Petroleum Experts Limited

11. - Workshop

- 73 –
Petroleum Experts Limited, Spectrum House, 2 Powderhall Road, Edinburgh, Scotland EH7 4GB.
Tel (44 131) 474 7030; Fax (44 131) 474 7031 ; email : [email protected]
Houston: Tel (1 281) 531 1121; Fax (1 281) 531 0810; email : [email protected]

IPMTCN. 5.00.21.07.2005
 Petroleum Experts Limited
There has been a new discovery and few data available on the discovery. You
want to study on basis of this information the best way to manage this field or
whether they proceed with the development of the field. The Platform
maximum capacity is 35000 bbl/d and the target Recovery Factor are 22% for
Res 1 and 35% for Res 2 in 10 years

Platform @
50000 ft
Sea level
60 F

600 ft

40 F
RES-1 RES-2

10000 ft

FIELD DESCRIPTION
Two subsea reservoirs in 600 feet of water.
Reservoir 1 is 10,000 feet away from reservoir 2.
Separator is 50,000 feet away.
Data: Reservoir 1 Reservoir 2
OOIP 150 85 (MMSTB)
Pressure 6500 11000 (psig)
GOR 500 1700 (scf/STB)
API 35 40
Gas gravity 0.7 0.72
Res Depth 14000 15000 (feet)
Permeability 50 500 (md)
Pay height 25 50 (feet)
Porosity 0.15 0.25 (fraction)
Swc 0.15 0.15 (fraction)
Salinity 100,000 10,000 (ppm)
Temperature 200 250 (F)
Wellbore radius 0.354 0.354 (feet)
Drainage Area 100 100 (Acres)

Res sat End Point Corey Exp

Krw: 0.15 0.7 0.8
Kro: 0.15 0.8 1.5
Krg: 0.02 0.9 1

- 74 –
Petroleum Experts Limited, Spectrum House, 2 Powderhall Road, Edinburgh, Scotland EH7 4GB.
Tel (44 131) 474 7030; Fax (44 131) 474 7031 ; email : [email protected]
Houston: Tel (1 281) 531 1121; Fax (1 281) 531 0810; email : [email protected]

IPMTCN. 5.00.21.07.2005
 Petroleum Experts Limited
GAP constrained Network Optimisation

The constrained optimisation module in GAP alters choke settings on naturally

flowing wells and gaslift amount on gaslifted wells to achieve maximum oil production
whilst meeting constraints – usually processing limits – placed at various levels in the
gathering system.

The Successive Linear Programming Algorithm

Firstly, consider a system (Fig 1) consisting of two wells connected via a manifold
and a pipework to a separator

Fig 1

The wells are naturally flowing, and their unchoked production is:

Oil Water Gas

Water Cut GOR
Production Production Production
Well 1 5000 1250 5 20 1000
Well 2 4000 3017 2 43 500
Totals 9000 4267 7

Processing limits at the manifold of 2500 STB/d water and 3.8 MMscf/day gas are
given, and the task is to choke back the wells to meet these limits in an optimal way,
where optimal in this case is define as maximising oil production.

This reduces to a mixing problem of a type frequently seen in all forms of process
industry, since each well supplies oil gas and water in its own particular proportions
- 75 –
Petroleum Experts Limited, Spectrum House, 2 Powderhall Road, Edinburgh, Scotland EH7 4GB.
Tel (44 131) 474 7030; Fax (44 131) 474 7031 ; email : [email protected]
Houston: Tel (1 281) 531 1121; Fax (1 281) 531 0810; email : [email protected]

IPMTCN. 5.00.21.07.2005
 Petroleum Experts Limited
and we are mixing the wells at the manifold. There are therefore many techniques
available for solving problems of this type.

Before we look at the actual technique used in GAP, let us solve the problem
manually. We therefore define x1 and x2 as the fraction of unchoked production from
each well, o1 and o2 as the unchoked oil production, w1 and w2 as the unchoked
water production and finally g1 and g2 as the unchoked gas production. The problem
can be stated with the following equations:

Maximise oil production = x1.o1+x2.o2

Subject to the constraints water production = x1.w1+x2.w2 <=2500
Gas production = x1.g1+x2.g2 <= 3.8
Non negative production = x1,x2 >= 0
Production <=unchoked = x1,x2 <= 1

We can draw a diagram of the problem (Fig 2) by plotting x1 along the x axis and x2
along the y axis. Any point within the square region defined by the 0,1 limits (OAHD)
represents a possible mixture of the two wells.

x2

F
J
L

H
A
E

D x1
O G I
Fig 2

Taking the water constraint first, if we plot a line on the diagram where the water
production from both wells adds up 2500, we get the line EI on the diagram. Any
point on or below this line will satisfy the water production constraint. We plot a
similar line representing the gas constraint, line FG. The lines intersect within the
square at point B. To satisfy both constraints, a point has to lie below or on both lines
simultaneously. The region containing all the possible mixtures which satisfy all the
constraints is the four sided region OEBG, the feasible region.

- 76 –
Petroleum Experts Limited, Spectrum House, 2 Powderhall Road, Edinburgh, Scotland EH7 4GB.
Tel (44 131) 474 7030; Fax (44 131) 474 7031 ; email : [email protected]
Houston: Tel (1 281) 531 1121; Fax (1 281) 531 0810; email : [email protected]

IPMTCN. 5.00.21.07.2005
 Petroleum Experts Limited
Now consider the oil production. Two lines representing all mixtures of the two wells
which produce 5000 and 4000 (JK and KL) are plotted. The 5000 line lies entirely
outside the feasible region, while the 4000 line divides it. If we visualise moving a
production contour from 5000 to 4000, it can be seen that the contour will first touch
the region at point B. This therefore must be the optimal point, since all points in the
region below must have lower oil production.

At point B, the actual production is 5031, and both the gas and water production are
at their limits. This corresponds to a fraction x1=0.513728 of unchoked production for
Well 1, and x2-0.615679 of unchoked production for Well 2. Note that the
combination of constraints has led a solution where both wells are choked.
Since we have performed curves for the wells relating production to WHP, we can
immediately look up the desired WHP for each well. This in turn gives us the choke
settings (as pressure differences), since they must equal the difference between the
manifold operating pressure and the desired WHP.

To get to this stage, we used the production data and constraints to form a set of
linear equations, and solved then simultaneously with a graphical method. It should
be clear that we can construct a similar set of equations for any system of naturally
flowing wells, with constraints at different levels affecting all of the wells connected
below. Since the equations are linear, this can be classed as a linear programming
problem, and GAP solves this using the simplex method, since this is reliable and
computationally efficient.

The Simplex Method.

A property of linear programming problems is that the solution always occurs on the
boundary of the region enclosed by the problem constraints, where two or more
constraints meet (i.e. a vertex of the region). Let us take a problem with N variables
(i.e. N wells) and M constraints.

To solve the problem therefore, we need to step through the points at the vertices,
ending with the point whose objective value is the highest. The simplex method is a
procedure which ensures that the objective increases at each step, and that the
optimum point is reached after a number of steps of order N (or M, whichever is
larger).

The first step is to express the system of equations in a standard form as follows:

z- 5000.x1 -4000.x2 =0 :Objective function

1250.x1 +3017.x2+y1 = 2500 :constraint 1
5.x1 +2.x2+y2 = 3.8 :constraint 2
x1 +y3 =1 :constraint 3
x2 +y4 =1 :constraint 4

They yi’s are called slack variables and are introduced to transform the inequality
constraints to equality constraints. All the variables are defined to be non-negative.
We now form a matrix representation of the equation (the ‘tableau’):

- 77 –
Petroleum Experts Limited, Spectrum House, 2 Powderhall Road, Edinburgh, Scotland EH7 4GB.
Tel (44 131) 474 7030; Fax (44 131) 474 7031 ; email : [email protected]
Houston: Tel (1 281) 531 1121; Fax (1 281) 531 0810; email : [email protected]

IPMTCN. 5.00.21.07.2005
 Petroleum Experts Limited
z x1 x2 y1 y2 y3 y4
Row 0 1 -5000 -4000 0 0 0 0 0
Row 1 0 1250 3017 1 0 0 0 2500
Row 2 0 5 2 0 1 0 0 3.8
Row 3 0 1 0 0 0 1 0 1
Row 4 0 0 1 0 0 0 1 1

As a starting point, take x1 and x2=0. This satisfies all constraints. We now want to
take a step which increases the objective. Choose the variable which has the largest
negative coefficient in row 0 – in this case x1. Let x2 stay at zero. As we increase x1,
the variables y1 will change value, but by definition, we cannot allow any y1 to
become negative. Taking each y1 in turn (and setting x2 to zero), we can say –

y1=2500-1250.x1 => y1 >= 0 for x1 <= 2500/1250 = 2

y1=2500-1250.x1 => y1 >= 0 for x1 <= 2500/1250 = 2
y2=3.8 – 5.x1 => y2 >= 0 for x1 <= 3.8/5 = 0.76
y3=1 – x1 => y3 >= 0 for x1 <= 1/1 = 1
y4=1 – x2 => y4 >= 0 for x1 <= any value of x1

Where the upper limits on x1 are simply the ratio of the limit to the coefficient. Clearly
we can set x1 to min{2500/1250, 3.9/5, 1/1} = 0.76, which will set y2 to zero. This is
called the ratio test, and in this case, row 2 (i.e. the y2 row) is the winner. We now
use matrix arithmetic to ‘pivot’ around the intersection of the x1 column and the y2
row, in other words use matrix arithmetic (row operations) to make the coefficient of
x1 1 in row 2 and 0 in all other rows. That is, we divide row 2 by 5 to make the row 2
coefficient 1, subtract 1250 times row 2 from row 1 to make the row 1 coefficient 0,
and so on. This yields the following tableau

z x1 x2 y1 y2 y3 y4
Row 0 1 0 -2000 0 1000 0 0 3800
Row 1 0 0 2517 1 -250 0 0 1550
Row 2 0 1 0.4 0 0.2 0 0 0.76
Row 3 0 0 -0.4 0 -0.2 1 0 0.24
Row 4 0 0 1 0 0 1 1 1

The step has made x1 non-zero, and the equation represented by the top row is now:
z – 2000.x2 + 1000.y2 = 3800, or z = 3800 + 2000.x2 – 1000.y2

- 78 –
Petroleum Experts Limited, Spectrum House, 2 Powderhall Road, Edinburgh, Scotland EH7 4GB.
Tel (44 131) 474 7030; Fax (44 131) 474 7031 ; email : [email protected]
Houston: Tel (1 281) 531 1121; Fax (1 281) 531 0810; email : [email protected]

IPMTCN. 5.00.21.07.2005
 Petroleum Experts Limited
Clearly, the next step in increasing z is to set y2 to zero, and increase x2. To find out
how much, we perform another radio test, using the four variables, y1, x1, y3 and y4 :

y1 = 1550 – 2517.x2 => y1 >= 0 for x2 <= 1550/2517 = 0.615812

x1 = 0.76 – 0.4.x2 => y2 >= 0 for x2 <= 0.76/0.4 = 1.9
y3 = 0.24 + 0.4.x2 => y3 >= 0 for any value of x2
y4 = x2 => y4 >= 0 for any value of x2

So, we can set x2 to min {1.9 , 0.615812 } = 0.615812, which will make y1 zero. We
now pivot around the y1 row (row 1) and the x2 column getting

z x1 x2 y1 y2 y3 y4
Row 0 1 0 0 0.794597 801.3508 0 0 5031.625
Row 1 0 0 1 0.000397 -0.09932 0 0 0.615679
Row 2 0 1 0 -0.00016 0.23973 0 0 0.513728
Row 3 0 0 0 0.000159 -0.23973 1 0 0.486272
Row 4 0 0 0 -0.0004 0.099325 0 1 0.384321

The step has made x2 non zero, and the equation represented by the top row is now:
z = 5031.625 – 0.794597.y1 – 801.3508.y2

We can stop here, since the only way to maximize z is to set y1 and y2 to zero. Thus
the optimal solution is x1=0.615679 , x2=0.513728 , z=5031 as we obtained
graphically above. A value of zero for y1 and y2 means that these constraints are
binding.

A good textbook on linear programming and the simplex method is [Introduction to

mathematical programming, W.L. Winston, Duxbury Press]

Special steps must be taken to cope with the situation where the back pressure
caused by a group of strong wells kills or severely attenuates production from a
weaker group, and the operator wishes the weaker wells to achieve a minimum
production. In practice, this minimum must be achieved by choking back the stronger
wells, thus reducing the back pressure and allowing the weaker wells to flow. The
algorithm described above, however, is not aware of the minimum production set by
the operator. GAP therefore detects when conditions could allow this, and sets the
weaker wells production level, artificially, to the minimum before performing the
simplex step. This allows the simplex algorithm to attenuate the stronger wells in
favour of the weaker ones.

We now have choked settings, but in general we would not obtain correct results if
we applied them and performed a network pressure calculation. This is because the
choke settings were based in the operating pressures as calculated for the unchoked
system. In general, these pressures will change as the wells production is modified,
because the pressure drops in the surface network will change. GAP therefore
performs a loop, calculating and applying choke settings as described above, then
- 79 –
Petroleum Experts Limited, Spectrum House, 2 Powderhall Road, Edinburgh, Scotland EH7 4GB.
Tel (44 131) 474 7030; Fax (44 131) 474 7031 ; email : [email protected]
Houston: Tel (1 281) 531 1121; Fax (1 281) 531 0810; email : [email protected]

IPMTCN. 5.00.21.07.2005
 Petroleum Experts Limited
performing a network pressure calculation until a stable situation is reached. This is
essentially linearising an inherently non-linear problem, and is known as a
Successive Linear Programming (SLP) approach.

The Nonlinear Programming Algorithm

The SLP algorithm described above can be applied to systems of naturally flowing
wells because the equations which describe the objective and the constraints are all
linear, allowing the use of the simplex algorithm at each stage. When we consider
systems containing gaslifted wells, however, this approach breaks down because the
gaslift constraint, i.e. that the sum of the individual wells gaslifts be limited to the gas
available, cannot be expressed as a linear equation involving the fractional products.
The problem thus unavoidably becomes a non-linear programming one (NLP). GAP
still uses the overall loop described above, but replaces the simplex step with a NLP
solution technique.

This is in fact requires a much more complex algorithm than simplex, since there are
many special cases and problems which have to be catered for, and the solution
almost always requires a significant number of search steps, complex logic to
maintain an ‘active constraint set’, as well as the gradients and even second order
derivatives of the objective function and the constraints. For descriptions of many
common techniques, see [nonlinear programming, M.S. Bazaraa, H.D. Sherali &
C.M. Shetty, Wiley]. Briefly, GAP solves the non-linear programming problem by
forming a quadratic approximation to the Lagrangian of the problem, solving this
approximation to obtain a direction to step in, then maximizing a merit function (The
L1 penalty function) along that direction, where the merit function measures both the
increase in the objective and the violation of the constraints.

These steps are repeated until an optimum is found. This can be quite
computationally expensive for some systems. The output of this step is a set of
desired productions for naturally flowing wells, and a set of gaslift amounts for
gaslifted wells. The naturally flowing wells then have their choke settings calculated.
If the algorithm decides to attenuate a gaslifted wells production, the well is
converted to a naturally flowing well and choked. To ensure a smooth transition from
gaslifted to choked regimes, which is required by the nonlinear programming model,
a blending function is used to merge the two.

As above, this step is followed by a network pressure calculation, and the two steps
are repeated until stability is reached.

- 80 –
Petroleum Experts Limited, Spectrum House, 2 Powderhall Road, Edinburgh, Scotland EH7 4GB.
Tel (44 131) 474 7030; Fax (44 131) 474 7031 ; email : [email protected]
Houston: Tel (1 281) 531 1121; Fax (1 281) 531 0810; email : [email protected]

IPMTCN. 5.00.21.07.2005