You are on page 1of 21

Chemical Engineering Science 60 (2005) 4431 – 4451

www.elsevier.com/locate/ces

A systematic approach to the synthesis and design of flexible


site utility systems
Zhigang Shanga,∗ , Antonis Kokossisb
a Department of Process and Systems Engineering, School of Engineering, Cranfield University, Cranfield MK43 0AL, UK
b Department of Chemical and Process Engineering, University of Surrey, Guildford, Surrey GU2 7XH, UK

Received 25 May 2004; received in revised form 15 October 2004; accepted 7 March 2005

Abstract
The paper presents a systematic approach for the synthesis of flexible utility systems satisfying varying energy demands. The approach
combines benefits of total site analysis, thermodynamic analysis and mathematical optimisation. A thermodynamic efficiency curve (TEC)
is developed, which gives an overview of the maximum thermodynamic efficiencies of all possible design alternatives. TEC and hardware
composites guide the selection of candidate structures in the superstructure, excluding uneconomic options from the synthesis model.
The integration of thermodynamics yields significant reduction in the synthesis model, addresses the impact of variable loads on the unit
efficiencies, and enables a compact formulation of the design problem over long horizons of operation. The optimisation is formulated as
a multi-period MILP problem that relies on new target models to describe the performance of steam turbines, condensing turbines, gas
turbines and boilers. Target models account for the variation of efficiency with unit size, load and operating conditions in a simple, yet
accurate way. As a result, these models are capable of accounting for the efficiency trends of realistic units.
䉷 2005 Elsevier Ltd. All rights reserved.

Keywords: Optimisation; Modelling; Process design; Utility systems; Thermodynamic analysis; Operational variations

1. Introduction demands requires a methodology that builds flexible utility


systems which operate efficiently (thermodynamic target),
Industrial utility systems operate as part of a site complex are capable to adjust to different conditions (combinatorial
servicing a pre-specified number of production processes challenge), and able to operate at minimum cost (economic
each demanding heat and power at different levels through- target in optimisation). Critical design decisions include the
out the year. The utility system is configured according to selection of steam levels and the layout of the site utility
the needs of the entire system and it is desired that its oper- system. The system consists of available steam turbines, gas
ation remains efficient but robust to the different variations turbines, boilers and other auxiliary units.
in demand. At times, such variations could be considerable, The optimal selection of steam levels has been discussed
following changes in the markets, changes in the supplies of in Shang and Kokossis (2004). Once the steam levels are
raw materials, product variability and seasonal changes. The determined, the design can proceed with the development
variability has an apparent impact on the efficiency, as large of the best structure to produce utilities. This task comprises
units would be under-used in low periods whereas small a large combinatorial problem. Candidate systems involve
units, being less efficient at full load, would make unfortu- layouts of the following units: (i) simple and/or complex
nate choices in periods of high demand. The variability of back-pressure (BP) turbines, (ii) simple and/or complex con-
densing turbines, (iii) reheat cycles, (iv) simple and/or re-
generative gas turbine cycles, and (v) boiler networks. Each
∗ Corresponding author. Tel.: +44 1234 754655; fax: +44 1234 750728. alternative configuration results in a different overall effi-
E-mail address: z.shang@cranfield.ac.uk (Z. Shang). ciency and a different capital cost. As the utility demands
0009-2509/$ - see front matter 䉷 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2005.03.015
4432 Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451

vary with time, it is important that the utility system main- synthesis formulations in the form of multi-period MILP
tain high efficiency over the entire variation range. On the problems. These include boiler hardware models, steam tur-
other hand, the optimum trade-off between flexibility and bine hardware models and gas turbine hardware model. They
capital cost needs to be identified. In order to evaluate the account for the variability of equipment efficiency with size,
alternative design options and distinguish amongst the as- load and operating conditions. Thermodynamics are em-
sociated efficiencies, the effects of the unit size, as well as ployed to reduce the size of the superstructure and integrate
load and operating conditions on the unit efficiencies, need only units required to achieve maximum performance. The
to be taken into account. These effects generally involve approach is illustrated in several design problems most of
non-linear relations that give rise to complex models and which represent real-life applications.
formulations.
In the last two decades, a number of approaches have
been reported for the synthesis and design of utility systems. 2. Problem definition
Papoulias and Grossmann (1983) proposed an MILP ap-
proach for the synthesis of flexible utility systems accounting Given are
for anticipated variations in process demands in the shape of
a multiperiod utility demand pattern. Hui and Natori (1996) (i) a set of chemical processes whose requirements for
presented a mixed-integer formulation for multiperiod steam and power are addressed by a system of fixed
synthesis and operation planning for utility systems and steam levels;
discussed the industrial relevance. A simulated annealing (ii) a presumed horizon of operation;
algorithm was used by Maia and Qassim (1997) for the (iii) power and steam demands at each level;
synthesis of utility systems with variable utility demands. (iv) a set of units that could generate steam and power.
Iyer and Grossmann (1998) proposed a multi-period MILP
It is assumed that power can be generated by gas turbines,
approach for the synthesis and planning of utility sys-
steam turbines and diesel engines or by importing electricity
tems under multiple periods. Oliveira Francisco and Matos
from the utility grid. Boilers include
(2003) extended the model by Iyer and Grossmann to in-
clude global emissions of atmospheric pollutants from fuels (1) very high pressure (VHP) boilers fired by fuel;
combustion. (2) heat recovery steam generators (HRSG) which recover
It is well documented that the appropriate selection of heat contained in the gas turbine or the furnace exhaust
the different superstructure units/components reflects on gases (supplementary firing is allowed for these units);
the actual value of the synthesis work. Unless containing (3) waste heat boilers recovering heat from chemical pro-
appropriate candidates, the superstructure approach deliv- cesses;
ers inferior results. At the same time, exhaustive or blind (4) medium pressure boilers fired by fuel, which reheat
integration of potential units is pointless and ineffective. steam in reheat cycle;
Multiple operational scenarios simply increase the size of
the synthesis model, whereas efficiencies are complicated Gas turbines are available through the following configura-
enough to be treated as parameters by the majority of re- tions:
searchers. None of the previous approaches have addressed (1) simple gas turbine cycles, and
the challenge to select superstructure units as a result of (2) regenerative gas turbine cycles.
operational variations. This is particularly important as unit
sizes directly relate—and have a major impact—on the net- Steam turbines may include
work efficiency. At the same time, size and efficiency count
as complicating variables since, if introduced to the formula- (1) BP steam turbines.
tion, they both lead to significant complexities. Mavromatis (2) extraction BP steam turbines, and
and Kokossis (1998a, b) have presented an approach to de- (3) condensing turbines.
couple the non-linear components of unit efficiencies with
the use of thermodynamic models. They accordingly man- The design problem is to determine the structure of the site
aged to formulate an MILP model for the design of steam utility system that minimises the total cost so that to satisfy
turbine networks that takes into account the impact of op- utility demands across the selected operation horizon. The
erational variations on the efficiencies. Tested on real-life optimisation of the steam levels has been addressed previ-
problems, and although limited to BP turbines, the approach ously by Shang and Kokossis (2004). The horizon is com-
reports improvements of 11% against cases that neglect posed by a number of periods whose length may be different.
variations or consider unit efficiencies as constant. The boiler capacities should be determined by the optimi-
This paper presents a methodology inspired by these last sation, and their efficiencies depend on their capacity, heat
developments. The approach is not limited to particular units load and operating conditions. Steam is collected and dis-
though but includes boilers, gas turbines and all types of tributed to the chemical processes, the steam turbines, the
steam turbines. Thermodynamic models are used to develop reheat cycles or to the next low pressure steam level through
Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451 4433

letdown valves. Gas and steam turbine efficiencies are vari-


Total Site Analysis Design targets
able and also depend on their size, load and operating con-
ditions. Exhaust gases may be used by a HRSG to generate
Minimum set of primal design
steam. If a condensing turbine is selected, a condenser and TEC components
a vacuum header have to be selected. Steam can be returned
as condensate to be collected in a header of a given pressure.
Superstructure synthesis
Auxiliary units include deaerator and pumps.
Model development +
Mathematical Optimisation
Optimisation
3. Targeting models
Post-optimisation
As unit loads and operating conditions vary in time, so
do unit efficiencies. The unit efficiencies are also influenced Fig. 1. Schematic showing the configuration design optimisation strategy.
by their capacities. Most synthesis models simplify these
dependencies by setting constant efficiencies and making
use of linear mass and energy balances around the units. and condensing turbines further processed into complex
Instead, this paper makes use of hardware models to account turbines).
for the variation of efficiency with load and capacity, as well
The approach combines the benefits of total site analysis,
as changes in the operating conditions. They include a
thermodynamics and mathematical optimisation. Total site
analysis (Dhole and Linnhoff, 1992; Raissi, 1994) identifies
• turbine hardware model (THM);
acceptable candidates where TCE reduces the size of opti-
• boiler hardware model (BHM);
misation, using efficiency as a guide in the selection since
• condensing turbine hardware model (CTHM);
the energy cost of a utility system is an overwhelming factor
• gas turbine hardware model (GTHM).
in the analysis of each year’s cost. The strategy has been ap-
The THM was originally proposed by Mavromatis and plied successfully in utility system design (Chou and Shih,
Kokossis (1998a) to describe the operation of BP turbines. 1987) and heat exchanger network design (Linnhoff and
The BHM proposed by Shang and Kokossis (2004) is ex- Turner, 1981).
ploited to describe the performance of boilers. The CTHM
and GTHM are proposed in this paper to describe the oper-
ations of condensing turbines and gas turbines, respectively. 5. Total site analysis
The development of the two models is given in Appendix A
and B. Primal candidate structures include steam turbine cycles,
condensing turbine cycles, simple gas turbine cycles, regen-
erative gas turbine cycles, combined steam and gas turbine
cycles with or without condensing turbine, diesel drivers,
4. The optimisation strategy and all of their different combinations. These different types
need be considered for all different sizes, different oper-
The solution strategy is schematically shown in Fig. 1. ating conditions (primarily pressure levels), and different
The proposed strategy comprises the following five stages: combinations of the primal structures. Industries that use
a very large amount of thermal energy, such as the petro-
1. Total site analysis that determines energy integration and chemical, food processing and the paper and pulp (Chou
cogeneration targets; and Shih, 1987), enforce practices whereby heat require-
2. Analysis through the thermodynamic efficiency curve ments are satisfied first and power demands are matched ex-
(TEC) deployed to determine a minimum set of primal actly. The work follows an approach consistent with such
design components to ensure maximum efficiency; practices.
3. Superstructure development: Primal components are in- The site composite curve (SCC) proposed by Raissi
tegrated as a synthesis model with BP steam turbines, (1994) reflects on the integration of chemical production
condensing steam turbines, reheat cycles, gas turbine processes and the site utility system. The curves represent
network, boiler network and auxiliary units; steam flows in the utility system as well as the heat ex-
4. Mathematical optimisation with models that account for changed between site processes and utilities. The curves are
mass balances, energy balances and hardware (BHM, illustrated in Fig. 2. The shaded area between steam levels
THM, CTHM and the GTHM) and take the form of a is proportional to the potential for power cogeneration. The
multi-period MILP model; curves assess heat recovery, target requirements for fuel
5. Post-optimisation analysis (where component units are and cooling duties and identify acceptable design options.
further integrated into practical arrangements e.g. BP Fig. 2 illustrates such an option, mainly of BP turbines that
4434 Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451

VHP VHP

HP HP
Q1
MP
MP

Q2
LP
LP

Q3
CW

Fig. 2. Identification of steam turbines of a site by using the SCC. CW

Fig. 3. Total site profiles of a plant.

is developed with the use of the SCC. The sizes and posi-
tions of the turbines can also be identified using the curves.
Qsteam Qsteam
Qfuel
Qfuel
Boiler Boiler VHP
6. Thermodynamic analysis

The objective of the thermodynamic analysis is to screen W1


out the ineffective options and define a minimum set of op- HP
W2
tions with capabilities to achieve the targets set by the pre- T2 Q1
vious stage. The use of the set reduces dramatically the size
MP
and the complexity of the optimisation problem. As heat and
W3
power are energies of different quality, the thermodynamic T3
Q2
efficiency is defined as a relationship that determines the ra-
tio of the useful part of the energy to the total fuel input. LP
The thermodynamic efficiency takes the form:
Q3
 p
W+ Qi
t = , (1)
Q fuel
CW
 p
where W is the shaft-work generated; Qi the sum of Fig. 4. Integration of the BBPT cycle.
the steam heat loads required by the chemical processes at
the different operational levels; Qfuel the net fuel heat input.
6.1.1. Integration of the boiler and back-pressure steam
The thermodynamic efficiency indicates fuel utilisation effi-
turbine (BBPT) cycle
ciency. Thermodynamics are used to subsequently calculate
The steam crossing each expansion zone can be used by
such efficiencies for each utility structure. The result is the
BP steam turbines to generate power (Fig. 4). The VHP
composition of an efficiency curve that provides a ranking
steam is raised by the VHP boiler. The useful energy is the
for the feasible structures of the previous stage. On the ba-
power output and the steam heat loads to processes and the
sis of the energy and power demand, such a ranking is used
efficiency is given by
for the composition of the superstructure that is explained
in the next section.   p
Wi + Qi
t
BBPT
=  B , (2)
Qfuel,i
6.1. Thermodynamic efficiencies and utility structures
where Wi is the shaft-work generated by the steam turbine;
p
Fig. 3 illustrates the total site profiles of a plant. The ther- Qi the steam heat load of the steam turbine to processes;
modynamic efficiencies for each different utility structure QBfuel,i the net fuel heat input of the boiler. As shown in
are explained next. Fig. 4, the exhaust heat of BP steam turbines can be used as
Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451 4435

B
Ff Qfuel
Qwaste

Qsteam
WB WB

Fa WG

GT QG
VHP VHP
+
VHP +
HP

HP MP
WC
MP LP

LP CW

CW VAC

Fig. 5. Integration of the GTWB cycle. Fig. 6. Integration of the BCT cycle.

process heat, therefore the VHP steam load is where T out can be calculated by using Eq. (B.18) (see Ap-
  p pendix B). The steam load of the boiler is given by
QBsteam = Wi + Qi . (3)

Substituting Eq. (3) into Eq. (2) gives: QG = B


t Qwaste , (8)

QB where Bt is the thermodynamic efficiency of the waste heat


BBPT
t =  steam = B
t , (4) boiler.
QB fuel,i

where B
t is the thermodynamic efficiency of the boilers. 6.1.3. Integration of the boiler and condensing turbine
(BCT) cycle
6.1.2. Integration of the gas turbine and waste heat boiler The BCT cycle is shown in Fig. 6. As the exhaust heat
(GTWB) cycle of the condensing turbine is lost to cooling water, the useful
The GTWB cycle is shown in Fig. 5. The gas turbine is energy is the power output of the condensing turbine and
integrated with the waste heat boiler to generate power, and the efficiency is calculated by
the waste heat boiler is used to raise steam. It is assumed that
the gas turbine works at full load (i.e., maximum efficiency WC
BCT
t = , (9)
load). The useful energy is the power output of the gas tur- QB
fuel
bine and the steam load raised by the waste heat boiler. The
efficiency is calculated as where W C is the power output of the condensing turbine
and QB fuel is the fuel consumption. It is assumed that the
W G + QG
GTWB
t = , (5) condensing turbine works at full load. Hence, on the basis
F f Hf of the CTHM (see Appendix A), the relation between the
power output and the steam load crossing the condensing
where W G is the power output, QG the steam load generated
turbine is
in the waste heat boiler; F f the fuel consumption. On the
basis of the GTHM, the fuel consumption is given by 1
M= (Ac + B c W C ). (10)
1 H is
Ff = (Ag + B g W G ). (6)
 Hf The fuel requirement is given by
As presented by Cohen et al. (1987), the stack temperature
1
is assumed as 170 ◦ C. The useful waste heat from the gas QB
fuel = qM, (11)
turbine to the boiler is B
t
 
Qwaste = F f
1+
1
Cp g (T out − 170), (7) where q is the specific heat load of the steam and B
t is the
f thermodynamic efficiency of the boiler.
4436 Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451

VHP
Ff Qwaste
Qsteam
WB
Fa WG

GT HP
VHP VHP
+ MP
HP Qsurp
MP
MP
WC
LP
LP

CW
Wsurp
VAC

Fig. 7. Integration of the GTWBCT cycle. VAC

Fig. 8. Integration of the SCT.


Combining Eqs. (9)–(11), the thermodynamic efficiency
of the BCT cycle is calculated as
ηt BBPT
H is W C B
t GTWB
BCT
t = . (12)
q(Ac + B c W C ) GTWBCT

SCT BCT
6.1.4. Integration of the gas turbine, waste heat boiler and IP
condensing turbine (GTWBCT) cycle PSCT PBBPT PGTWB PGTWBCT PBCT PIP
The GTWBCT cycle is shown in Fig. 7. If the gas turbine
and the waste heat boiler cycle produce more steam than re- PA PB PC PD Power (MW)
quired, a condensing turbine is integrated to generate addi-
tional power by using the surplus heat. The thermodynamic Fig. 9. The TEC.
efficiency for the cycle is accordingly
WC + WG either buy W impt power or Qfuel fuel to generate power with
GTWBCT
t = , (13)
F f Hf the same cost). The thermal efficiency for generating power
using Qfuel fuel is equal to the power generated over Qfuel .
where W C and W G are the power outputs of the condensing Similarly, the imported power efficiency is defined as
turbine and the gas turbine respectively. W C is calculated by
applying the CTHM and W G is calculated by the GTHM. W impt CF
impt = = P, (15)
Qfuel C
6.1.5. Integration of the surplus steam condensing turbine
(SCT) where C F is the unit fuel price and C P is the unit power
The SCT is defined as the condensing turbine using sur- price. The concept provides the efficiency for importing
plus steam from chemical processes as shown in Fig. 8. Con- power and is used to compare operation with alternative
densing turbines are integrated to generate power with the structures.
surplus heat. As the SCT uses surplus heat, it is the one to
first consider for power and the efficiency is given by 6.2. Thermodynamic efficiency curve (TEC)
surp W surp
t = , (14) The TEC plots efficiencies vs. power demands for the dif-
Qsurp
ferent utility structures discussed earlier. Its graphical pre-
where W surp is the power output of the condensing turbine sentation is given in Fig. 9. Its first horizontal section ac-
and Qsurp is the surplus heat of the chemical processes. The counts for the units using surplus heat from the chemical
W surp can be calculated by using the CTHM. processes (SCT). The step length is determined by the max-
imum capacity of the units involved. From left to right—and
6.1.6. Import of power (IP) with the exception of its first section—the TEC expands to
The thermodynamic efficiency for importing power fol- sections with lower efficiency.
lows the top level analysis (Makwana, 1997) that assumes Once the energy requirements are determined (by the to-
that the cost of Qfuel fuel can buy W impt power (one could tal site analysis), the appropriate structures and sizes of the
Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451 4437

units to integrate become available from the curve. Structures


corresponding to higher demands than the ones required are
simply less efficient and are excluded. Fig. 9 explains the
case for a power requirement PA ; the SCT and the BBPT
cycles are the two structures sensible to integrate and scope;
the other options are excluded as inefficient. If the power
requirement increases from PA to PB , the additional consid-
erations would include a gas turbine and waste boiler cycle.
Unit sizes are not determined uniquely though. The selec-
tion of sizes to integrate follows Mavromatis and Kokossis
(1998a). Applied to our example, the power capacity of the
gas turbine is equal to the difference between the power re-
quirement and PA . Additional power requirement from PB Fig. 10. Complex turbines are considered as a cascade of simple turbines.
to PC introduces the GTWBCT cycle. The power capacity
of the cycle is equal to the difference between the power
demand and PB . Extending the power requirement from PC capital cost of a series of simple turbines is more expensive
to PD includes the option for the BCT cycle. The capacity than a complex/multistage turbine with the same capacity.
in power of the cycle is equal to the difference between the Economic benefits to further integrate simple turbines into
power requirement and PC . Beyond the power requirement pass-out and complex turbines are addressed in the post-
PD , the residual power has to be imported from the utility optimisation stage.
grid. The sizes of component turbines for each scenario are
The TEC is a screening tool that is based on thermody- determined at the thermodynamic analysis stage. For mul-
namics and, more specifically, on energy efficiency and in- tiple operation scenarios the number and sizes of simple
tegration. Extensions to TEC may include exergy efficien- component cylinders are identified for each expansion zone
cies, and/or the adoption of weights for the different types by using the discretisation method proposed by Mavromatis
of fuel and power. Such extensions are straightforward and and Kokossis (1998b). The method suggests that turbines
do not alter the concept behind TEC, which is to use ther- are sized to match the loads of every scenario as well as all
modynamics to simplify the optimisation without forfeiting their possible combinations. The case of two scenarios with
on important design decisions. Whether exergy or energy ef- the steam flows across an expansion zone is illustrated in
ficiencies make better options is not apparently part of this Fig. 12. The first combination involves the selection of tur-
work. Similarly, the screening could be relaxed to include bine T1 sized to the scenario B, while operating at part load
one more (or two more structures) than the ones dictated by under scenario A. This option features the lowest capital
TEC. This could be done in the spirit that a number of other cost, but lower part load efficiency for scenario A. Alterna-
operational features (e.g. flexibility, dynamics, etc.) are not tively, turbine T2 can be installed to size scenario A. This
addressed at this early stage of design. option achieves the highest overall efficiency but requires
the highest capital cost. In option 3, turbine T2 can be se-
lected, along with turbine T3 sized to take up the remaining
7. Superstructure development load for scenario B. The efficiency for scenario B will be
smaller than the first two options, but achieves the highest
For each operation scenario the TEC is constructed and efficiency for scenario A and requires lower capital cost than
the curves are applied to identify candidate structures and the second option.
potential capacities of the utility units. The steps to generate
the superstructure are presented as follows. 7.2. Superset of gas turbines

7.1. Superset of BP steam turbines The capacity of the gas turbine for each scenario is deter-
mined by using the TEC. For multiple operation scenarios
As discussed by Mavromatis and Kokossis (1998b), both the number, sizes and types of candidate gas turbines of the
complex turbines and multistage turbines are equivalent to a superset depend on the specific problem, as explained in the
cascade of simple turbines, each taking up potential from a following section.
single expansion zone, as shown in Fig. 10. On the grounds
of the equivalence, all possible combinations of turbine lay- 7.2.1. Types of gas turbines in the superset
outs can be reduced to a single superset of component cylin- The types of the gas turbine cycles are concerned with
ders as illustrated by Fig. 11. This superset of design com- simple and regenerative gas turbine cycles. The major differ-
ponents is adequate to achieve the targets expected by the ence between the simple and regenerative gas turbine cycles
BP steam turbine network. The assumption is apparently is the addition of a recuperator for heat exchange between
correct from the thermodynamic point of view, as the total the turbine outlet and the compressor outlet as shown in
4438 Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451

Fig. 11. Decomposition of complex steam turbines.

Scenario B 0.5
Scenario A 40MW
HP HP 60MW
0.4
T1 T2 T3 20MW
0.3

ηe
MP MP 0.2
T2
T1
T3 Superset of BP steam turbines 0.1

0
η η η 0 10 20 30 40 50 60
W (MW)
T1 T2 T2
T1 T3
Fig. 14. The effect of part load operation prevails over the increase of
Q efficiency with gas turbine size.
Q Q
Option 1 Option 2 Option 3

Fig. 12. The candidate BP steam turbines for the case of two scenarios.
7.2.2. Number and sizes of gas turbines in the superset
The GTHM relates the power output to the fuel load and
the size of the gas turbine as shown in Eq. (B.17) in Ap-
pendix B. As both the load and size of the unit need to
Heat-exchanger be optimised, straightforward modelling would result in an
MINLP formulation. The discretisation method enables in-
stead the formulation of an MILP problem. The economic
Fuel analysis of the gas turbine operation suggests the discrete
Air sizes and number of the candidate gas turbines. As elec-
trical efficiency increases with size but decreases with par-
Power tial load, the highest efficiency is obtained for gas turbines
Compressor Turbine
sized and operating at full load. This is shown in Fig. 14.
Fig. 13. The regenerative gas turbine. In terms of efficiency, the optimum size for each scenario
exactly matches its power demand. Following the discretisa-
tion scheme used for BP steam turbines, a similar discreti-
sation method is proposed whereby gas turbines are sized
Table 1
to match the power demand of every scenario as well as all
P /H characteristics of gas turbine cycles
their possible combinations.
Gas turbine cycle P /H ratio

Simple gas turbine cycle 0.65 7.3. Superset of boilers


Regenerative gas turbine cycle 0.85
The BHM (Shang and Kokossis, 2004) is used to describe
the performance of each fired boiler and waste heat boiler.
The design model is given by
Fig. 13. Following Chou and Shih (1987), the types of the
Qfuel = (Cp Tsat + q)((1 + b)M + aM max ), (16)
gas turbines are screened by the characteristic value of power
to heat ratio, P /H . Table 1 shows the ratio corresponding to where Qfuel is the fuel requirement; Cp the specific heat
each gas turbine cycle. of boiler water; Tsat temperature difference between the
Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451 4439

Waste heat namic analysis, rather than for the exhaustive structures. The
binary variables account for the selection of units and their
Fuel Fuel Fuel operation status at each scenario. The continuous variables
relate to the stream flowrates (steam, fuel), the power out-
Water
VHP puts and the operating and capital costs.
The optimisation problem involves the following defini-
Fig. 15. Superset of boilers. tions for sets, parameters and variables:

Sets
saturation temperature of the steam generated in the boiler
and the temperature of the boiler inlet water; q the specific I B = {ib|candidate boilers},
heat load of the steam generated in the boiler; M the steam I T = {it|candidate BP steam turbines},
load; M max the maximum steam load; a and b the regression I C = {ic|candidate condensing turbines},
parameters. The above expression relates to the fuel flowrate V C = {vc|candidate VHP condensing turbines},
with the steam load and the boiler size. The superset of the P = {pi|power generation units},
boilers is shown in Fig. 15. The waste heat is the waste heat I = {i|selected units},
from gas turbine cycles. The number, sizes and fuel require- I G = {ig|candidate gas turbines},
ments of the boilers are determined by the optimisation. K = {k|operation scenarios},
Z = {z|expansion zones}.
7.4. VHP condensing steam turbines, surplus steam
condensing turbines, and reheat cycles Parameters

For each scenario, the power output of the VHP condens- a, b regression parameters of the BHM
ing turbine is determined by the TECs. For a single scenario, Aic , Bic regression parameters of the CTHM
the optimum size matches exactly the power demand. For Ag , B g regression parameters of the GTHM
multiple scenarios, the number and sizes of candidate con- Az , Bz regression parameters of the THM for BP steam
densing turbines of the superset for multiple operation sce- turbines of expansion zone z
narios are determined by the discretisation scheme followed Cp specific heat of boiler water
for the gas turbines. Cpg specific heat of gas turbine flue gas
For each steam level, the surplus heat of the processes is Cpa specific heat of gas turbine combustion air
obtained by total site analysis. The number and sizes of the Cpf specific heat of gas turbine fuel
surplus condensing turbines are determined in a similar way EI S ic isentropic enthalpy change of condensing tur-
as VHP condensing steam turbines. For multiple operation bine ic
scenarios, the surplus steam condensing turbines are sized EI S z isentropic enthalpy change of BP steam turbines
to match the demands of each individual scenario as well as of expansion zone z
f,max
all their different combinations. Fig maximum fuel load of gas turbine ig
In a reheat cycle, steam is first expanded to some interme- H operating hours per year
diate pressure and then reheated in the boiler. It next expands Hf specific fuel combustion heat
in the turbine to the exhaust pressure. Following Chou and LB lower bound of boiler capacities
Shih (1987), the reheat cycle can improve the overall ther- BT ,max
Mib maximum steam load of boiler ib
mal efficiency only if the thermal efficiency contributed by BT ,max
the reheat part is greater than that of the remaining parts. The Mz,it maximum steam load of BP steam turbine it of
required large heat-exchange area and the increased com- expansion zone z
CT ,max
plexity in system design detract from the gain in efficiency Mic maximum steam load of condensing turbine ic
c
Mz,k total steam load across each expansion zone z
due to reheating. The reheat cycle, therefore, will be of in-
terest to site utility system design only when a lot of heat is under scenario k
exhausted to cooling water. qib specific heat load of the steam generated in
boiler ib
Ta temperature of inlet air of gas turbines
8. Optimisation model Tf temperature of gas turbine fuel
Tib,k
sat temperature difference between the saturation
In this section, a multi-period MILP model is presented temperature of the steam generated in the boiler
for the minimisation of capital investment and operating and the temperature of the boiler inlet water
cost. The model incorporates the BHM, THM, CTHM and TkS time fraction of scenario k
GTHM models. The optimisation is a screening tool for the UB upper bound of boiler capacities
f
selected alternative design options by using the thermody- Uk unit cost of fuel for boilers under scenario k
4440 Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451

GT ,f
Uk unit cost of fuel for gas turbines under scenario The above bilinear model is replaced by the following mixed
k integer linear model:
Wkdem power demand of site processes under scenario B,f
k Qib,k + QB,w
ib,k = (Cp Tk + q)
sat

B
Binary variables ((1 + b)Mib,k + aX B
ib,k ), ib ∈ I B, k ∈ K. (18)

B,e Constraints include


yib integers to denote the selection of the boilers
B,o B,max B,o
yib,k integers to denote the operation of the boilers Mib − U B (1 − yib,k )
BT ,e
yz,it integers to denote the selection of the BP steam B B,max B,o
 Xib,k  Mib − LB (1 − yib,k ), ib ∈ I B, k ∈ K.
turbines
BT ,o (19)
yz,it,k integers to denote the operation of the BP steam
turbines B B,max
CT ,e
The above logical constraint denotes Xib,k equals Mib
yic integers to denote the selection of the condensing while the boiler is operating
turbines
CT ,o B,o B,o
yic,k integers to denote the operation of the condensing LB yib,k B
 Xib,k  U B yib,k ,
turbines ib, ∈ I B, k ∈ K. (20)
GT ,e
yig integers to denote the selection of the gas turbines
GT ,o B = 0 while the
The above logical constraint denotes Xib,k
yig,k integers to denote the operation of the gas turbines
boiler is not in operating status.
Continuous variables B,max B,e B
Mib − yib U  0, ib ∈ I B, (21)
C B,f total annual fuel cost of all boilers B,max B,e B
C GT ,f total annual fuel cost of all gas turbines Mib − yib L  0, ib ∈ I B. (22)
C c,tot total capital cost of all the selected units
The above logical constraints denote the boiler capacity
Cic capital cost for each unit
should be zero if it is not selected and the boiler capacity
C tot total annual cost
f should be larger than its lower bound and smaller than its
Fig,k fuel load of gas turbine ig under scenario k upper bound.
B
Mib,k steam load of boiler ib under scenario k
BT B B,o B
Mz,it,k steam load of BP steam turbine it under scenario k Mib,k − yib U  0, ib ∈ I B, k ∈ K. (23)
CT
Mic,k steam load of condensing turbine ic under
scenario k The above logical constraint denotes the load of the boiler
CT
Mvc,k steam load of VHP condensing turbine vc under should be zero if it is off
scenario k B B,max
t Mib,k − Mib  0, ib ∈ I B, k ∈ K. (24)
Mz,k amount of steam throttled through the let down
valve of expansion zone z under scenario k The above constraint denotes the load of the boiler should
B,f
Qib,k fired fuel load of boiler ib under scenario k not be larger than its capacity
QB,w
ib,k waste heat load from gas turbines to boiler ib in
B,o B,e
scenario k yib,k − yib  0, ib ∈ I B, k ∈ K. (25)
QGT
ig,k
,w
waste heat load of gas turbine ig under scenario k
BT The above logical constraint denotes the boiler cannot op-
Wz,it,k power output of BP steam turbine it of zone z under
erate if it is not selected.
scenario k
Buy (b) BP steam turbines: The THM applied for the power
Wk power import under scenario k output of a BP steam turbine in zone z under scenario k
CT
Wic,k power output of condensing turbine ic under sce- yields
nario k  
GT
Wig,k power output of gas turbine ig under scenario k BT 6 1 Az
Wz,it,k = EI S z − BT ,max
5 Bz Mz,it
Given the parameters, sets and variables above, the design  
model includes consideration for the following models: BT 1 BT ,max BT ,o
× Mz,it,k − Mz,it yz,it,k ,
(a) Boilers: The BHM yields 6
z ∈ Z, it ∈ I T , k ∈ K. (26)
B,f
Qib,k + QB,w
ib,k = (Cp Tib,k + qib )
sat If the turbine is not selected, it cannot operate:
B
((1 + b)Mib,k + ay B,o B,max
ib,k Mib,k ), ib ∈ I B, k ∈ K. (17) BT ,o
yz,it,k BT ,e
− yz,it  0, z ∈ Z, it ∈ I T , k ∈ K. (27)
Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451 4441

The load of the steam turbine should not be larger than its valves (in case the installation of a turbine is not cost effec-
capacity: tive). The mass balances give
BT ,o BT ,max

BT BT t c
Mz,it,k − yz,it,k Mz,it  0, z ∈ Z, it ∈ I T , k ∈ K Mz,it,k + Mz,k = Mz,k , z ∈ Z, k ∈ K. (36)
(28) it∈I T

(c) Condensing steam turbines: The CTHM applied for The VHP steam requirement is equal to the sum of the
the power output of a condensing turbine under scenario k amount of steam across the first expansion zone plus the
yields: amount of the steam through the VHP condensing turbines.
   
B c CT
6 1 Aic Mib,k = M1,k + Mvc,k , k ∈ K. (37)
CT
Wic,k = EI S ic − CT ,max ib∈I B vc∈V C
5 Bic Mic
  (f) Power balance: The electricity balance under scenario
CT 1 CT ,max CT ,o
× Mic,k − Mic yit,k , k is expressed as
6

ic ∈ I C, k ∈ K. (29) buy
Wpi,k + Wk = Wkdem , k ∈ K. (38)
If the condensing turbine is not selected, it cannot operate: pi∈P

CT ,o CT ,e (g) Costs: The total annual fuel cost of the boilers is


yic,k − yic  0, ic ∈ I C, k ∈ K. (30)
 f B,f
The load of the condensing turbine should not be larger than C B,f = Uk Qib,k TkS H . (39)
its capacity: k∈K
ib ∈ I B
CT CT ,o CT ,max
Mic,k − yic,k Mic  0, ic ∈ I C, k ∈ K. (31)
The total annual fuel cost of the gas turbines is
(d) Gas turbines: The GTHM applied for the power output  GT ,f f
of a gas turbine under scenario k yields: C GT ,f = Uk Fig,k TkS H . (40)
  k∈K
1 Ag ig ∈ I G
GT
Wig,k = g Hf − f,max 
B F The total capital cost incurred for the installation of the
ig
f f,max GT ,o
equipment is
× ((1+n)Fig,k −nF ig yig,k ), ig∈I G, k∈K, 
(32) C c,tot = Cic . (41)
i∈I
f,max
where the maximum fuel flowrate Fig can be calculated
The capital costs of the units are calculated using the func-
by Eq. (B.14) ) (see Appendix B).
tions presented by Bruno et al. (1998). They are given in
If the gas turbine is not selected, it cannot operate:
Table 2 along with linearized expressions for the boiler cost.
GT ,o GT ,e It is important to note that the proposed discretisation meth-
yig,k − yig  0, ig ∈ I G, k ∈ K. (33)
ods treat the capacities of the steam turbines, the gas turbines
The load of the gas turbine should not be larger than its and the electric generators as parameters in the optimisa-
capacity: tion model. Therefore, the nonlinear capital cost functions
for these units are applied.
f GT ,o f,max
Fig,k − yig,k Fig  0, ig ∈ I G, k ∈ K. (34) (h) Objective function: The objective function minimises
the total annual cost:
The waste heat from the gas turbine is given by
 min C tot = C B,f + C GT ,f + C c,tot . (42)

QGT ,w  1 Cpa T a + Cpf T f + Hf


ig,k =
The total annual cost consists of the capital cost and the fuel
f cost. The optimisation model consists of linear constraints
  and integer variables, and comprises a multi-period MILP
1 A g
f
−(1 + n) g Hf − f,max  Fig,k , model. The structure and the operation strategy are opti-
B F mised to minimise the total cost consisting of capital cost
ig
and operating cost. The development of the MILP model
ig ∈ I G, k ∈ K. (35)
requires the following information:
(e) Steam mass balances: The mass balance across each
expansion zone z for scenario k involves the steam through • steam level specifications;
the turbines and the steam throttled through the let down • data on total site profiles for each scenario;
4442 Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451

Table 2
Capital cost data (Bruno et al., 1998)

Unit Type of cost function Investment cost ($/year)

Large package boiler Nonlinear 4954F 0.77 fp2


F : steam flowrate (t/h) fp 2 = 1.3794 − 0.5438P + 0.1879P 2
P : Pressure (MPa) Linear (P = 9 MPa) 495, 384 + 13, 861F
Heat Recovery Boiler Nonlinear 941Ffg0.75
Ffg: flue gas flowrate (t/h) Linear 6996+211.5Ffg
Steam turbine Nonlinear 2237Wst 0.41
Wst: power (kW)
Gas turbine Nonlinear 952Wgt 0.76
Wgt: power (kW)
Electric generator Nonlinear 176Weg0.49
Weg: power (kW)
Deaerator Nonlinear 904FB0.62
FB : BFW flowrate (t/h)

• power demand for each scenario; Table 3


• cost correlations for the utilities; Summary of operating conditions
• capital cost correlations for the units. Unit Operating conditions

Even though the primitive approach to the problem yields an VHP header Saturation temperature: 303 ◦ C
HP header Saturation temperature: 275 ◦ C
MINLP formulation with a very large number of variables,
MP header Saturation temperature: 210 ◦ C
the use of total site analysis and the TCE reduces it into a LP header Saturation temperature: 140 ◦ C
reasonably sized MILP. The optimisation yields a layout of Vacuum header Temperature: 60 ◦ C, Pressure: 0.02 MPa
simple turbines that are post-processed to synthesise com- Deaerator Vent ratio: 0.0015
plex or multi-stage turbines. For two cylinders to merge into
a complex unit, they both have to be loaded during the same
scenario. Depending on whether the steam flow through the Table 4
upper cylinders is larger or smaller than that in the lower Utility data
sections, the turbines can be of an extraction or induced type. Demineralized water Fuel (natural gas) Electricity

Temperature: 27 ◦ C LHV: 13856 kW h/t


Cost: 0.24$/t Cost: 223$/t Cost: 0.1$/kW h
9. Case studies

When considering different operating scenarios, Shang the aim is to find the optimal configuration of the site utility
and Kokossis (2004) presented an approach to optimise the system that satisfies the utility demands and minimises the
steam levels of a total site. Having specified the optimal annual total cost.
steam levels, the optimal configuration of the site utility sys-
tem is determined with the approach presented in this work. 9.1.1. Totals site analysis
Two case studies are selected to illustrate the capabilities of The SUGCCs are given in Fig. 16. The steam across
the methodology. The operating conditions of the four steam each expansion zone is obtained in Table 6. LP steam is in
levels, the vacuum header and the deaerator are shown in surplus and used by the condensing turbines. BP turbines
Table 3. The pressures for VHP, HP, MP and LP are, re- are installed in the steam expansion zones of VHP–HP and
spectively, 9, 6, 1.9, 0.36 MPa. The steam used for heating HP–MP. The power outputs for the possible SCT and BBPT
returns as condensate. The case studies feature different util- cycles are calculated using CTHM and THM models based
ity demands. The utility costs are shown in Table 4 and the on the inlet and outlet steam conditions and the resulting
capital costs are presented in Table 2. values are shown in Table 7.

9.1. Case study 1 9.1.2. Thermodynamic analysis


The characteristic values of power to heat ratio for the
The site utility grand composite curves (SUGCCs) of Fig. three scenarios are presented in Table 8. The heat demand
16 reflect the steam demand/generation of the site under is the total demand of HP, MP and LP. The P /H values are
three operation scenarios. The power demands for the sce- lower than the characteristic value of the simple gas turbine
narios are given in Table 5. On the basis of the SUGCCs, cycle. A simple gas turbine cycle is then employed instead
Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451 4443

T T Scenario B T Scenario C
Scenario A 220 260
180 VHP VHP
VHP 40 40
HP HP HP
50 MP MP
MP 300
130 50 260 60
LP 30 LP LP
VAC VAC VAC

H(t/h) H(t/h)
H(t/h)

Fig. 16. The SUGCCs of a site.

Table 5 9.1.3. Generation of the superstructure


Power demands of Case 1 For each different scenario, the amount of steam expanded
Scenario A Scenario B Scenario C in each zone determines the candidate sizes for the BP steam
turbines shown in Table 10. The surplus heat provided by the
Power demand (MW) 22 35 42
processes determines the candidate sizes of the condensing
turbines given in Table 11. Simple gas turbines are used. Ex-
pressed in terms of power capacities, the candidate sizes are
given in Table 12. VHP boilers are fired by fuel and heated
Table 6
Steam amount across each expansion zone of Case 1 (t/h)
by the waste heat available from the gas turbine cycles. Two
VHP (B1 and B2) candidate boilers are included in the su-
Scenario A Scenario B Scenario C perstructure, as for multiple scenarios it might prove more
VHP–HP 180 220 260 economic to operate multiple small boilers than a single big
HP–MP 130 260 300 boiler. An HP waste heat boiler and an MP waste heat boiler
MP–LP recover surplus heat from the processes. Without heat ex-
LP–Vacuum header 30 50 60 hausted to cooling water, the reheat cycle is excluded as an
option. The generated superstructure is shown in Fig. 18.

9.1.4. Optimisation
Table 7
The superstructure is formulated as an MILP model. The
Power outputs of possible SCT and BBPT cycles of Case 1
optimisation minimises the total annual cost. The model is
Scenario A Scenario B Scenario C developed using GAMS and the optimisation is conducted
SCT (MW) 2.5 4.1 5 by employing the OSL solver. The model involves 195 con-
BBPT (MW) 13 22.9 26.7 tinuous variables, 92 binary variables and 242 constraints.
The optimum configuration is given in Fig. 19. The se-
lected units include three BP steam turbines, one condens-
ing turbine, one gas turbine, a VHP boiler, an HP waste
Table 8 heat boiler, an MP waste heat boiler and the deaerator. The
P /H characteristics of Case 1 HP and MP waste heat boilers are selected to produce HP
Scenario A Scenario B Scenario C
and MP steam, respectively. One of the BP turbines oper-
ates between the VHP and the HP levels. The other two
P /H ratio 0.22 0.285 0.29 BP turbines operate between the HP and MP levels. They
supply power by exploiting the cogeneration potential. The
condensing turbine is employed to generate power by using
surplus heat from the site processes. The BP turbines and
of a regenerative gas turbine cycle. By calculating the ef- the condensing turbine do not meet the power demands. The
ficiency and the maximum power output for each potential gas turbine is installed to supply the remaining power. The
cycle, the TEC is constructed for each one of the three op- total annual cost is 33.341 mUS$. The annual fuel cost is
eration scenarios as shown in Fig. 17. On the basis of the 27.718 mUS$ and the annual capital cost is 5.623 mUS$.
TECs and the utility demands, the promising candidate util- The capacities of the selected units are given in Fig. 19.
ity structures include the BBPT cycles, the SCT cycles and The optimal loads of the units are obtained under each
the GTWB cycles; all other design options are excluded as scenario. These are given in Table 13. The BP turbine BT1
less efficient. By using the TECs, the power outputs of the (HP–MP) operates during A, but not during B and C. Turbine
GTWB cycles are obtained for all the scenarios as shown in BT3 (HP–MP) operates during B and C, but not during
Table 9. A. During A turbine BT1 (HP–MP) is more efficient than
4444 Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451

Scenario A ηe Scenario B
ηe BBPT
BBPT GTWB
GTWB

GTWBCT GTWBCT
BCT BCT
Simple GT Simple GT
IP
SCT IP SCT

2.5 13MW 6.5MW 4.1MW 22.9MW 8MW

P (MW) P (MW)

ηe Scenario C
BBPT
GTWB

GTWBCT
BCT
Simple GT
IP
SCT

5MW 26.7MW 10.3MW

P (MW)

Fig. 17. TECs for Case 1.

Table 9 As shown in Table 14, no steam turbines have the same


Power outputs of the GTWB cycles of Case 1 operation schedule. Therefore, no complex turbines can be
Scenario A Scenario B Scenario C synthesized.
Power output (MW) 6.51 7.98 10.32 9.2. Case study 2
The SUGCCs account for the steam demand/generation
of the site under the three operation scenarios of Case 1.
Table 10
Candidate sizes of BP turbines of Case 1 (t/h)
These are shown in Fig. 16. The power demands for this
case are given in Table 14.
BT1 BT2 BT3 BT4 BT5 Based on the TECs of Fig. 17 and the power demands,
VHP–HP 180 220 260 40 80 the candidate utility structures include the BBPT cycles and
HP–MP 130 260 300 40 170 the SCT cycles. Since the power demands for all three sce-
narios can be totally satisfied by these two cycles, the other
power units are thus excluded because of their lower thermal
efficiencies. The superstructure is shown in Fig. 20 and is
Table 11 smaller than the one studied in Case 1. The candidate sizes
Candidate sizes of condensing turbines of Case 1 (t/h)
of the turbines are apparently the same.
CT1 CT2 CT3 CT4 CT5 The MILP model is optimised against the total annual
cost. The model involves 97 continuous variables, 60 binary
Size (t/h) 30 50 60 10 20
variables and 157 constraints.
The optimum configuration is shown in Fig. 21. It in-
cludes two BP steam turbines, one condensing turbine, one
Table 12 VHP boiler, one HP waste heat boiler, one MP waste heat
Candidate sizes of gas turbines of Case 1 (MW) boiler and the deaerator. One BP turbine operates between
the VHP and the HP steam levels. Another one operates be-
GT1 GT2 GT3 GT4 GT5 GT6
tween the HP and the MP steam levels. The condensing tur-
Capacity (MW) 6.51 7.98 10.32 1.47 3.81 2.34 bine generates power using surplus heat from the site pro-
cesses. There is no need for a gas turbine. The total annual
cost is 30.751 mUS$. It is lower than Case 1 because of the
lower power demand. The annual fuel cost is 26.325 mUS$
and the annual capital cost is 4.426 mUS$. The capacities
turbine BT3 (HP–MP) and during B and C turbine BT3 of the units are given in Fig. 21. The optimal loads for each
(HP–MP) is more efficient than turbine BT1 (HP–MP). The scenario are given in Table 15. The BP steam turbines BT2
HP waste heat boiler shuts down during B and C because (VHP–HP) and BT3 (HP–MP) maintain identical operation
there is no waste heat during these periods. All other units schedules as shown in Table 15. Therefore, the two turbines
operate for all three scenarios. can be synthesized as a single complex turbine.
Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451 4445

GT1 GT2 GT3 GT4 GT5 GT6

B1 B2
VHP
BT1 BT2 BT3 BT4 BT5

HP
BT1 BT2 BT3 BT4 BT5
HPWHB
MP

LP

CT1 CT2 CT3 CT4 CT5


LPWHB LP MP HP
CONS. CONS. CONS.

VAC.

DW COND.

Deaerator

Fig. 18. Superstructure of Case 1.

Max 10.32MW
GT3

Max 260t/h

B1
VHP
BT3 Max 6.91MW
Max 50t/h
HP
BT1 Max 8.38MW BT3 Max 19.78MW
HPWHB
MP

Max 60t/h
LP

LPWHB CT3 Max 4.99MW LP MP HP


CONS. CONS. CONS.

Max 360t/h VAC.

DW COND.

Deaerator

Fig. 19. Optimal structure of Case 1.


4446 Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451

Table 13 Table 16 compares the size of the formulation with and


Optimal loads of the units of Case 1 without the thermodynamic screening. Assuming a discreti-
Unit Scenario A Scenario B Scenario C sation across the power range of Case 1 (2–42 MW; incre-
ment 0.5 MW), the model without screening requires 1628
BT3 (VHP–HP) 180 t/h 220 t/h 260 t/h
binary variables and 2697 continuous variables. With the
(steam load/ power output) 4.36 MW 5.64 MW 6.91 MW
BT1 (HP–MP) 130 t/h 0 t/h 0 t/h proposed approach the model requires only 92 binary vari-
(steam load/ power output) 8.38 MW 0 MW 0 MW ables and 195 continuous. A similar comparison for Case 2
BT3 (HP–MP) 0 t/h 260 t/h 300 t/h (range: 2–30 MW) yields a problem that, without screening,
(steam load/ power output) 0 MW 16.61 MW 19.78 MW would require 1140 binary variables and 1905 continuous
CT3 30 t/h 50 t/h 60 t/h
(Table 17). Instead, the proposed model requires 60 binary
(steam load/ power output) 2.0 MW 4.0 MW 5.0 MW
GT3 (power output) 7.26 MW 8.76 MW 10.32 MW variables and 97 continuous variables.
VHP B1 (steam load) 180 t/h 220 t/h 260 t/h The three scenarios discussed in the two case studies re-
HP WB (steam load) 50 t/h 0 t/h 0 t/h flect on the real-life dimensions of the industrial problems.
LP WB (steam load) 30 t/h 50 t/h 60 t/h However, the number of design candidates may become
Deaerator (water load) 210.3 t/h 310.5 t/h 360.5 t/h
large for cases of many operation scenarios. For a problem
less than eight scenarios, it is not a serious burden on an
MILP formulation. For cases of large number of scenarios,
Table 14 a range of unit sizes at constant capacity increments can be
Power demands of Case 2 used to alternatively develop superstructures or, as a more
Scenario A Scenario B Scenario C practical approach, the sizes of these units could be those of
the standard units available in the market, within the range
Power demand (MW) 12 25 30
of capacities involved in the specific problem.

9.3. Remarks and discussion 10. Conclusions

Although the SUGCCs of the two studied cases are the A systematic methodology is presented for the optimal
same, the optimal configurations of the site utility systems design of flexible site utility systems. The methodology
are different because of the different power demands. The combines the benefits of total site analysis, thermodynamic
studies imply that different P/H ratios assume different op- analysis and optimisation techniques. The approach ac-
timal structures. counts for the interactions between the site utility systems
A reduced superstructure is obtained by using thermody- and the site processes. The design task is addressed in view
namic analysis to screen out inefficient options. The reduc- of the anticipated variations in the process demands and the
tion is significant as it further replaces the need to solve the effect of the unit capacities and varying loads on the efficien-
MINLP with a need to solve an MILP problem instead. cies of the selected units. These aspects normally give rise to

B1 B2
VHP

BT1 BT2 BT3 BT4 BT5

HP
BT1 BT2 BT3 BT4 BT5
HPWHB
MP

LP

CT1 CT2 CT3 CT4 CT5


LPWHB LP MP HP
CONS. CONS. CONS.

VAC.

DW COND.

Deaerator

Fig. 20. Superstructure of Case 2.


Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451 4447

Max 260t/h

B1
VHP
BT2 Max 5.78MW

Max 50t/h
HP
BT3 Max19.78MW
HPWHB
MP

Max 60t/h
LP

CT3 Max 5.0MW


LPWHB LP MP HP
CONS. CONS. CONS.

VAC.

DW COND.

Deaerator

Fig. 21. Optimal structure of Case 2.

Table 15 The total site analysis is employed to screen and identify all
Optimal loads of the units of Case 2 possible design options. A thermodynamic curve is proposed
Unit Scenario A Scenario B Scenario C in this approach. It is a useful tool to identify the promising
candidate design options to be included in the superstructure
BT2 (VHP–HP) 180 t/h 220 t/h 220 t/h
by screening among all alternative design options. The size
(steam load/ power output) 4.52 MW 5.78 MW 5.78 MW
BT3 (HP–MP) 119 t/h 243 t/h 300 t/h of the optimisation problem can be reduced by screening
(steam load/ power output) 5.49 MW 15.23 MW 19.78 MW out the uneconomic design options. The proposed approach
CT3 30 t/h 50 t/h 60 t/h has been illustrated with two case studies. The optimisation
(steam load/ power output) 2.0 MW 4.0 MW 5.0 MW problems of the two cases have been reduced significantly
VHP B1 (steam load) 180 t/h 220 t/h 260 t/h
compared to the full superstructure approach. As a result, the
HP WB (steam load) 50 t/h 0 t/h 0 t/h
LP WB (steam load) 30 t/h 50 t/h 60 t/h optimal solutions have been obtained in a reasonable time
Deaerator (water load) 210.3 t/h 310.5 t/h 360.5 t/h for the proposed approach.
By using the engineering knowledge and analytical in-
sight, a discrete scheme is proposed to identify the sizes of
Table 16 the candidate steam turbines, condensing turbines and gas
Results for the two approaches for Case 1 turbines. The optimisation problem is formulated as a multi-
Method Number of Number of period MILP model that relies on the THM, CTHM, GTHM
binary variables continuous variables and the BHM to describe the performance of the BP steam
Reduced superstructure 92 195 turbines, condensing turbines, gas turbines and boilers. The
Full superstructure 1628 2697 models account for the efficiency variations with operating
conditions and capacity. It should be emphasised that, had
conventional models for the units been applied, the use of
Table 17 an MINLP formulation would be inevitable.
Results for the two approaches for Case 2
Method Number of Number of
binary variables continuous variables

Reduced superstructure 60 97
Full superstructure 1140 1905 Appendix A. Condensing turbine hardware model
(CTHM)

The condensing turbine hardware model (CTHM) ex-


highly complex and large problems. The proposed method- tends the THM for BP steam turbines, originally proposed
ology utilises total site analysis and thermodynamic analy- by Mavromatis and Kokossis (1998a). The model as-
sis to reduce the size and complexity of the design problem. sumes Willans lines (Church, 1950) and its shaftwork is
4448 Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451

0.85 0.35
0.8 0.3
0.25
0.75 0.2

Ac
0.7 0.15
ηis,max

0.65 0.1
14bar 0.05
0.6 28bar
41bar 0
0.55 62bar 150 200 250 300 350
83bar Tsat (°C)
0.5 103bar
0.45 Fig. 23. Regression parameter Ac as a function of inlet saturation tem-
0.1 1 10 100 perature.
Emax(MW)

Fig. 22. Typical data on the efficiency of condensing turbines (Peterson 1.35
and Mann, 1985).

1.3

Bc
given by
  1.25
6 1 Ac
W= H is −
5 Bc M max
  1.2
1 150 200 250 300 350
× M − M max . (A.1)
6 Tsat (°C)

W stands for the shaft-work of the condensing turbine; Ac , Fig. 24. Regression parameter B c as a function of inlet saturation tem-
B c are the regression parameters, H is the isentropic en- perature.
thalpy change; M the steam flowrate; and M max the steam
turbine capacity.
The isentropic efficiency is given by 0.25
  
6 1 Ac 1 M max 0.2
cis = 1 − 1 − . (A.2)
5 Bc H is M max 6 M 0.15
Ac

The differences between the CTHM and the THM relate to 0.1
the regression parameters Ac and B c . Typical data on the
efficiency of condensing turbines are found from plots of 0.05
the maximum efficiency, as shown in Fig. 22 (Peterson and 0
Mann, 1985). The regression parameters are derived from 150 200 250 300 350
the data on the figure. By definition, the maximum efficiency Tsat(°C)
is
Fig. 25. Regression parameter Ac as a function of inlet saturation tem-
E max perature (E max  1.5 MW).
is,max = . (A.3)
His M max
Hence
are given in Figs. 23 and 24, respectively. These parameters
E max are in turn approximated by the following expressions:
H is M max
= . (A.4)
is,max
Ac = −0.0896 + 0.0013T sat , (A.6)
E max is the maximum power output of the turbine. For each
potential inlet pressure, the curves of Fig. 22 are represented B c = 1.1752 + 0.0003T sat , (A.7)
by
where the inlet saturation temperature is in ◦ C. However,
E max the above expressions do not give accurate estimates of
His M max = = Ac + B c E max . (A.5) the maximum efficiency when the power output is below
is,max
1.5 MW. More accurate estimates for the efficiency are ob-
Parameters Ac and B c are extracted by regression for each tained by conducting regression below 1.5 MW and above
inlet pressure. The plots of parameters Ac and B c against that 1.5 MW. The two segments’ regression gives efficiency es-
saturation temperature corresponding to the inlet pressure timates within 3% error (Figs. 25–28). The corresponding
Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451 4449

1.4 The condensing turbine hardware model follows similar


principles with the THM. It has the capacity to accurately
1.38
consider the effect of turbine size, load and the operating
1.36 conditions on the efficiency of typical condensing turbines.
The maximum efficiency plots represent typical condens-
Bc

1.34 ing turbines. More accurate estimates can be obtained by


1.32 applying more segments’ regression analysis.

1.3
150 200 250 300 350 Appendix B. Gas turbine hardware model (GTHM)
Tsat(°C)
The GTHM makes use of basic thermodynamic princi-
Fig. 26. Regression parameter B c as a function of inlet saturation tem-
perature (E max  1.5 MW).
ples and available information for gas turbine equipment.
Fig. 29 illustrates the basic structure of a gas turbine cycle.
The model relates the power output (W ), the total power
0.5
loss of the gas turbine cycle (W loss ), the fuel mass flowrate
(F f ), the fuel temperature (T f ), the air mass flowrate (F a ),
0.4 the air temperature (T a ) and the outlet temperature T out in
0.3
the expression:
Ac

0.2 W + W loss = F a Cpa T a + F f Cp f T f


+ F f Hf − (F a + F f )Cp g T out . (B.1)
0.1

0
Cpa is the specific heat of air, Cp f the specific heat of
150 200 250 300 350 fuel; Cpg the specific heat of flue gas; Hf the specific fuel
Tsat(°C) combustion heat, and W loss mechanical losses, heat loss and
so on.
Fig. 27. Regression parameter Ac as a function of inlet saturation tem- If we define f = F f /F a , Eq. (B.1) becomes
perature (E max  1.5 MW). 
1
W= Cp a T a + Cp f T f + Hf
f
  
1.29 1
− 1+ Cp g T out F f − W loss . (B.2)
1.28 f
1.27 Defining,
Bc

1.26
1
1.25 h = Cp a T a + CP f T f + Hf
f
1.24
 
1
− 1+ Cp g T out . (B.3)
1.23 f
150 200 250 300 350
Tsat(°C) Eq. (B.2) yields

Fig. 28. Regression parameter B c as a function of inlet saturation tem- W = hF f − W loss . (B.4)
perature (E max  1.5 MW).
The parameters h and W loss depend on the size of the gas
turbine. The electrical efficiency increases with size with
expressions for the regression parameters are indicative dependencies shown in Fig. 30 (Marechal and
for E max < 1.5 MW Kalitventzeff, 1998). The plot illustrates the influence of
turbines size on efficiencies at full load but not on partly
Ac = −0.0981 + 0.001T sat , (A.8) loaded units. The curve fitting of the plot in Fig. 30 yields
the following expression:
B c = 1.2059 + 0.0006T sat , (A.9)
W max
= Ag + B g W max . (B.5)
for E max > 1.5 MW e,max

Ac = −0.0376 + 0.0014T sat , (A.10) The parameters Ag and B g are obtained by conducting re-
gression for the curve in Fig. 30. The resulting values of Ag
B c = 1.1718 + 0.0003T sat . (A.11) and B g are 6.7571 and 2.4381, respectively. But the above
4450 Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451

F f,Tf It follows that

Tout W max
e,max = . (B.11)
Hf F f,max
Combustor
It is assumed that
Fa,Ta
W W loss = nW max (B.12)
Compressor Turbine
Wloss n is a constant parameter. According to Eqs. (B.4) and (B.12)
the maximum power output is equal to
Fig. 29. Simple gas turbine cycle.
W max = hF f,max − nW max . (B.13)

Substituting Eq. (B.11) into Eq. (B.5),


0.45
1
F f,max = (Ag + B g W max ). (B.14)
0.4 Hf
0.35
By combining Eqs. (B.11)–(B.13), the following expressions
ηe,max

0.3 for the parameters of the model are derived:


 
0.25 1 Ag
h = (1 + n) g Hf − f,max (B.15)
B F
0.2
0 10 20 30 40 50 60 70 and
1  
Wmax(MW)
W loss = n g
Hf F f,max − Ag . (B.16)
Fig. 30. Typical data on the electrical efficiency of gas turbines as a B
function of size (Marechal and Kalitventzeff, 1998).
Substituting Eqs. (B.14) and (B.15) into the Eq. (B.3) gives
 
1 Ag
W = g Hf − f,max
expressions do not give accurate estimates of the maximum B F
efficiency when the power output is below 6.9 MW by using × ((1 + n)F f − nF f,max ). (B.17)
the parameters. More accurate estimates for the efficiency
are obtained by conducting regression below 6.9 MW and According to Eq. (B.2) and Eq. (B.14), the outlet temperature
T out is given by

(1/f )Cp a T a + Cp f T f + Hf − (1 + n)(1/B g )/(Hf − Ag /F f,max )


T out = . (B.18)
(1 + 1/f )Cp g
Eq. (B.17) relates the power output of the gas turbine to
above 6.9 MW. The two segments’ regression gives effi- the size and the load of the turbine, as well as its operating
ciency estimates within 2% error. The corresponding regres- conditions. Eq. (B.18) relates the outlet temperature of the
sion parameters are for W max < 6.9 MW gas turbine to the size of the turbine and its operating con-
ditions, expressed through T a , T f , f , Hf . Given the ex-
Ag = 2.0836, (B.6) pected operating conditions, the performance of the turbine
can be estimated merely on the basis of its size. While the
B g = 3.1724, (B.7) non-linear variation of the efficiency is accounted for, the
relation of the power output to the load is linear. The pre-
for W max > 6.9 MW dicted electrical efficiency can be derived from Eqs. (B.5)
and (B.16):
Ag = 8.817, (B.8)   
1 Ag F f,max
e = g 1 − (1 + n) − n .
B g = 2.3905. (B.9) B Hf F f,max Ff
(B.19)
From the definition of the electrical efficiency
By plotting the predicted efficiency against the fuel load F f
W (or W) at various turbine size F f,max , Fig. 31 shows the
e = . (B.10)
Hf F f variation of the electrical efficiency with load and the effect
Z. Shang, A. Kokossis / Chemical Engineering Science 60 (2005) 4431 – 4451 4451

0.5 Church, E.F., 1950. Steam Turbines. McGraw-Hill Book Company,


40MW 60MW New York.
0.4 Church, E.F., 1950. Steam Turbines. McGraw-Hill Book Company,
20MW
0.3 New York.
Cohen, H., Rogers, G.F.C., Saravanamuttoo, H.I.H., 1987. Gas turbine
ηe

0.2 theory, Longman Scientific & Technical, Harlow.


Dhole, V.R., Linnhoff, B., 1992. Total site targets for fuel, co-generation,
0.1
emissions and cooling. Computers & Chemical Engineering 17,
0 s101–s109.
0 10 20 30 40 50 60 Hui, C.W., Natori, Y., 1996. An industrial application using mixed
W (MW) integer-programming technique: a multi-period utility system model.
Computers & Chemical Engineering 20, s1577–s1582.
Fig. 31. The GTHM accounts for the effect of gas turbine size and load Iyer, R.R., Grossmann, I.E., 1998. Synthesis and operational planning
on efficiency. of utility systems for multiperiod operation. Computers & Chemical
Engineering 22, 979–993.
Linnhoff, B., Turner, J.A., 1981. Heat recovery networks: new insights
yield big savings. Chemical Engineering 88, 56–70.
of the turbine size on the turbine efficiency as it would be Maia, L.O.A., Qassim, R.Y., 1997. Synthesis of utility systems with
expected by a reliable and realistic model. variable demands using simulated annealing. Computers & Chemical
Engineering 21, 947–950.
The GTHM is based on basic thermodynamic principles
Makwana, Y., 1997. Energy retrofit and debottlenecking of total sites.
and considers a maximum efficiency plot, assuming fixed Ph.D. Thesis, Department of Process Integration, UMIST, Manchester,
losses W loss to the maximum power output. Thermodynam- UK.
ics principles ensure heat and mass balances and the max- Marechal, F., Kalitventzeff, B., 1998. Process integration: selection of
imum efficiency plot accounts for a standard performance the optimal utility system. Computers & Chemical Engineering 22,
s149–s156.
for the gas turbine. In addition to capacity, load and oper-
Mavromatis, S.P., Kokossis, A.C., 1998a. Conceptual optimisation of
ating conditions, the efficiency of gas turbine depends on a utility networks for operational variations—1: targets and level
series of other factors, such as it’s type, technology and age. optimisation. Chemical Engineering Science 53, 1585–1608.
Consequently, it is not realistic to expect all gas turbines to Mavromatis, S.P., Kokossis, A.C., 1998b. Conceptual optimisation of
fit in the same set of curves. It is sensible to obtain plots utility networks for operational variations—2: network development
and optimisation. Chemical Engineering Science 53, 1609–1630.
for the particular class of turbines that are under consider-
Oliveira Francisco, A.P., Matos, H.A., 2003. Multiperiod synthesis and
ation and extract the corresponding regression parameters operational planning of utility systems with environmental concerns.
that will more accurately describe the specific class. ESCAPE -13, Computer-Aided Chemical Engineering 14, 281–286.
Papoulias, S.A., Grossmann, I.E., 1983. A structural optimization approach
in process synthesis—I utility systems. Computers & Chemical
References Engineering 7, 695–706.
Peterson, J.F., Mann, W.L., 1985. Steam system design: how it evolves.
Bruno, J.C., Fernandez, F., Castells, F., Grossmann, I.E., 1998. A rigorous Chemical Engineering 14, 62–74.
MINLP model for the optimal synthesis and operation of utility plants. Raissi, K., 1994. Total site integration. Ph.D. Thesis, Department of
Chemical Engineering Research & Design 76, 246–258. Process Integration, UMIST, Manchester, UK.
Chou, C.C., Shih, Y., 1987. Thermodynamic approach to the design and Shang, Z, Kokossis, A.C., 2004. A transhipment model for the optimisation
synthesis of plant utility system. Industrial and Engineering Chemistry of steam levels of total site utility system for multiperiod operation.
Research 26, 1100–1108. Computers & Chemical Engineering 28, 1673–1688.

You might also like