SPE-199923-MS

Application of Mesh Adaptive Derivative-Free Optimization Technique for

Gas-Lift Optimization in an Integrated Reservoirs, Wells, and Facilities
Modeling Environment

Mohammad Khoshkbarchi, Mohammad Rahmanian, Jonas Cordazzo, and Long Nghiem, Computer Modelling
Group Ltd.

Copyright 2020, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE Canada Heavy Oil Techncial Conference originally scheduled to be held in Calgary, Alberta, Canada, 18 – 19
March 2020. Due to COVID-19 the physical event was postponed until 29 September – 2 October 2020 and was changed to a virtual event. The official proceedings
were published online on 24 September 2020.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents
of the paper have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect
any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written
consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may
not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract
Gas-lift is an important artificial lift strategy for increasing the production of hydrocarbons from heavy
oil and offshore reservoirs with declining pressure. The optimum design and operation of gas-lift has a
considerable impact on the optimum production and economics of the entire field and can only be achieved
by considering all related variables in connected reservoirs, gas-lifted wells and facilities. Therefore, it
is essential to formulate the gas-lift model as an optimization problem within an integrated modeling
environment, where all the time dependent physical and operational constraints of reservoirs, wells and
facilities can be collectively taken into account during the time evolutionary modeling of the asset. However,
this could pose a computationally challenging problem for most derivative based optimizers, as some of
the governing equations and models representing the gas-lifted asset could be very nonlinear, physically
or mathematically discontinuo us, and the system may not have a solution under some of the conditions
proposed by the optimizer due to physical operational restrictions or infeasibility. To overcome these
problems we applied a mesh adaptive direct search (MADS) derivative free optimization technique to
optimize the gas-lift problems in an integrated reservoir, wells and production facilities environment. The
results proved the suitability of the MADS strategy for optimizing these systems particularly for systems
with narrow or discontinuous boundaries of physical operation.

Introduction
When the reservoir pressure is inadequate for the wells to naturally flow to the surface or a well production is
lower than the desired rate, it is necessary to enhance the production by implementing a suitable artificial lift
strategy to supplement the energy of the reservoir and reduce the backpressure at the wellbore. Gas-lift is an
efficient artificial lift strategy, routinely used to overcome the low productivity of the wells. A typical gas-
lift set up involves injecting compressed gas into the produced reservoir fluid through the annulus connected
to an orifice valve, commonly referred to as gas-lift valve, placed at the middle and/or lower part of the
production string. The introduction of the gas into the produced reservoir fluid decreases its average density
2 SPE-199923-MS

and therefore lowers the hydrostatic pressure in the well tubing. This also leads to higher production rate by
lowering the bottom-hole pressure, leading to higher pressure difference across the well completion and the
drainage reservoir region which constitutes the main driving force for production. Although the reduction
of the hydrostatic pressure in the well as a result of injection of gas works in favor of the production,
the increase in the friction pressure drop as a result of the two-phase flow in the well contrasts this gain.
At high enough gas rate, the balance between these two effects imposes an adverse impact on the well
production rate. Therefore, for every gas-lifted well there is an optimum gas injection rate beyond which the
production decreases with an increase in the lift gas injection rate. For a single well, the relation between
the oil production rate and the gas injection rate is represented by the gas-lift performance curve shown
in Figure 1.

Figure 1—Lift performance curve for gas-lift optimization

For an isolated single well gas-lift study, where there is no constraint on the lift gas supply, the
maximum point on this curve determines the optimum amount of gas injection rate (Nishikiori, et al.
1989). Traditionally, a variation of this approach using nodal analysis has been applied to optimization of
production systems (Beggs 1991). Although, gas-lift optimization based on performance curves is simple
and easy to implement, it suffers from several shortcomings as it ignores flow interactions among reservoirs,
wells and production facilities. This assumption could lead to erroneous field design decisions as, for
even a simple system containing two wells and one common flow lines, it has been shown that the flow
interactions are very significant (Dutta-Roy and Kattapuram 1997). From a simulation point of view, this
in turn renders the mathematical models of the system a highly non-linear problem that strongly requires
a non-linear optimization technique to allocate gas optimally to gas-lifted wells. This problem intensifies
as the gas allocation problem needs to optimally consider the requirements of gas-lifted wells, as well as
other areas such as reservoir pressure maintenance by gas recycling, sales gas constraints, etc. There have
been several attempts to address this problem ranging from applying simple heuristic algorithms (Hong
1975, Kanu, Mach and Brown 1981, Buitrago, E. Rodriguez and Espin 1996, Weiss, et al. 1990), linear
programming, derivative based optimization techniques such as SQP (Dutta-Roy and Kattapuram 1997,
Davidson and Beckner 2003, Wang, Litvak and Aziz 2002, Fujii and Horne 1995), and derivative free
optimization techniques such as genetic algorithms, Tabu search (Vázquez, et al. 2001, Stoisits, et al. 1999,
Wang, Litvak and Aziz 2002) with various levels of success. A review of the developed strategies has been
given by (Rashid, Bailey and Couët, 2012). The heuristic optimization strategies are usually based on ad-
hoc rules such as repeatedly scaling the well rates to meet the gas constraints and then considering the water,
liquid, and oil rate constraints sequentially and handling the gas-lift optimization separately from well rate
optimization using. These procedures, although simple, may not yield the optimal solution. The optimization
strategies based on gradient based optimization are powerful methodologies to find the optimum point for
allocation problem if the problem is convex, continuous and the derivatives of the models with respect to
SPE-199923-MS 3

decision variable are all available and continuous. However, in problems such as gas allocation for assets
containing gas-lifted wells, the models could present physical discontinuities during optimization such as
flow patterns changes in the pipes and annulus and tubing, choke phenomenon in the gas-lift valve, critical
fluid phenomenon in injected gas, etc. Also, as the optimizer is searching the space for the optimal solutions,
there could be a set of proposed values that do not lead to a physical solution of the model resulting in
failing to calculate an objective function value and evaluation of the constraints. Therefore, there is a need
for a robust optimization strategy that could deal with these situations and find the optimum point despite
the existence of discontinuities and possible failure of the model. Mesh adaptive direct search (MADS)
algorithm (Audet and Hare 2017) is a promising derivative free optimization strategy which could perform
optimization without a need for derivatives for systems with noisy or discontinues structure that could fail
to solve for some conditions even those that are feasible.
In this study the MADS optimization strategy has been used to perform gas allocation optimization in an
integrated asset consisting of reservoir, a gas-lifted well and a gas injection well to study the suitability of
this optimization technique to deal with this problem. To perform this study, CoFlow (Hiebert, et al. 2011),
a next generation, fully integrated, simulator for simultaneous solution and optimization of integrated assets
comprising a connected set of reservoirs, production and injection wells and surface facilities has been used
as the modeling framework. The NOMAD optimization software (Le Digabel 2011, Abramson, et al. 2011)
that implements the MADS algorithm has been integrated as an optimizer option in the CoFlow to perform
the optimization on the integrated asset.

Derivative Free Optimization Using Mesh Adaptive Direct Search (MADS)

Mesh adaptive direct search algorithm is a generalized version of the General Pattern Search (GPS)
technique. The MADS algorithm is a direct search method based on a stencil-based optimization procedure
which only require objective function evaluations without a need for gradient information. It iterates on
a series of underlying meshes on the domain space. The goal of iteration is to find unfiltered points in
the variable space. In every iteration of the optimizer the algorithm consists of a search and a poll step,
performing an adaptive search on the meshes and refining them. To understand this better the algorithm can
be considered as a means for generating trial points on the underlying mesh and adapting the fineness of the
mesh to approach a local optimum point. During the search step the algorithm searches the solution space
for a better point in every iteration of the optimizer. When the search step is unsuccessful, the poll step is
performed. The poll step is a local exploration around the last successful solution. A mesh size parameter
is used to scale the directions of the generated trial points. The mesh size parameter is increased or held
constant when either the search or poll step is performed successfully; otherwise the last successful solution
is used to decrease the mesh size parameter.
Figure 2 illustrates the polling sequence and evolution of the optimization using MADS algorithm.

Figure 2—Illustration of polling sequence

4 SPE-199923-MS

Some of the advantages of using the MADS method are that it does not require the gradients of the
objective functions or the constraints during optimization process and therefore can handle systems with
discontinuities or hard to evaluate. It parallelizes naturally as the evaluation of the objective during the poll
points in a particular iteration can be performed in a distributed manner. It also does not require guaranteed
successful evaluation of the objective function in all iterations where the system may not have a physical
solution. The main drawback of the MADS algorithm is it requires O(n) function evaluations at each iteration
which could result in performance degradation.

Modeling Framework
To test the ability of the MADS optimization algorithm to optimize the gas allocation in integrated systems
comprising a set of reservoirs connected, via production and injection wells and surface facilities, CoFlow,
a next generation, fully integrated, simulator for simultaneous solution and optimization of integrated assets
has been employed. In the CoFlow structure the mathematical representations of all asset components
such as reservoirs, wells and production facilities, and all their contained components, are encapsulated
in separate models consisting of equations and variables. Equations are the main building blocks of the
models, and they define the (non-linear) relationships between the various variables in the system which
represent the operating or physical attributes of asset components such as gas injection flow rates, pressures,
etc. From a mathematical point of view, a system is made up of a collection of models M:
(1)
where the m-th model M depends on a set of equations E and a set of variables V as shown:
(2)
Each individual equation E must be expressed as a function of variables chosen from its set of n variables
V. The evaluation of equations is expected to return residuals, and the simultaneous satisfaction of all such
residuals corresponds to the desired solution. The equations supply derivatives with respect to their own
variables. If these variables are not the primary solution variables for the whole system, the framework
provide the necessary chain ruling to re-express the natural variables in terms of the primary solution
variables. Provisions are made within the framework to allow different solution techniques to be used on
different portions of the Jacobian matrix, which permits use of multi-level pre-conditioner strategies. For
the system to be solvable, the degree of freedom must be fulfilled which means that the number of free
variables must be equal to the number equations.
The connectivity between the production and injection facilities, wells and reservoirs is achieved by
defining the system as an undirected graph where every element such pipes, pumps, etc. is essentially a
node in the graph whose equations for a node with n inlets and m outlets is defined as follows:

(3)

(4)

with

(5)
where F, H and P denote mass flow rate, enthalpy and pressure, respectively; k refers to inlets and outlets
and j refers to components in the system.
SPE-199923-MS 5

The CoFlow structure also contains an optimization framework where some of the variables of the system
can be designated as decision variables with upper and lower bounds along with a set of equality or inequality
constraints to be solved by an optimizer with the goal of minimizing an objective function which is also a
function of the variables in the system.
(6)
By optimizing decision variables xi:
(7)
Subject to:
(8)
(9)
The optimization problem is set up by the optimizer, in every iteration, proposing a set of decision
variables which would render the entire simulation solvable using a Newton solver discussed by Hiebert et
al. 2011. After the successful completion of the entire asset solution, the objective function and constraints
are evaluated and optimization moves to the next set of proposed decision variables until the convergence
is achieved. It should be noted that the MADS algorithm can handle the failed solution of the asset for
unphysical situations that can arise from a particular set of decision variables.

Results and Discussions

To increase the oil production of petroleum assets several enhanced oil recovery (EOR) and/or improved
oil recovery (IOR) methods could be employed. Nonetheless, applying suitable and efficient production
strategies on each well such as gas lift, submersible pumps, acid jobs, hydraulic fracturing and etc. can also
assist reaching the overall optimum production. One of the interesting and common optimization problems is
how to allocate limited amount of available gas between pressure maintenance strategy and gas-lift operation
to achieve the maximum oil recovery in an asset. However, this could pose a complex optimization problem
due to physical and operational constraints which could render the asset to only be physically possible in
a narrow range of operating conditions. In this study we have implemented and studied the application
and abilities of MADS algorithm (implemented in NOMAD optimizer used by CoFlow) to optimize such
gas allocation problems in an integrated environment of reservoirs, wells and production and injection
facilities. In the following, three different cases are studied under different sets of operational restrictions
to achieve optimal production by allocating limited gas supply between gas-lifted wells and gas injection
to the reservoir.
Case (I): The reservoir model used in this case is a black oil model, adapted from (Odeh and Aziz 1981).
There is one producer and one injector well in this case which are connected to a surface facility as illustrated
in Figure 3. Although, the presented case here can be conceptually extended to a group of producer and
injector wells connected to any complex surface facilities networks, a simpler case as used in this study can
give us a better perspective and more in depth understanding of the new optimization strategy proposed in
this study and issues associated with it. The well trajectories specifications along with their equipment and
perforation sections are shown in Appendix A. These wells are connected to a surface facility which supports
both gas recycling and also gas-lift processes. All the related specification for modelling of surface facility
network are summarized in Appendix B. The fluid from producer is directed eventually to a three-phase
separator where the fluid is flashed at lower pressure and moderate temperature conditions. The produced
water and oil phases are transferred to their individual sinks. The gas phase, on the other hand, is split into
two streams, one used for gas-lift operation on the same producer and the other stream is recycled into the
reservoir via the injector well. The gas-lift operation increases the oil production by reducing the density
6 SPE-199923-MS

of the reservoir fluid leading to a reduction in the hydrostatic pressure of the fluid column in the well.
Therefore, the well can operate at lower bottom-hole pressure which in turn leads to more fluid production
due to more flow potential difference around the wellbore. On the other hand, gas recycling maintains the
reservoir pressure and reinforces the depletion driving force. Since the amount of produced gas is limited
and also changes by time, the optimization objective is to optimally split the produced gas at the distributor
to gain the maximum oil rate. In this case, the production term is defined to be 10 years and the decision
variable could be either the fraction of produced gas used for gas-lift process or for gas recycling.

Figure 3—Surface Facility of case (I)

Before starting the optimization process, a sensitivity analysis was performed to find out the effect of the
split ratio on the ultimate oil production with the results shown in Figure 4. The results of sensitivity analysis
showed that the amount of gas used for the gas-lift or re-injection has a significant impact on the final oil
production of the asset. It also showed that an increase in the fraction of the produced gas re-injected into
reservoir would lead to an increase in the oil production. This indicates that by only relying on sensitivity
analysis results to decide what portion of the gas should be injected to the reservoir, one could conclude that
about 60% to 70% of the gas should be recycled back to the reservoir as shown in Figure 7. On the other hand,
using an optimization strategy provides higher oil recovery compared to sensitivity analysis and suggests
different philosophy of allocating the produced gas for recycling and gas-lift process during production life
of the asset as shown in Figure 5. Figure 6 shows that the optimum fraction of gas to be split between gas
recycling and gas lift processes. In the first half of production life, to compensate for the hydrostatic pressure
of producing fluid in producer well, the optimizer finds that it is more optimal to dedicate significant portion
of the produced gas to the gas-lift process. However, as the reservoir depletes and more gas is evolved in
the reservoir, the produced fluid becomes lighter and therefore the gas-lift process loses its importance and
pressure maintenance plays a more dominant role in providing higher oil rate production. Figure 7 illustrates
that the optimized solution provides higher oil production.
SPE-199923-MS 7

Figure 4—Oil production sensitivity analysis with respect to split fraction of produced gas between
gas-lift and gas recycling (legends indicates the fraction of gas recycled back to the reservoir); case I.
8 SPE-199923-MS

Figure 5—Oil rate production after optimization of fraction of produced gas to be recycled into reservoir; case I
SPE-199923-MS 9

Figure 6—Optimized split fraction of produced gas (case I)

Figure 7—Ultimate oil production vs. recycled fraction of produced gas

compared with oil production after optimization (red dash line); case I

Case (II): In this case the asset is built exactly the same as the asset in case (I) except that additional
time-dependent constraints are imposed on the available gas to be split between gas recycling and gas-lift
processes. Referring to Figure 8, a gas sink is added to the surface network immediately after the separator.
A set of 6-month interval periodic gas sale constraints is defined on the inlet of this sink, imposing more
demand of gas in cold seasons of the year as shown in Figure 9. Considering this extra need for the produced
gas, both the sensitivity analysis and optimization solution, similar to case (I), are applied to this case with
results shown in Figure 10 to Figure 12. The results show a clear dependency of oil production on split ratio
10 SPE-199923-MS

of available gas between recycling and gas-lift process. Figure 12 shows that based on sensitivity analysis the
maximum oil recovery take place if ∼60% of available gas is re-injected back into the reservoir. However,
the optimized split fraction yields higher oil production. Furthermore, the split fraction ratios of produced
gas, as shown in Figure 13, reveals interesting behaviors that the entire cyclic change on available gas,
caused by periodic gas sale constraints, is absorbed by recycling fraction. This intermittent sudden change
in the available gas did not impact the gas-lift fraction. In fact, the optimizer maintains the required gas for
gas-lift intact similar to what was required for case (I). Regardless of how much gas is available and how it
fluctuates with time, optimizer puts more priority on gas-lift process in the first half life of production.

Figure 8—Surface Facility of case (II)

Figure 9—Periodic gas sale constraint in case (II) and case (III)
SPE-199923-MS 11

Figure 10—Oil production sensitivity analysis with respect to split fraction of produced gas between
gas lift and gas recycling; case (II) (legends indicates the fraction of gas recycled back to the reservoir)
12 SPE-199923-MS

Figure 11—Oil rate production after optimization of fraction of available gas to be recycled into reservoir; case (II)
SPE-199923-MS 13

Figure 12—Ultimate oil production vs. recycled fraction of produced gas

compared with oil production after optimization (red dash line); case (II)

Figure 13—Optimized split fraction of produced gas (case (II))

Case (III): The asset in this case is built identical to the case (II) except that the maximum values on
periodic constraint sale are doubled as shown in Figure 9. The intention of studying this case is to see how
the magnitude of periodic changes (discontinuity) on available gas would affect the split ratios and also
overall oil production. The results are shown in Figure 14. Similar to the case (II), all the periodic changes
on the shortage of available gas is compensated for by the recycling fraction; the gas-lift fraction is kept the
14 SPE-199923-MS

same as previous two cases. In the same way, the optimizer prioritizes the gas-lift process over gas recycling
to achieve the maximum oil production.

Figure 14—Optimized split fraction of produced gas for case (III)

Conclusion
This study proves that the MADS algorithm as a derivative free optimization methodology is a suitable
approach for optimizing gas allocation in integrated asset modeling. To study the impact of initialization on
the optimization final results, the presented cases in this study were executed with different initial values
of decision variables while keeping the seeding strategy of the NOMAD constant. It was observed that the
proposed optimization strategy shows insignificant sensitivity to initial values of the decision variables with
similar convergence pace and behavior. The optimization was successful for the same objective function
but with different sets of decision variables. It was also observed that this type of optimization technic could
successfully tolerate discontinuity and noisy objective functions with periodic changes imposed on them
during the search for optimum solution. As the optimizer searches through the defined space, it could endure
failures of evaluating the objective function without any interruption in the whole process and revive itself
into the possible global solution.
This study shows that using sensitivity analysis or optimization approach to determine the gas allocation
in an asset can lead to different strategies for asset development and it is prudent to consider both modeling
approached before making any fundamental decision for future development of any asset. In the three
studied cases, the sensitivity analysis led to a constant gas recycling ratio of 60%-70% of available gas
during the entire production life. For the same cases, however, the optimization put more emphasis on gas-
lift process in the early depletion stage of the reservoir when the produced fluid is heavier. However, for
the second half life of the reservoir, the pressure maintenance consumed the larger portion of the available
gas. The results show that the difference between the operation decisions made using sensitivity analysis
approach and optimization approach could be significant in any case configuration and operation conditions.
Therefore, it would be crucial to incorporate the results of both approaches into the making decision process.
SPE-199923-MS 15

The cases studied in this research show, for an asset contacting both gas-lifted wells and injection wells
with limited available gas, the gas-lift operation should have a higher weight on early stage of production
life. Figure 15 illustrates that in all 3 cases, with and without limitation on available gas, the optimizer tends
to allocate the same amount of gas to the gas-lift process. It could be inferred that the operations that could
improve the well production immediately or in short term, should have higher priority rather than those
which affect the integrated system and their impacts will be felt on longer terms (e.g. pressure maintenance
via gas recycling).

Figure 15—The comparison of gas lift fraction in all cases

The ultimate oil recovery (10 year production term) of the 3 cases studied in this research is compared
in Figure 16. Overall, the recovery is low in all cases; however, the effect of limitation of available gas on
oil production is significant. In case (III), the amount of gas taken out from the system is higher than the
other cases and the final recovery is lower.

Figure 16—Ultimate oil recovery of all cases

16 SPE-199923-MS

References
Abramson, Mark A., G. Couture Charles Audet, John E. Dennis Jr, and and C. Tribes. Sébastien Le Digabel. 2011. "The
NOMAD project Technical report". Project.
Audet, Charles, and Warren Hare. 2017. Derivative-free and blackbox optimization. Springer.
Beggs, H. Dale. 1991. Production optimization using nodal analysis.
Buitrago, S., E. Rodriguez, and D. Espin. 1996. "Global optimization techniques in gas allocation for continuous flow gas
lift systems." SPE gas technology symposium. Society of Petroleum Engineers.
Davidson, Jeffrey E., and Bret L. Beckner. 2003. "ntegrated optimization for rate allocation in reservoir simulation." PE
Reservoir Simulation Symposium. Society of Petroleum Engineers.
Dutta-Roy, Kunal, and James Kattapuram. 1997. "A new approach to gas-lift allocation optimization." SPE western
regional meeting: Society of Petroleum Engineers, Jan 1.
Fujii, Hikari, and R. N. Horne. 1995. "Multivariate optimization of networked production systems." SPE Production &
Facilities 165–171.
Hiebert, Allan David, Mohammad Khoshkbarchi, Peter H. Sammon, Ibere Nascentes Alves, and Alexander Jeroen Belien,
Benjamin Howell, Fredrik E. Saaf, and Per Valvatne Jose Rodrigues. 2011. "An Advanced Framework for Simulating
Connected Reservoirs, Wells and Production Facilities." SPE Reservoir Simulation Symposium: Society of Petroleum
Engineers.
Hong, Harry T. 1975. "Effect of the variable on optimization of continuous gas lift system". PhD diss. University of Tulsa.
Kanu, Eni P., Joe Mach, and Kermit E. Brown. 1981. "Economic approach to oil production and gas allocation in
continuous gas lift." Journal of Petroleum Technology (Society of Petroleum Engineers) 33: 887,892.
Le Digabel, Sébastien. 2011. "Nonlinear optimization with the MADS algorithm." ACM Transactions on Mathematical
Software (TOMS) (ACM) 37 (4): 44.
Nishikiori, Nobuo, RA Redner, DR Doty, and Z Schmidt. 1989. "An Improved Method for Gas Lift Allocation
Optimization". Society of Petroleum Engineers, 8–11 October.
Odeh, A, and S Aziz. 1981. "Comparison of solutions to a three-dimensional black-oil reservoir simulation problem."
Journal of Petroleum Technology 33: 13–25.
Rashid, Kashif. 2010. "Optimal allocation procedure for gas-lift optimization." Industrial & Engineering Chemistry
Research 2286–2294.
Rashid, Kashif, William Bailey, and Benoît Couët. 2012. "A Survey of Methods for Gas-Lift Optimization." Modelling
and Simulation in Engineering (Hindawi Publishing Corp.) 24.
Stoisits, RF, KD Crawford, DJ MacAllister, MD McCormack, AS Lawal, and DO Ogbe. 1999. "Production optimization at
the Kuparuk river field utilizing neural networks and genetic algorithms." SPE mid-continent operations symposium.
Society of Petroleum Engineers.
Vázquez, Manuel, Alexis Suarez, Hugo Aponte, Leonardo Ocanto, and José Fernandes. 2001. "Global optimization of oil
production systems, a unified." SPE annual technical conference and exhibition. Society of Petroleum Engineers.
Wang, Pengju, Michael Litvak, and Khalid Aziz. 2002. "Optimization of production operations in petroleum fields." SPE
annual technical conference and exhibition. Society of Petroleum Engineers.
Weiss, JL, WH Masino Jr, GP Starley, and JD Bolling. 1990. "Large-scale facility expansion evaluations at the Kuparuk
river field." SPE California Regional Meeting. Society of Petroleum Engineers.
SPE-199923-MS 17

Appendix A:

Figure 17—Well Trajectories; equipment and perforation

Table 1—Well modeling specifications

18 SPE-199923-MS

Appendix B:
Table 2—Pipe modeling specifications referring to Figure 3 and Figure 8

Table 3—Valve and compressor specifications referring to Figure 3 and Figure 8