CHEMICAL ENGINEERING TRANSACTIONS
VOL. 52, 2016
A publication of
The Italian Association
of Chemical Engineering
Online at www.aidic.it/cet
Guest Editors: Petar Sabev Varbanov, Peng-Yen Liew, Jun-Yow Yong, Jiří Jaromír Klemeš, Hon Loong Lam
Copyright © 2016, AIDIC Servizi S.r.l.,
ISBN 978-88-95608-42-6; ISSN 2283-9216
Sequential Initialisation Technique for Synthesising Multi-
Period Multiple Utilities Heat Exchanger Networks
Adeniyi J. Isafiade*a, Oludare J. Odejobib
aDepartment of Chemical Engineering, University of Cape Town, South Africa
bDepartment of Chemical Engineering, Obafemi Awolowo University, Ile-Ife, Nigeria.
Aj.isafiade@uct.ac.za
This paper presents a new method for synthesizing heat exchanger networks involving multi-period operations
and multiple utilities. Since process parameters, such as streams’ flowrate and supply/target temperatures, in
multi-period operations usually fluctuate around some nominal conditions, the synthesis models developed for
their optimisation need to ensure that the heat load requirement of every stream in every period of operation is
satisfied. Solving problems of this nature is not a trivial task due to the presence of multiple periods of
operations. The complexity is further compounded in situations where multiple options of utilities are involved
in the optimisation task. Hence in this paper, a new synthesis method which involves firstly, solving single
period models for each period of operation in the problem, after which the selected matches from each of the
single period solution are systematically used to initialise the multi-period superstructure. Contrary to existing
methods where utilities are placed or appended only in the first and last stage of the multi-period
superstructure model, the newly presented method positions utilities in all stages of the superstructure where
process streams of the opposite kind exist. The new approach is applied to two examples and the solutions
obtained demonstrate the benefits of the method.
1. Introduction
Energy using processes such as those found in chemical and petroleum industries are under pressure to
reduce the amount of energy consumed in their operations. This is due to the highly uncertain costs and
availability of fossil based fuels. One other key reason for which energy usage should be reduced, especially
those that are fossil based, is the associated environmental impact which is gradually having its toll on the
world’s environment. A large number of studies have been carried out on ways of achieving the
aforementioned reductions in energy usage and one of them is through designing efficient networks of heat
exchangers. However, most papers presented in this area have been based on the assumption that process
operating parameters are fixed, as done in the paper presented by Short et al. (2015). However, in reality,
process stream parameters do vary within certain ranges due to issues such as changes in environmental
conditions, plant start-ups/shut downs, changes in feed/product quality, etc. Hence, this implies that heat
exchanger network synthesis (HENS) design methods need to be capable of handling the aforementioned
changes in process operating conditions. Networks, that fulfil these criteria, are called multi-period networks.
The work of Verheyen and Zhang (2006) addressed the synthesis of multi-period heat exchanger networks
using a modified version of the stage-wise superstructure (SWS) of Yee and Grossmann (1990). Isafiade and
Fraser (2010) also presented an approach for solving multi-period problems using their interval based
superstructure approach. Other studies which addressed the multi-period design problem are that of Kang et
al. (2015), where a 2-step approach was used, and that of Jiang and Chang (2015), where the time sharing
scheme was used. Although these methods addressed the multi-period HENS problem, however, they are
limited in that they considered only one option of utility, and this utility was placed or appended to the first and
last intervals (for hot and cold utilities respectively) of the stage-wise superstructure. The work of Isafiade et al.
(2015), which also used the multi-period SWS model, however considered multiple utilities in solving HEN
multi-period problems. This approach was limited due to the fact that it placed all hot utilities only in the first
interval of the multi-period superstructure and all cold utilities only in the last interval. In as much as this
DOI: 10.3303/CET1652125
Please cite this article as: Isafiade A. J., Odejobi O. J., 2016, Sequential initialisation technique for synthesising multi-period multiple utilities
heat exchanger networks, Chemical Engineering Transactions, 52, 745-750 DOI:10.3303/CET1652125
745
superstructure definition approach has the benefit of simplifying the solution generation process by virtue of
the fact that the number of potential discrete variables are reduced, its main drawback is that for cases where
the multiple utilities have significantly different temperature ranges, they will not have the opportunity to
exchange heat with streams of the opposite kind through an optimal use of driving forces. Hence in this paper,
a new approach which positions every utility in all intervals of the multi-period superstructure where process
streams of the opposite kind exist, is presented. The authors of this paper are aware that this superstructure
definition approach has the tendency to increase the computational intensity involved in solving the problem,
as it will result in an increase in the degree of non-linearity of the model, as well as increase in the number of
discrete variables. However, in order to circumvent this potential problem, a systematic initialisation
procedure, which is newly generated in this paper is adopted.
2. Problem statement
The problem solved in this paper can be stated as follows: Given a set HP of hot process streams, and a set
CP of cold process streams, with their supply temperatures, target temperatures and flowrates at specified P
periods of operations. The hot and cold streams have to be cooled and heated respectively. Given also are a
set of hot (HU) and cold (CU) utilities which can be used in any of the specified periods of operations. The task
is to synthesize a heat exchanger network which would satisfy the heat demand of all process streams in all
periods of operations while using utilities and heat exchangers in a cost effective manner.
3. Model development
Figure 1 shows the modified multi-period stage-wise superstructure used in this paper. This representative
superstructure involves 2 hot process streams, 2 cold process streams, 2 hot utilities and 2 cold utilities. The
process streams are represented as thick lines while the utility streams are represented as dashed lines in the
figure. What is unique to this superstructure is that hot utilities 1 and 2 are both made to participate in stages
2, 3 and 4, where cold process streams 1 and 2 exist. Also, cold utilities 1 and 2 are made to participate in
stages 1, 2 and 3 where hot process streams 1 and 2 are present. However, the hot and cold process streams
run from their starting intervals, as shown in the figure, to the end of the superstructure, while the utilities may
or may not cross intervals. The utilities will not cross intervals in cases where they do not have a significant
temperature change, e.g. steam, while they will cross intervals in cases where they have significant
temperature change, e.g. hot oil. This new representation of multi-period stage-wise superstructure implies
that for the specific case of Figure 1, process heat exchangers can only exist in stages 2 and 3 (represented
as two black circles connect with a line) while utility exchangers (represented as two white circles connect with
a line) can exchange heat in all stages of the superstructure (stages 1, 2 and 3 for cold utilities and stages 2,3
and 4 for hot utilities). The same assumption of equal temperature mixing for split streams, as presented by
Yee and Grossmann (1990) for the single period scenario, is also adopted in this paper for both single and
multi-period scenarios. It should be known that Figure 1 is illustrated for a single period case, however in order
to convert it to a multi-period model, the index 𝑝, which represents each period of operation, is included in the
single period model equations.
HU1 HU1 HU1
HU2HU2HU2
CU2 CU2 CU2
CU1 CU1 CU1
HP1
HP2
CP1
CP2
Stage 1 Stage 2 Stage 3 Stage 4
Figure 1: Modified multi-period stage-wise superstructure used in this work
746
Due to space limitation, the detailed set of model equations will not be shown here, however, the reader can
get the details from the work of Verheyen and Zhang (2006) and later Isafiade et al. (2015). Only the objective
function used is shown in this paper which is illustrated in Eqs(1) and (2).
𝐴𝑖,𝑗,𝑘 ≥
𝑞𝑖,𝑗,𝑘,𝑝
(𝐿𝑀𝑇𝐷𝑖,𝑗,𝑘,𝑝) (𝑈𝑖,𝑗 )
(1)
𝑚𝑖𝑛 {∑ (
𝐷𝑂𝑃𝑝
∑ 𝐷𝑂𝑃𝑝
𝑁𝑂𝑃
𝑝=1
∑ ∑ ∑ 𝐶𝑈𝐶𝑗 ∙ 𝑞𝑖,𝑗,𝑘,𝑝
𝑘𝜖𝐾𝑗𝜖𝐶𝑈𝑖𝜖𝐻𝑃
+
𝐷𝑂𝑃𝑝
∑ 𝐷𝑂𝑃𝑝
𝑁𝑂𝑃
𝑝=1
∑ ∑ ∑ 𝐻𝑈𝐶𝑖 ∙ 𝑞𝑖,𝑗,𝑘,𝑝
𝑘𝜖𝐾𝑗𝜖𝐶𝑃𝑖𝜖𝐻𝑈
)
𝑝𝜖𝑃
}
+ 𝐴𝐹 (∑ ∑ ∑ 𝐶𝐹𝑖𝑗 ∙ 𝑧𝑖,𝑗,𝑘 +
𝑘𝜖𝐾𝑗𝜖𝐶𝑖𝜖𝐻
∑ ∑ ∑ 𝐴𝐶𝑖𝑗 ∙ 𝐴𝑖,𝑗,𝑘
𝐴𝐶𝐼
𝑘𝜖𝐾𝑗𝜖𝐶𝑖𝜖𝐻
) ∀ 𝑖𝜖𝐻𝑃, 𝑗𝜖𝐶𝑃, 𝑘𝜖𝐾, 𝑝𝜖𝑃 (2)
In Equation 1, 𝐴𝑖,𝑗,𝑘 represents the area (m
2) of the heat exchanger that connects hot stream 𝑖 and cold stream
𝑗 in interval 𝑘 of the superstructure, 𝐿𝑀𝑇𝐷𝑖,𝑗,𝑘,𝑝 represents the logarithmic mean temperature difference (K)
between hot stream 𝑖 and cold stream 𝑗 in stage 𝑘 and period 𝑝 of the superstructure, 𝑞𝑖,𝑗,𝑘,𝑝 represents the
amount of heat (kW) exchanged between hot stream 𝑖 and cold stream 𝑗, in period 𝑝, 𝑈𝑖,𝑗 represents overall
heat transfer coefficient between hot stream 𝑖 and cold stream 𝑗 ( 0.1 W/m2°C). In Eq(2), which is the objective
function, 𝐷𝑂𝑃𝑝 represents duration of period 𝑝 while 𝑁𝑂𝑃 represents the number of periods, 𝐶𝑈𝐶𝑗 represents
cost of cold utility ($/(kW∙y)), 𝐻𝑈𝐶 represents cost of hot utility ($/(kW∙y)), 𝐴𝐹 represents the annualisation
factor (0.2), 𝐶𝐹𝑖𝑗 represents the cost of installing each unit of heat exchanger (8,333.3 $), 𝑧𝑖,𝑗,𝑘 represents the
binary variable whose purpose is to show the existence, or otherwise, of a match between any hot and any
cold streams in the superstructure, 𝐴𝐶𝑖𝑗 represents area cost of heat exchangers (641.7 $/m
2) while 𝐴𝐶𝐼
represents heat exchanger area cost index. Equations 1 and 2 show that the maximum area approach as
used by Verheyen and Zhang (2006) and later by Isafiade et al. (2015) is also used in this paper. The
maximum area ensures that the area used for the calculation of heat exchanger sizes in the objective function
is the area of the maximum of heat exchanger areas exchanging heat between the same pair of streams
existing in different periods of the superstructure for interval 𝑘. Apart from the modified multi-period stage-wise
superstructure used in this paper, another unique feature of the new method is the solution approach adopted.
The solution technique entails basically using a systematic approach to initialise the multi-period
superstructure at various times, so as to obtain a good solution in reasonable time. This is necessary because
apart from the nonlinear expressions in the model, the presence of multiple periods of operations also
complicates the solution generation process. A further complexity inherent in the model is the fact that the
number of discrete variables are significantly increased since the utilities are made to participate in every
interval where process streams exist. However these difficulties are overcome in this paper using the solution
procedure outlined next.
Generate a single period superstructure model for each period of operation using the modified single period
SWS superstructure shown in Figure 1.
i. Solve each of the single period models generated in the first step (using the same number of
superstructure stages for each period) as a mixed integer non-linear program (MINLP) and identify the
selected matches. Matches which are selected in more than one period should be represented only
once in the list of identified matches.
ii. Use the set of identified matches in the second step to initialise the multi-period model, which is a
multi-period superstructure version of Figure 1. It should be known that despite the fact that the multi-
period model at this stage is initialised with a set of matches, the model is still solved as an MINLP
because the initialising matches represent the binary variables. The multi-period model at this stage
should have the same number of stages as used in the second step for the individual single period
models.
iii. Systematically add or remove matches in the reduced multi-period superstructure of the third step
while solving until an optimal solution is obtained.
It is worth stating that Isafiade et al. (2015) used a somewhat similar solution generation approach to that
outlined above. However, their method restricted utilities to only the first and last stages of the superstructure.
Also, Isafiade et al. (2015) did not solve individual single period models so as to generate the set of initialising
matches, instead they solved a full multi-period model a number of times, and then identified the selected
matches in about 2 or 3 of the best solutions depending on the size of the problem. The identified matches
were then used to initialise the multi-period model. A key shortcoming of this approach is that since the utilities
are restricted to only one interval of the superstructure, there may not be an optimal use of driving forces,
747
especially for intermediate utilities. Furthermore, chances of getting good solutions at the stage of solving the
full multi-period model may be reduced due to the highly non-linear nature of MINLP multi-period multiple
utility models.
4. Examples
The newly developed method is applied to 2 examples both having equal period durations. The first example
is taken from Isafiade et al. (2015) while the second example is generated in this paper so as to fully
demonstrate the benefits of the new approach.
4.1 Example 1
The process stream data for this example are shown in Table 1, while the cost data for its utilities are: HU1 =
70 €/(kW∙y), HU2 = 50 €/(kW∙y), HU3 = 40 €/(kW∙y), CU1 = 1.3 €/(kW.y). Other capital cost data are presented
with Equation 1. Applying the 1st and 2nd steps of the solution procedure gives the set of matches (and total
annual costs, TAC) for each of periods 1, 2 and 3 shown in Table 2. Based on these set of selected matches,
a new set of matches which are common to periods 1 to 3, or present in at least one of the periods, were
chosen for use in the 3rd step to initialize the multi-period model. These new set of matches, known as
initializing matches, are shown in the last column of Table 2. Using these set of initializing matches in the
multi-period model greatly reduces the number of discrete variables, hence simplifies the solution generation
process. Solving the model at this stage gave a TAC of € 4,416,756 with 13 units. This solution was obtained
in 2.12 s of CPU time. The last step was then carried out by removing matches (HP1,CP1,3) and (HP1,CP2,3)
and the model was solved again. At this last step, a solution with a TAC of € 4,413,854, having 9 units, was
obtained in 1.67 s of CPU time. It is worth stating that exactly the same TAC and number of units was
obtained by Isafiade et al. (2015), however the structures of the solution networks are different. Further
excluding matches (HP2,CP2,2) and (HP3,CP1,3) does not have any effect on the solution obtained.
However, if any of the other matches were excluded, the TAC goes up. This shows the importance of having
the right set of matches, alongside other initializations for key variables, at the initialization stage, in solving
multi-period problems in a mathematical programming environment. Systematically initializing the model at the
third step, which is the step at which the multi-period model is solved, helps position the solver within the right
region of finding good solutions in reasonable time. The structure obtained using the method of this paper is
shown in Figure 2.
Table 1: Process stream data for Example 1
Streams Periods
Period 1 Period 2 Period 3
Ts (°C) Tt (°C) F (kW/°C) Ts (°C) Tt (°C) F (kW/°C) Ts (°C) Tt (°C) F (kW/°C)
HP1 393 60 201.6 406 60 205 420 60 208.5
HP2 160 40 185.1 160 40 198.8 160 40 175.2
HP3 354 60 137.4 362 60 136.4 360 60 134.1
CP1 72 356 209.4 72 365 210.3 72 373 211.1
CP2 62 210 141.6 62 210 141.0 62 210 140.5
CP3 220 370 176.4 220 370 175.4 220 370 174.5
CP4 253 284 294.4 250 290 318.7 249 286 271.2
HU1 490 490 - 490 490 - 490 490 -
HU2 438 438 - 438 438 - 438 438 -
HU3 375 375 - 375 375 - 375 375 -
CU1 0 10
In this figure, the representative areas (i.e. the maximum areas) for each exchanger is shown above the
exchanger. It can be seen in the figure that HU3 and CU1 are both used in intermediate stages, which is stage
2. This is unlike the solution of Isafiade et al. (2015) where utilities are only used in the first and last intervals,
which are the only available utility stages. In the solution of Isafiade et al. (2015), only HU2 and HU3 were
used as hot utilities, while CU1 was used as cold utilities.
4.2 Example 2
The process stream data for this example are shown in Table 4 while the cost data for its utilities are HU1 =
70 €/(kW∙y), HU2 = 30 €/(kW∙y), HU3 = 20 €/(kW∙y), CU1 = 1.3 €/(kW∙y), CU2 = 1.0 €/(kW∙y). Other capital
cost data are presented with Equation 1. Note that this example has 2 cold utilities, unlike Example 1 which
has 1 cold utility. Applying the 1st and 2nd steps of the solution procedure gives the set of matches, for each of
the periods, shown in Table 3. It can be seen that the set of matches obtained for each of the periods in step 2
748
contains a large number of selected matches that are unique to each of the periods. Therefore, when the
initializing matches were identified from the pool of matches obtained in step 2, a total of 23 initializing
matches was obtained. Since this is a large number for a moderately sized problem, a systematic approach
was used in reducing these set of initializing matches. The approach used entails solving a multi-period
version of the problem using the modified multi-period SWS model but with all multi-period binary variables
present in the model. A solution having the following matches (HU1,CP1,1), (HU1,CP3,2), (HU2,CP1,2),
(HU2,CP4,2), (HU3,CP2,1), (HP1,CP2,2), (HP1,CU2,4), (HP2,CP2,3), (HP2,CU1,4), (HP3,CP1,3),
(HP3,CU2,4) with a TAC of € 6,331,908 was obtained. After a series of repeated addition/elimination of
matches, using the resulting selected matches from the conventional solution approach as a guide, the set of
initializing matches were ultimately reduced to those shown in the last column of Table 3. Solving the multi-
period model with these final set of initializing matches gave a solution having 11 units with a TAC of €
6,240,266. This solution was obtained in 0.97 s of CPU time. Solving this problem using the conventional
method gives a solution having a TAC of € 7,507,781 with 10 units. The solution was obtained in 7.49 s of
CPU time. Figure 3 shows the final solution network obtained for Example 2. This solution has 2 process heat
exchangers, 7 heaters, and 3 coolers. It can be seen that 4 of the heaters are used in the intermediate stages.
This implies that process streams such as CP1, CP3 and CP4, all have the opportunity of having their
intermediate temperatures determined as a result of heat exchange with the intermediate placement of hot
utilities. This would not have been possible if the utilities had been restricted to only the first and last stages of
the superstructure.
Table 2: Selected matches for individual periods for
Example 1
Period 1 Period 2 Period 3 Selected
matches for
step 3
HU2,CP3,1
HU3,CP4,3
HP1,CP1,2
HP1,CP1,3
HP1,CU1,4
HP2,CU1,3
HP3,CP1,3
HP3,CP2,3
HP3,CU1,4
TAC:
€ 4,801,926
HU2,CP1,1
HU3,CP4,3
HP1,CP1,2
HP1,CP2,3
HP1,CP3,2
HP1,CU1,4
HP2,CU1,2
HP2,CU1,3
HP3,CP1,2
HP3,CP1,3
HP3,CU1,4
HP3,CP3,2
TAC: €
4,877,509
HU1,CP1,1
HU1,CP3,1
HU2,CP1,2
HU2,CP3,2
HU2,CP4,1
HP1,CP1,3
HP1,CU1,4
HP2,CP2,3
HP2,CU2,4
HP3,CP2,2
HP3,CU2,3
TAC: €
6,377,848
HU2,CP1,1
HU2,CP3,1
HU3,CP4,3
HP1,CP1,2
HP1,CP1,3
HP1,CP2,3
HP1,CP3,2
HP1,CU1,4
HP2,CU1,2
HP2,CU1,3
HP3,CP1,2
HP,CP1,3
HP3,CP2,3
HP3,CP3,2
HP3,CU1,4
Table 3: Selected matches for individual periods for
Example 2
Period 1 Period 2 Period 3 Selected
matches for
step 3
HU1,CP1,1
HU1,CP3,2
HU2,CP1,2
HU2,CP3,3
HU2,CP4,2
HP1,CP2,3
HP1,CU1,4
HP2,CU2,2
HP3,CP2,2
HP3,CU1,4
TAC:
€ 6,390,959
HU1,CP1,1
HU1,CP3,1
HU2,CP1,2
HU2,CP3,2
HU2,CP4,3
HU3,CP2,1
HP1,CP2,2
HP1,CU2,3
HP2,CU2,2
HP3,CP1,3
HP3,CU2,4
TAC:
€ 6,461,437
HU2,CP3,1
HU3,CP4,3
HP1,CP1,3
HP1,CU1,4
HP2,CU1,2
HP2,CU1,3
HP3,CP2,3
HP3,CP3,2
HP3,CU1,4
TAC:
€ 4,636,293
HU1,CP1,1
HU1,CP3,1
HU2,CP1,2
HU2,CP3,3
HU2,CP4,2
HU3,CP2,1
HP1,CP2,2
HP1,CU1,4
HP2,CU1,4
HP2,CU2,2
HP3,CP1,3
HP3,CU1,4
5. Conclusions
Presented in this paper is a new synthesis method for multi-period heat exchange network problems involving
multiple utilities. The method positions every utility in the problem in every stage of the superstructure where
process streams of the opposite kind exist, so that there can be an optimal use of available driving force.
Using this superstructure generation approach increases the complexity of solution generation, therefore the
new method uses a solution approach where the multi-period problem is initialized using selected matches
from each period’s optimal solution network. This new method overcomes some of the drawbacks associated
with current multi-period synthesis techniques that are based on the SWS model, in that good solutions can be
obtained in reasonable times. One of the main drawbacks associated with this new method is that there are no
specific rules to follow concerning addition/removal of matches in the last step of the new method, hence this
step may become tedious to handle for large problems.
749
Table 4: Process stream data for Example 2
Streams Periods
Period 1 Period 2 Period 3
Ts (°C) Tt (°C) F (kW/°C) Ts (°C) Tt (°C) F (kW/°C) Ts (°C) Tt (°C) F (kW/°C)
HP1 213 60 201.6 206 60 205 220 60 208.5
HP2 160 40 185.1 160 40 198.8 160 40 175.2
HP3 254 60 137.4 262 60 136.4 260 60 134.1
CP1 72 356 209.4 72 365 210.3 72 373 211.1
CP2 62 210 141.6 62 210 141.0 62 210 140.5
CP3 220 370 176.4 220 370 175.4 220 370 174.5
CP4 253 284 294.4 250 290 318.7 249 286 271.2
HU1 490 490 - 490 490 - 490 490 -
HU2 338 338 - 338 338 - 338 338 -
HU3 275 275 - 275 275 - 275 275 -
CU1 0 10
CU2 15 30
Acknowledgements
This work is based on research supported by the National Research Foundation (NRF) of South Africa for
rated researchers, grant numbers: 87744 and 85536.
References
Isafiade A., Bogataj M., Fraser D., Kravanja Z., 2015, Optimal synthesis of heat exchange networks for multi-
period operations involving single and multiple utilities, Chemical Engineering Science, 127, 175-188.
Isafiade A.J., Fraser, D.M., 2010, Interval based MINLP superstructure synthesis of heat exchanger networks
for multi-period operations, Chemical Engineering Research and Design, 88(10), 1329 - 1341.
Jiang D., Chang C.-T., 2015, An algorithmic approach to generate timesharing schemes for multi-period HEN
designs, Chemical Engineering Research Design, 93, 402-410.
Kang L., Liu Y., Hou J., 2015, Synthesis of multi-period heat exchanger network considering characteristics of
sub-periods, Chemical Engineering Transactions, 45, 49 – 54.
Short M., Isafiade A.J., Fraser D.M., Kravanja Z., 2015, Heat exchanger network synthesis including detailed
exchanger designs using mathematical programming and heuristics, Chemical Engineering Transactions,
45, 1849-1854.
Verheyen W., Zhang N., 2006, Design of flexible heat exchanger network for multi-period operation, Chemical
Engineering Science 61, 7760-7753.
Yee T., Grossman I., 1990, Simultaneous optimization models for heat integration—II. Heat exchanger
network synthesis, Computers & Chemical Engineering, 14(10), 1165-1184.
Stage 4 Stage 3 Stage 2 Stage 1
490
490490490490490
438
438438438438438
375
375375375375375
0 0 010 10 10
HP1
HP2
HP3
CP1
CP2
CP3
CP4
900.5
1560
7038
13272861
1229
2864
1156
1991
Figure 2: Final multi-period network obtained for
Example 1
HU1
HU2
HU3
CU1
Stage 4 Stage 3 Stage 2 Stage 1
490
490490490490490
338
338338338338338
275
275275275275275
0 0 010 10 10
HP1
HP2
HP3
CP1
CP2
CP3
CP4
2864
1991
15 15 1530 30 30CU2
Figure 3: Final multi-period network obtained
for Example 2
750