Professional Documents
Culture Documents
This paper was prepared for presentation at the SPE Asia Pacific Oil & Gas Conference and Exhibition held in Brisbane, Australia, 23–25 October 2018.
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
Several gas fields in South East Asia share some common traits among them, obviously on their geological
features but also on their complex field operation. With a large number of small gas accumulations spreading
across a large area with high degree of lateral compartmentalization, production from these fields are usually
accomplished by hundreds of wells through multi-branches field networks. The scope of this paper is to
present the challenging journey of the company's in-house innovative methodology which resulted in the
development of a robust software to address the above challenges. The main objective of the software is to
optimize field production under numerous constraints present in these fields.
With the target to optimize field production and enhance predictive capability, the company integrates
the experiences from operating several fields and proposes an innovative approach to tackle these field
management challenges. The resultant software optimizes and solves the network calculation by simplifying
and formulating the production network into a system of linear equations, then applying optimization
techniques as large-scale simplex and mixed-integer linear programming algorithms, to search for the best
production scheme while taking user-selected objective function into consideration. The workflow was
developed using MATLAB optimization toolbox to work in conjunction with a familiar Excel-formatted
input. Moreover, with the incorporation of the Decline Curve Analysis (DCA), it is also applicable for
generating long term production forecast. The tool was further combined with Production Data Management
System (PDMS) to provide a more efficient automated workflow. It was used to maximize condensate
production in Arthit field, where the main constraints are to capture the production loss from CO2 removal
unit and to limit mercury concentration in sales condensate. While, in Zawtika field, the application exploits
quadratic programing to minimize the sum of gas production rate square hence controlling wells to produce
at optimal rate, mitigating sand production problem.
In this paper, successful implementation examples and benefits gained will be discussed. It ensures that
the condensate production in Arthit field is kept at optimal level compared with about 91% efficiency when
subjected to conventional practices while, in Zawtika, applying the workflow and operation resulted in
dramatically lower sand production problem. For future forecast, a look-back study was performed to make
sure that the method of calculating future potential is accurate. Not only does this new tool provided a
2 SPE-192080-MS
more efficient way for the teams to manage their assets but, more importantly, it also helps to save costs by
reducing man-hours through its rapid computation.
Introduction
In operating any gas fields within PTTEP, there are three production prediction practices that must be
performed regularly: snapshot optimization; medium-term forecast; long-term forecast. In the following
Snapshot Optimization
The objectives of this process are to evaluate the gas deliverability of a field while honoring every
operational constraint and to find optimal production scheme. Typically, after collecting potential of each
well from production tests, engineers will rely on trial and error process or some basic solvers in MS Excel
spreadsheets to find the suitable scheme. However, due to the limitation and the empirical nature of the
process, the result was never fully optimized and subjective to the person performing the task. Moreover,
since these fields are produced with a large number of producers, using commercialized software to perform
optimization is tedious and time-consuming.
Medium-Term Forecast
This is to assure that gas deliverability can be maintained above the Daily Contractual Quantity (DCQ) for
the next 6-12 months with the current operation schedule. Figure 1 depicts the example of a medium-term
profile which is subjected to an arbitrary well intervention schedule. The conventional process primarily
relied on performing DCA of all wells. Then the DCQ would be achieved by trying to produce wells
proportionally without selective production. Therefore, it was laborious and did not capture the nature of
field management in which wells are selectively produced based on their values in petroleum fluids.
(2)
(3)
(4)
(5)
(10)
By firstly obtaining the nominal decline rate (d) from the current reserves of each producing unit and its
respective production rate, at the end of each time step, we can update our parameters as follows:
• Update the remaining reserves by subtracting the production of the previous time step, the result
of the linear programming problem, from the primary reserves
SPE-192080-MS 5
• Update the potential of the well by using Eq. 10 where Gp and qi are the production and the potential
of the previous time step, respectively
Wherever the history of the field suggests that condensate-gas ratio (CGR) and water-gas ratio (WGR)
tend to change with cumulative gas production (Yu 2014), equations similar to Eq. 10 for those two
parameters can be constructed and applied in the same manner. The processes to update gas potential, CGR,
Figure 3—The process to update gas potential (qg, max), CGR, and WGR when the cumulative
production from the unit increases from Gp, t at time step t to Gp, t+1 at time step t+1
Moreover, whenever field operations, drilling or well services operation, are expected to be performed,
gas potentials together with the fluid characteristics of those wells can be added to the linear programming
problem (newly drilled wells) or updated accordingly (intervened wells). The workflow of a single time
step snapshot optimization and production forecast is illustrated in Figure 4 below.
(11)
After relying on production plan from LINOPT, the condensate production would reach what should
really be the true productivity of the fields.
Achievability of the target from LINOPT. For any software, another point that should be validated is the
practicability of the result. In this case, we need to ensure that the condensate target from LINOPT that we
plan to produce is really obtainable with real field practices. In other words, the scheme provided from the
software and the predicted result is pragmatic. Therefore, after we applied LINOPT, the actual condensate
production is compared with the target resulted from LINOPT as per Eq. 12.
(12)
The achievability over 100% shouldn't pose any problems since it can be the case when the max potential
of some wells are not fully captured. However, the figure much lower than 100% might suggest either the
appearance of some unexpected operational problems or the impractical production target from the software.
Potential forecast error. Another aspect that should be validated is the future potential forecast method
on which the medium-term and the long-term forecasts are based. This is achieved by comparing the gas
potential from medium-term forecast done in the past with the actual potential realized at each time step as
per Eq. 13 where Q*g denotes gas potential.
(13)
Man-hour reduction. Implicitly, the man-hour spent on any tasks is subjective to the ability and the
experience of each operator. In eliminating possible biases, each field and for each type of forecasts, the
persons generally do the tasks are asked to perform production forecasts with both the conventional approach
and LINOPT separately. Then, the amount of time spent with each methods is compared and reported. This
process is meant to roughly evaluate the program numerically in term of how it makes the process become
easier and faster.
SPE-192080-MS 7
Arthit
One of the challenges in this area is that gases from several reservoirs in different areas show much higher
CO2 content than our sales-gas specifications. As a result, CO2 removal unit was implemented requiring
operators to control the feed CO2 level at a specific value in order to serve the CO2 specs at sales condition
(Suebsook et.al 2011). Producing it at lower than this value of feed CO2 may cause loss in value if some
reservoirs with higher CO2 are left unblended and non-producible. In contrast, producing at higher than this
feed specification lessen the value of the sales gas.
In addition, the condensate from this field shows a wide range of mercury (Hg) content
(Charoensawadpong et.al 2018). Therefore, to meet the sales-condensate specifications, condensate
blending has to be performed.
Software Application. In Arthit field, the optimization objective is to maximize the BOE while maintaining
DCQ resulting in the maximum revenue. Eq. 14 is the formula for the objective function, where Conv is the
unit conversion factor to convert from BTU to BOE. Since generally the GCV for each well is quite close
to that of the others, with this objective function, condensate production would be maximized as well.
(14)
For constraints, other than general pipeline constraints that must be obliged in every asset, this field also
strives to control the Hg content of its condensate as well as CO2 in the gas stream. In order to limit the Hg
content, an equation similar to Eq. 15 can be used in the definition of linear programming problem.
8 SPE-192080-MS
(15)
Implementation Results.
1. Comparison between the condensate target from the conventional approach and the actual condensate
potential from LINOPT
For illustrating the benefit of the software in snapshot optimization, we collect the well data which we
In the figure, the sharp drop in condensate sales at the start of month 5 was due to planned shutdown for
maintenance. Compared with the actual production, the target from the conventional approach was generally
achieved with mean absolute percentage error (MAPE) of 4.84%. When compared with the actual potential,
the conventional approach tended to provide production scheme at only 91.39% of the true potential. This
gap represents the portion of the field management that could be optimized further by LINOPT.
2. Achievability of the target from LINOPT
After applying LINOPT in optimizing the field production, the condensate production and its respective
production target from LINOPT were collected over the six-month period and plotted in Figure 7.
SPE-192080-MS 9
As the figure above suggests, the condensate target from LINOPT which must be the maximum
possible condensate production at any time periods can be achieved successfully. Numerically, the target
achievability is 100.03% in average with the MAPE between the predicted rate and the actual sales of
6.52%. Therefore, it can be concluded that the production scheme provided from the software is reasonably
achievable.
3. Potential forecast error
In order to ensure that the method of potential decline is consistent with the reservoirs’ characteristics,
the potential profile generated from medium-term forecast in LINOPT is plotted together with the actual
gas potential at each period of time in Figure 8 below.
Figure 8—Art hit look-back study on accuracy of medium-term forecast compared with the actual data
The reason why the actual potential was generally lower than the medium-term potential is that, in
performing snapshot optimization, we must take some additional unexpected restrictions into account. This
includes emergency shutdown of some platforms, the element that is unpredictable, and hence hard to
capture for medium-term forecast.
Considering the overall result, even though the field, at this period of time, was subjected to lots of well
intervention activities every day, it can be observed that the general trend of actual gas potential is captured
10 SPE-192080-MS
in the medium-term forecast. This is the good indicator that the way that engineers calculate potential gains
or losses from each operation and also the method that the software uses to calculate potential decline are
accurate. The medium-term potential forecast error according to Eq. 13 during that period was on average,
5.76%.
4. Man-hour reduction
After using LINOPT in all three optimization tasks, the comparison of the time required to perform each
Figure 9—Arthit man-hour reduction from using LINOPT (green) instead of the conventional approach (blue)
For any of these three tasks, using LINOPT significantly shortens the time to perform optimization and
construct production profile. While the traditional snapshot optimization and medium-term forecast take
more time than that with the new software because they require some manual calculations, the long-term
forecast takes more time due to a more computationally expensive approach adopted by material balance
software than that of LINOPT.
Note that the reason why petroleum engineers usually take longer time in performing medium-term
forecast than the long-term forecast is that in medium term forecast, we are required to optimize drilling
and well services schedule, a trial-and-error process. Therefore, the task is more complicated and requires
more simulation cases than coming up with the long-term profile.
Zawtika
Field Overview. Zawtika field, a dry gas field in offshore Myanmar, started production in 2014 with DCQ
over 250 MMscf/D. While it is not relatively heterogeneous or geologically complicated compared with
the reservoirs in Arthit, it has several unconsolidated reservoirs resulting in high risk of sand production
(Junmano et.al 2016). Consequently, the main challenge here is to make sure that all wells, while serving
enough gas for DCQ, are producing at sand free rates. The Zawtika map is shown in the Figure 10.
SPE-192080-MS 11
Software Application. As per the nature of the unconsolidated sands, the maximum allowable production
rate for each well are constrained in the program in order to mitigate the sand problem. Moreover, the
optimization objective is set to minimize the sum of gas production rate square (Eq. 16 where i is an integer
from 1 to the number of wells n). This would further help us ensure that fewest wells are forced to produce
at high draw down pressures resulting from high production rates. Even though this objective forces us to
rely on quadratic programming instead of linear programming, it is easy to implement since the solver of
this programming problem is already available in MATLAB library.
(16)
Implementation Results. Since the reservoirs in this area are dry gas reservoirs, the objective function of
the optimization is not intended to maximize condensate production. Therefore, any evaluations involving
condensate production are not meaningful. Consequently, only mid-term potential forecast error to confirm
the validity of the decline approach and the man-hour reduction are analyzed. Since the final goal of
optimizing this field is to obtain the production scheme with lower sand problems, the sand production
result after implementing the software is duly addressed.
1. Potential forecast error
In evaluating the error of the gas decline method, as performed in Arthit, the potential profile generated
from medium-term forecast in LINOPT is plotted together with the actual gas potential in Figure 11 below.
12 SPE-192080-MS
During the year of observation, there were several efforts initiated to prolong the productivity of the field.
The increase in potential was therefore due to additional perforation. The medium-term potential forecast
error according to Eq. 13 during that time period was on a low average of 1.72%. This shows that both the
method to calculate the additional potential from future wells and the gas decline approach being used in
LINOPT are accurate.
2. Man-hour reduction
In this field, we have been implementing LINOPT in doing medium-term and long-term forecasts since
the beginning; therefore, the time spent on those forecasts with conventional approaches is not available.
The result is reported Figure 12.
Figure 12—Zawtika man-hour reduction from using LINOPT (green) instead of the conventional approach (blue)
Considering snapshot optimization task, using LINOPT shortens the time required significantly.
3. Effect of the optimization approach on sand production
In evaluating the effectiveness of the objective function to help mitigate sand production problem in this
field, the yearly average of the sum of square of daily gas production rate is plotted together with the sand
SPE-192080-MS 13
production rate in Figure 13. During this period, the field was produced at the same DCQ; therefore, the
reduction in this parameter was not due to the change in DCQ.
It is clear from the results that as the sum of square of daily gas production rate reduces, the sand
production rate reduces as well. Since late Year 2, when this production scheme has been strictly honored,
the sand production problem has been reported to be at minimum.
Conclusions
1. LINOPT software has been developed internally in PTTEP for years and now it has been used for
snapshot optimization and future production forecast in several gas fields in the Gulf of Thailand
where PTTEP is the main operator.
2. The core idea, which eventually leads to the success, is to formulate the field production system as a
linear optimization problem where a variety of production goal can be selected and most, if not all,
operational constraints can be taken into account.
3. Regarding the snapshot optimization, which is used to devise the daily production strategy, the
software is proven to provide a more optimal production plan than those from any conventional
approaches applied.
4. The production target obtained from the software is practical and achievable with normal field
practices.
5. For future production forecast, the implementation of the gas potential decline approach similar to the
exponential decline in DCA is justified by the low error in future potential forecast.
6. As a result of simple but robust software structure including user-friendly workflow, the man-hour
spent on each task can be reduced significantly.
Acknowledgement
This research was supported by PTTEP. We are truly grateful to PTTEP management for their continued
support and review on the project. We also would like to express our deepest gratitude and appreciation
to the following for their technical advice and assistance; Saran Sirimongkolrat, Sujaree Worapotpisan,
Mohamad Kamal Bin Hamdan, Sittichod Srihirunrusmee, Vartit Tivayanonda, Watanapong Ratawessanun,
and Chuayrach Wongrattananon, PTTEP.
14 SPE-192080-MS
Nomenclature
qg,i = gas production rate of ith well
qo,i = condensate production rate of ith well
WGRi = Water-Gas-Ratio of ith well
CGRi = Condensate-Gas-Ratio of ith well
LGRi = Liquid-Gas-Ratio of ith well
References
Arps, J. J. (1945, December 1). Analysis of Decline Curves. Society of Petroleum Engineers. doi:10.2118/945228-G
Charoensawadpong, P., Chatwarodom, P., Manit, P., Atibodhi, N., Yongmanitchai, M., Hayook, C., … Suwanvesh, K.
(2018, March 20). Unlocking Field Potential from High Mercury Wells: New Opportunities, Challenges and Solutions.
Offshore Technology Conference. doi:10.4043/28524-MS
John C. Nash, (2000, January 1). The [Dantzig] Simplex Method for Linear Programming. Computing in Science and
Engineering (CiSE) vol. 2. doi:10.1109/5992.814654
Junmano, T., Lee, B., Grant, G., Raipairi, P., Viriyasomboon, N., & Nopsiri, N. (2016, November 12). Sand Production
Management the Critical Challenge in Zawtika Gas Production. International Petroleum Technology Conference.
doi:10.2523/IPTC-18839-MS
Lo, K. K., Starley, G. P., & Holden, C. W. (1995, February 1). Application of Linear Programming to Reservoir
Development Evaluations. Society of Petroleum Engineers. doi:10.2118/26637-PA
Suebsook, J., Eksaengsri, A., Lertsakulsup, S., Suwanvesh, K., Iswigrai, J., & Sillapacharn, T. (2011, January 1).
Process Optimization in Acid Gas Removal System. International Petroleum Technology Conference. doi:10.2523/
IPTC-14641-MS
Yu, S. (2014, April 1). A New Methodology to Predict Condensate Production in Tight/Shale Retrograde Gas Reservoirs.
Society of Petroleum Engineers. doi:10.2118/168964-MS