceduconsists of gross  error detection. In  this step, invalid measurments due to instrumentation malkmctioning are identified antreated. Afterward, the small differences in the mass and energ

balances are spread among measurements throughout the prcess model, taking into consideration the statistical uncertaintiof the instruments and redu ndant measurements.

Page 4: Multivariable Control and Real-time Optimization - An Industrial Practical View (2005)

8/12/2019 Multivariable Control and Real-time Optimization - An Industrial Practical View (2005)

http://slidepdf.com/reader/full/multivariable-control-and-real-time-optimization-an-industrial-practical 4/8

SPECIALREPORT PROCESS AN D PLANT OPTIM IZ ATION

process over a wide range of operation conditions. Parameters ofsuch models are updated to compensate for nonm casured distur-bances and changes in process performance caused by factors suchas catalyst deactivation, heat exchanger fouling, and furnace andcolumns efficiency reduction. By analogy to a traditional PID con-troller, model parameter updating using plant data corresponds

to process Feedback.• Optimization algorithm. After the model is properly fit-

ted ro real plant data, optimization is performed. To attain theoptimal solution of the  optimization problem, a nonlinear pro-gramming (NLP) algorithm   is used. Such a solution correspondsto the maximum profitability of the unit inside the operatingregion limited by process constraints. The NLP algorithm usu-ally employed in industrial applications is successive quadraticprogramming (SQP).^

Th e real-time optimization problem is defmed as:

Diesel ASTM 85

Atmosphe t i c

residue

Delta die5el

(economic benefit)

V o l

F I G .  4 Diesel production is a function of the fractionationconstrained by the m aximum  STM 85 specification.

subject to the constraints:

 « «•

 9

(10

(11

(12

(13

where:

- Constraints limits

- Parameters

- Instrumentation data

Scheduling/planning

- Economic data- Constraints: feed/products

Data base/information system

Optimizer

Targets

Processmodel

NLPsolver

Multivariablecontroller 1

Multivariable

controller 2

Setpoints Setpoints Setpoints

DCS  - regulatory control

Traditional real-time optimization structure.

i steady-state disturbances vector,̂,y =  economic objective functionh^  economic model constraintshp  nonlinear model constraints

  vector of constraints on the steady state.

Integrating the real-time optimizer w ith the M VRTO determines the optimum operating point to which tprocess must be driven. This solution cannot be implementdirectly in the DC S due to dynamic con straints. High-frequendisturbances would destabilize the process and move it outsiits constrained region. This can happen because the low-frquency optimizer is not able to deal with such disturbances. its characteristics, MV C is the adequ ate tool to move the procein a robust way to such an operating optim um .

The optimizer solution is normally made a\'ailable to the MVin the form of optimum steady-state targets of its manipulatand/o r controlled variables (Fig. 6).

Fig. 6 illustrates the classical structure of RTO integrated

the M VC , available in most industrial ap plications. This stratedisplays two main functions:

a) Optimization of the  steady state, which is accomplishedthe RTO in a relatively low frequency (superior layer).

b) Implementing the optimal solutiachieved in (a) above by the MVC (intermdiate and lower layers), which is responsibby the dynamic driving of the  process frthe present state to the optimum operatipoint calculated by the RTO.

MVC execution is divided in two part• Dynam ic control, which is executed

a relatively high Frequency and is responsib

for keeping che process inside the enveloestablished by its  constraints and for movthe unit to its optimum operating point.

• Linear optimization, which  is execuin the same frequency as the dynamic cotrol, whose function is to send the RTsolution to the MVC, but making smaadjustments to it due to disturbances enting the process inside the RTO executiinterval.

Thus, each task of the  control and omization strategy is distinct and displays own algorithm, which is executed sequetially at the same MV C run.

Integrating RTO with MVC is achiev

Data reconciliationmodel updating

Multivariable

controller N

Page 5: Multivariable Control and Real-time Optimization - An Industrial Practical View (2005)

8/12/2019 Multivariable Control and Real-time Optimization - An Industrial Practical View (2005)

http://slidepdf.com/reader/full/multivariable-control-and-real-time-optimization-an-industrial-practical 5/8

SPE I LREPORT PRO SS  ND PUNT OPTIMIZ TION

RTO  {4h)

- steady-state rigorous model

- economic model of process- nonlinear program ming {NLP)

IOptimum values: Manipulated variables, u' '

Controlled variables, y' '

Linear optimizer  1  min}

- steady-state linear model-  QP or LP algorithm

Optimum values: u'', y''

Controller  1  min

-dynamic linear model- predictive multivariable algorithm

Setpoints manipulated), u Variables

y

DCS -  regulatory control

F I C .  6  The optimizer solution is normally made available to

the m ultivariable controller in the form of optimumsteady-state targets of its manipulated and/or controlledvariables.

linear optimization layer. In this way, when the RTO is active,Eq.  is modified to:

mmU, Ji V

 14)

subject to constraints represented by Eqs. 2- 5 , where:

u  vector of  optimal values of rhe m anipulated variablesdetermined by the RTO.

The first term of Eq.  14,  which represents implementationof the optimizer solution, does not possess the tuning parameterpresent in Eq. 1. In this case, the  information embedded in the

vector of economic weights, W,, is informed through the productprice.s in the RTO economic function.

C a s e s t u d i e s . Next, some variables of a crude distillation unitand a. flui catalytic cracking  FCC) converter are analyzed. Inthe first case studied, it is shown that MVC  optimization mustbe complemented by RTO. In the other cases, the op timizationtask is exclusively made by RTO.

Cases th at can be optimized exclusively by MVC, i.e., theiroptimum  is on the constraints, are not analyzed. M aximizing jetfuel in a petroleum fractionator and pressure balance on an FCC

converter when compressors/blowers are operating in  their fullcapacity are examples of such cases.

Crude atmospheric distillation unit:  OT and overflashflowrate. CO T  is manipulated  to  optimize diesel production,

generally constrained by m aximum diesel ASTM 85 % distillationtemperature. Fig. 7  illustrates diesel production and overflashcontrol once furnace outlet temperature has been fixed. Overflash

Feed TTTOverflash (optimizer

Overflash is adjusted by the setpoint of the external reflu

The heavy diesel prod uction is determiend by a massbalance over that particular region of the column.

production is determined by a mass balance over that particuregion of the column .

The MV C of the fractionator column increases the CO T unthe diesel ASTM 85% distillation temperature becomes activFig. 8 illustrates the influence of  furnace outlet temperature aoverflash flowrate on the diesel quality.

When operating with a deficient overflash,  SV~ the MV

increases the COT  until operation point 1 is attained, wheVi is the diesel flowrate and the diesel ASTM 85% distillatitemperature constraint becomes active.

As the overflash flowrate is optimized, SV' , increasing tcontroller flowrate setpoint without increasing the COT, a n

operating condition represented by point 2 is achieved in whithe mass and energy balances of the heavy diese region reduthe diesel flowrate to Vj. In this cond ition , there is a giveawaythe ASTM 85%  diese speciflcation.

Starting from point 2, the MVC increases the COT until tdiesel ASTM 85% distillation temperature constraint becomactive again at point 3. In  this condition, diesel flowrate, V'3increased by A V'  in relation to initial point 1.

Operating with excessive overflash, SV'', the operating porepresented  by 4, the economic gain with  the  additional dieproduction, A^'*, does not compensate for the add itional enerexpenditure.

In this case, we verify that the MVC optimizers beneflts—bmaximizing the COT—depends on the overflash flowrate, whooptimum value can be determined by the RTO since it depenon the feed quality and many other variables of the atmosphecolumn.

Determining the optimum overflash flowrate, as shown, is part of the scope of the MV C, thus the  fractionation problerequires an RTO based on a rigorous model.

Distribution  of pumparounds in a  fractionation colum

Distribution of  energy withdrawal through  the  pumparoun{Fig. 9) is exclusively an RTO problem. An economic balan

exists b etween the energy recovery by the pum parounds anthe internal reflux along the column that are responsible for t

products fractionation.

Page 6: Multivariable Control and Real-time Optimization - An Industrial Practical View (2005)

8/12/2019 Multivariable Control and Real-time Optimization - An Industrial Practical View (2005)

http://slidepdf.com/reader/full/multivariable-control-and-real-time-optimization-an-industrial-practical 6/8

SPECIALREPORT PROCESS AN D PLANT OPTIM IZ ATION

umparounds d istribution:

- Frattitniation x energy recoveryD ies e l ASTM 8 5

Optimum stripping steam:- Kerosene flash point- Bottom light components

Heavy naphtha

_ T^ Steam

D iesel

Delta diesel

  e o n o m i RTO

benefit

When operating w ith a deficient overflash, the

multivariable control increases the COT until operation

point 1 is attained.

must consider the following internal refluxes:

• i ] and i g , responsible for the fractionation, respectively, ofthe kerosine light and h ea\y ends

•  L^j responsible for the fractionation between diesel andatmospheric residue

• I46, that corresponds to overflash flowrate.

w w w , G E A J E T . C O M

More than

just vacuum

We provide optimal, process in-

tegrated solutions for any type of

vacuum system, using our well

known jet pumps together with

other types of vacuum pumps.

Our jet vacuum pum ps are used

in oil refineries, steel degassing,

deodorizing and dehumidifyingof vegetable oils/fatty acids, sea

water desalination and vjriou.s

applications in the chemical, phar-

maceutical and food industries.

The opiimum design of a jet

vacuum pitmp requires a lot of

experience. W e have it.

GEA Jet Pumps

GEA Jet Pumps Gm bH • Einsteinstrasse 9-15 • 76275 Et tling en • German y

Tel.:  +49 7243 705-0 • E -mai l: in fo®g ea je t . de • I n te rne t ; www ,gea je t . com

Atmospheric residue

D istribution of energy withdrawa l through the

pumparounds is exclusively a real-time optimization

problem.

As consequence of the strong system nonlinearities, the coumn duty distribution is not conveniently handled by MVwhich operates hased on linear m odels. In cases where the M Vmanipulates pum paround flowrates heir setpoint values remaalmost permanently on their operating limits, not  eing effectiv

manipulated MVC variables.Additionally, heat load distribution must consider pressu

drops in the pumparounds regions to optimize pressure in tvaporization zone. This task is exclusive of an optimizer w ithrigorous model because MVC does not take into account tcolumn tray hydraulics for flooding detection.

Stripping steam  flowrate Kerosinc .stripping steam is meantbe the fine-tuning factor of tbe product flash point property; meawhile, stripping steam in the bo ttom of the fractionator removes light components of tbe atmospheric residue. Only through a rigous process model is it possible to determine the optimum strippisteam flowrates. Tbe optimizer can  determine an optimum produsteam ratio, which depends on crude quality, and the steam flowrcan be controlled by regulatory control in the DCS.

Stripping steam manipulation cannot he bandied by MVbecause the process model obtained from plant tests) depenon the steam rate at the very moment of  the  test. For instancethe steam flowrate is above a certain value, the controlled property is not sensible to it. Therefore, this is a system w ith stronnonlinearities in w hicb tbe econom ic gain may change its signFor instance, sometimes it is economically convenient to increathe steam injection; other times the optimum is obtained by topposite procedure.

Preheat trains and furnace feed allocation One of the opquestions in optimizing distillation unit operation consists

allocating the crude feed to preheat trains and furnaces.W hen active constraints exist, the allocation problem can

solved with MVC alone by keeping tbe process variables equal

Page 7: Multivariable Control and Real-time Optimization - An Industrial Practical View (2005)

8/12/2019 Multivariable Control and Real-time Optimization - An Industrial Practical View (2005)

http://slidepdf.com/reader/full/multivariable-control-and-real-time-optimization-an-industrial-practical 7/8

P RO C ES S A N D P L A N T O P T I M I Z A T I O N

On the other hand, when there is no active constraint, feed

allocation becomes a rigorous model optimization with an eco-

nomic objective function that minimizes energy consumption.

Column reboi ler duty In a d is t i l lat ion colum n with top

and bot tom products (for ins tance, s tabi l izer and crude pre-

flash), normally the effective manipulated variables are reflux

flowrate and r eboi ler duty , and the con trol le d variab le is aprope rty of th e top prod uct . In such a s i tuat ion, several pairs

of reflux flowrate and reboiler duty values satisfy the same top

product specificat ion.

The difference among those pairs of manipulated variables is

translated into larger or smaller distillate amounts. It becomes

clear that a point of economic opt imum must exis t . Operat ing

outside this point causes distillate product loss or excess energy

consum ption. This opt im um point can be determined by a r igor-

ous model and an RTO, but cannot be determined by MVC since

it does not consider the trade-off between energy consumption

and product fractionation.

F C C   converter: reaction temperature To keep a converters table, MVC normally constrains the operat ing react ion tem-

perature to a value in the range of 1°C to 2°C. This is practically

equivalent to keeping such a variable at a fixed setpoint.

This operating procedure is adopted because the MVC optimizer

tends to augm ent the conversion by increasing reaction temperature

up to the compressor limit. Nevertheless, the operating optimum

can occu r before such a constraint is activated. In this case, operat-

ing on the constraint (compressor limit) causes overcracking, result-

ing in inade quate yield profile or high olefin co ntent in the cracked

naphtha. Hence, determining the optimum reaction temperature

is an optimization problem w ith a rigorous model.

In practical operation, the values that bound the narrow reac-

tion temp erature operatin g range are calculated offline hy theprocess analysis enginee ring as a function of an econom ic ohjec-

tive and sent to the MVC to be sought.

R egen erator dense phase tem pe ratu re In the totalcombu stion FC C converter, the regenerator dense phase tempera -

ture is a consequence ofthe burned coke on the spent catalyst. The

regenerator control is perform ed through the excess of oxygen.

O s c a r R o t a v a   is  senior process control engineer for Petrobras.

He graduated as a chemical engineer at the Federal U niversity of

Rio de Janeiro and join ed P etrobras, Brazil s largest oil com pany,

the same year and worked in the training department, lecturing

on fluid mechanics (pipe and pump design), thermodynamics and

process control. Presently Dr. Rotava works w ith the op timization group, commission-

ing multivana bie process control systems on d istillation an d FCC units in several of th e

company s refineries. He is in charge of the corporate mass balance impleme ntation

program. Dr. Rotava got an MSc degree in chemical engineering from the Federal

University of Rio de Janeiro and a PhD degree in process control fro m the Im perial

College, London.

A n t o n i o C a rl o s Z a n i n   is a senior process con trol engineer

for Petrobras automation group. He received his BSc in chemical

engineering fro m the Federal University of Rio Grande do Sul. Dr,

Zanin holds an MSc (dissertation in predictive multivariabie control)

and PhD (thesis in real-time optimization] degrees in chemical

engineering fro m the University of Sao Paulo. He worked in developing Petrobras smultivariable control technology. Presently, Dr, Zanin is responsible for developing

and implementing inferential property algorithms, advanced control and real-time

In the part ial comhust ion FCC converter , the regenerator

dense phase temperature is related to the CO/CO2 ratio in fuel

gases and to coke conte nt in the regenerated catalyst.

In the latter case, regenerator temperature optimization must

consider the catalyst/oil ratio and the coke content on the regen-

erated catalyst, which affect the conversion in opposite ways. For

instance, a reduction in the regenerator temp erature increases thecatalyst/oil ratio and, consequently, the conversion. O n the oth er

han d, the coke conten t on the regenerated catalyst Increases, too.

Therefore, the combined action can decrease conversion. Balanc-

ing these effects can only he accomplished through an optimizer

with a rigorous model. HP

LITERATURE CITED

' Cutler, C. R. and R. T. Perry, Real time optimization with multivariablecontrol is tequired to maximize profits, omputers an d  hemical EngineerinV. 7, n. 5. pp, 663 -667 , 1983.

^ Marlin, E. T. and J. R Fotbes, Selecting the proper location for economicoptimization: multivariabie control or RTO , NPRA Com puter C onference,National Petroleum Refiners Association, paper CC-95-125, Nashville, Nov.6-8, 1995.

^ Hardin, M. B., R. Sharun, A. Joshi and J. D. Jones, Rigorous crude unitoptimization, NPRA Com puter Conference, National Petroleum RefinersAssociation, paper CC-95-122, Nashville, Nov. (i-8, 1995.

• Friedman,  Y. Z., Closed-loop optimization update—We are a step closer tofulfilling the dream, Hydrocarbon Processing HPIn Control, January 2000

'' White, D. C , Online optimization; what, where and estimating ROK Hydro

carbon Ib-ocessing pp . 43-51, June 1997.

PcTROL

www petrolab com

Page 8: Multivariable Control and Real-time Optimization - An Industrial Practical View (2005)

8/12/2019 Multivariable Control and Real-time Optimization - An Industrial Practical View (2005)

http://slidepdf.com/reader/full/multivariable-control-and-real-time-optimization-an-industrial-practical 8/8