Вы находитесь на странице: 1из 8

Risk Analysis of Drilling productivity.

1. Background

Drilling operations are subject to more uncertainty, in terms of time,


costs and schedule, than most other components of a project. Despite
this inherent risk, drilling personnel struggle to quantify and
communicate risk consistently, performance has therefore not always
been properly accounted for to acknowledge uncertainties that exist.

A risk analysis tool characterises project uncertainty based on the risks


present in the individual components of the project. In the context of
well construction in this review the project consists of a large sample of
exploration and multi-well development, essentially however, the same
process applies.

A simple process for risk analysis is thus presented.

2. Steps in the Process

2.1 Select Data set of Wells

A set of wells broadly similar was used.

The risk analysis was then based on the engineering experience and
knowledge of the wells. The similar nature of data set selected underpin
this process and provide an auditable backup.

2.2 Create a Well Plan

A plan simple listing the main activities required to achieve the well
objectives and the relevant depths was constructed.

Task Start End


Drill 36" Hole 36 1654
Install 30" Conductor
Drill 17 " Hole 36 1654
Install 13-3/8" Casing
Drill 12 " Hole 36 1654
Install 9-5/8" Casing
Drill 8-1/2" Hole 1654 3125
Log 8-1/2" Hole
Install 5 " Liner

1
2.3 Analyse Offset Performance

Drilling times on each data set of wells were broken down into
performance summaries at a level of detail which was considered fit for
purpose. This data was then used to estimate the duration of the
planned well.

Non-productive time (NPT) including waiting on weather and major


sources of NPT due to single incidents (train-wrecks) were stripped out.

A train-wreck is defined as any single incident which exceeds 15% of


total well duration.

NPT and train-wrecks are included as specific line items in the risk
analysis and are treated separately for two main reasons:

they are both major sources of uncertainty (see 2.7),

their removal produces more consistent productive times which, as a


result, are easier to model.

Due to the nature of the working environment being reviewed, it is


necessary to include waiting on weather as a third component of lost
time, along with NPT and train-wrecks.

The tables below summarise the performance on actual wells from,


Block 204 Exploration wells, Foinaven and Schiehallion developments.

The figures for the hole sections (36, 26, 17 12 and 8 ) are
expressed in metres per day (m/d). The performance was kept simple
to illustrate the risk analysis process. The train-wrecks both refer to
stuck pipe, casing and equipment related incidents and were by far the
largest single sources of NPT. Note that Total Days includes the train-
wreck duration.

Job A B C D E F
12-1/4" (m/d) 426 406 499 490 398 367
9-5/8" Casing 1.13 1.17 1.33 1.17 1.26 1.18
8-1/2" (m/d) 323 348 253 366 391 417
Log 1.50 1.46 2.52 1.54
7" Liner 0.86 1.19 1.17 0.98 0.96 1.02
%NPT 13.6 1.2 5.7 0.7 5.6 2.4
Total Days 14.57 20.88 21.42 13.89 13.73 13.31
Train-Wrecks 4.92 4.06

There is always some element of subjectivity involved when using a


selection of data to model future performance, e.g. changes in rig
specification, different mud systems, more challenging directional plans etc.

2
It is obviously important to allow for these when compiling the estimate,
however, it is equally important to allow for the uncertainty that such
adjustments may introduce into the estimate. This can be done by
careful selection of the distributions and parameter ranges, as discussed
in the following sections.

2.4 Define Statistical Distributions

This is the key step in the risk analysis process because it defines the
input distributions to the subsequent Monte Carlo simulation. Monte
Carlo simulation is the basis of most risk analyses and is an effective method for
modelling the uncertainty of real problems which are invariably too complex to
be solved analytically.

Based on the offset data, summarise each component of the forecast in


terms of a statistical distribution. Try to ensure that a sufficiently broad
range of outcomes is characterised by each distribution.

2.5 Monte Carlo Simulation

Most real-world problems involving elements of uncertainty are too


complex to be solved by strict analytical methods and so Monte Carlo
simulation is adopted instead. When you run a simulation, random
numbers are generated for each input variable according to its
underlying statistical distribution. Each iteration of random numbers
effectively represents a single what-if scenario. Running through a
large number of different random scenarios characterises the
uncertainty of total well time. The distribution of total time based on a
Monte Carlo simulation with 1,000 iterations is shown below for the
example dataset. The bimodal (two peaked) shape of the distribution
reflects the 33% chance of a train-wreck on the planned well.

Forecast: Well Time

1,000 Trials Frequency Chart 0 Outliers


.070 70

.053 52.5

.035 35

.018 17.5

.000 0

10.0 12.8 15.5 18.3 21.0

Conventionally, output distributions are summarised in terms of three


percentiles: P10, P50 and P90. The P10 is the time below which 10% of
values lie. The P50 value (or median) is the time below which and

3
above which 50% of values lie. The P90 is the time above which 10% of
values lie. In this case, the three percentiles are as follows.

P10 = 12.0 days


P50 = 13.8 days
P90 = 18.2 days

Note the skewness of the results, again reflecting the train-wreck risk.
The cumulative distribution is shown below.

Forecast: Well Time

1,000 Trials Cumulative Chart 0 Outliers


1.000 1000

.750 750

.500 500

.250 250

.000 0

10.0 12.8 15.5 18.3 21.0

It is common to take the P50 value as the time or cost estimate with the
P10 and P90 values characterising the uncertainty. However, be careful
when summing P50 values over a number of wells as the sum of many
P50 values may yield a total which is significantly less than a P50 due to
the skewness of the underlying distributions. This problem does not
occur when dealing with means and, for this reason, some practitioners
of risk analysis prefer the mean to the P50 (the GIAPPS process is
based on the mean). This problem can be avoided by setting up a
separate risk analysis for the whole well programme using the
uncertainty estimates of each individual well as the inputs. Be careful to
take account of dependency when building such a model (see 2.8).

2.7 NPT and Train-Wrecks

Invariably, the two most influential components of a drilling risk analysis


are NPT and train-wrecks. The amount of NPT on a series of apparently
similar wells typically varies wildly. Stripping out train-wrecks and
considering these separately greatly reduces this uncertainty but there is
still usually a need to include a wide range of NPT in the risk analysis
model.

It is also important to make allowance for train-wrecks because these


are the usual cause of drilling time and cost overruns. There are two
recommended approaches for incorporating the risk of train-wrecks: the
generic method and the specific method.

4
The Generic Method

According to this approach, train-wrecks on the offset wells are treated


as generic incidents. In other words, the various incidents are
considered as a homogeneous group and the probability of occurrence
and time impact derived accordingly.

The table below summarises, by area, actual train-wreck frequency and


impact based on 208 BPX wells drilled world-wide in 1996 and 1997. A
train-wreck is as defined in Section 2.3: any single incident which
exceeds 15% of total well duration. The impact is quoted in terms of
percentage of total well duration and the analysis refers to dryhole time
only.

Area Frequency (%) Impact (%)


UKCS 23 23
Overall 19 25

The Specific Method

In this approach, a list of potential risks is identified for the planned well.
This could be based on a detailed analysis of hazards experienced on
offset wells or on the results of a brainstorming session of engineers
familiar with the area to be drilled. In either case, the output is a list of
potential risks. The second stage is to go through each risk and
estimate the probability of occurrence and likely impact based on a
consensus of those involved. Multiplying the probabilities by the impacts
gives the risked impact, which may be used to eliminate risks of little
consequence. The remaining risks are then included in the risk analysis
in the manner described previously.

2.8 Dependency

Dependency, or correlation, takes into account relationships between


input variables. For example, it may be that the time to run 13-3/8
casing is linked to the time to run 9-5/8 casing in that poor 13-3/8
performance will, more likely than not, be followed by poor 9-5/8
performance.

Another example concerns the cost of a 10-well development. If the


cost estimate of one well turns out to be optimistic then the estimates of
the other nine well costs will also, more likely than not, be optimistic.
The same argument holds for estimates that are pessimistic. Combining
individual well costs without taking dependency into account will result in
an unrealistically tight uncertainty range for the total development cost.

Crystal Ball allows dependency to be simply defined using correlation


coefficients between input variables. As a rough guide, if dependency is
anticipated use a value of at least 0.75. For correlations between wells
on a multi-well development a figure as high as 0.9 is recommended.

5
2.9 Sensitivity Analysis

Sensitivity analysis allows the influence of each input variable on the


final outcome to be assessed. The analysis is conventionally
summarised in terms of a plot showing each input variable in decreasing
order of sensitivity as shown below for the example model described
previously. The results show that, as expected, the probability of a train-
wreck has the most influence on total well time, that is, whether or not
there is a train-wreck. The coefficient in the centre column of the plot is
a measure of the correlation between the input variables and total well
time. The sign (positive or negative) of this coefficient is unimportant
and simply indicates whether increasing the input variable produces
either an increase or a decrease in total well time.

The main advantage of sensitivity analysis is that it indicates which


variables have a major impact on project uncertainty and which have
little impact. The former may merit additional focus while the latter may
reasonably be ignored (e.g. 9-5/8 casing and 7 liner below).

Sensitivity Chart

Target Forecast: Well Time

Prob of Train-Wreck .70

8-1/2" Hole -.32

%NPT .23

12-1/4" Hole -.21

Log .16

Train-Wreck .12

9-5/8" Casing .08

7" Liner .08

-1 -0.5 0 0.5 1

Measured by Rank Correlation

6
2.10 Communication of Results

Having performed a risk analysis it is obviously important to


communicate the results effectively. Any report should include the P10,
P50 (or mean) and P90 values as a minimum. This should be
supported by a plot of the frequency distribution (top figure in Section
2.5). Other disciplines are familiar with the use of percentiles, for
example, reserve estimates are commonly quoted in terms of P10, P50
and P90 values. When discussing P10 and P90 results, remember that
there is a relatively high chance (20%) of being outside the range they
define. They do not represent minimum and maximum figures. If
appropriate, run the full economic model on the P10 and P90 outcomes
as well as the P50.

It may also be worth referencing other performance indicators such as


the technical limit, industry best in class, industry upper quartile and so
on. Benchmarking data from other operators is available on the Intranet
from both A D Little and Rushmore studies.

7
3. Bullets

q Base time estimates on actual offset data whenever possible and


avoid fudge factors that have not been calibrated against real
performance.

q Treat NPT and train-wrecks separately. Both are major sources of


uncertainty and their removal makes productive performance more
consistent and easier to model.

q Characterise drilling performance in terms of statistical distributions


of each component. Estimate uncertainties based on the offset data
and experience. If in doubt use triangular distributions.

q Allow for a wide range of NPT. Estimate the probability and impact
of train-wrecks using offset frequency, a detailed study of potential
risks or a combination of both.

4. Software Tools

Time Forecasting
Available in DEAP, automatically breaks down performance on wells
which have passed through Phase/Job Allocation. Uses resultant data
to generate a forecast for a planned well based on Monte Carlo
simulation. Very powerful tool but requires training to use effectively.

Crystal Ball
Inexpensive add-in to Excel which performs Monte Carlo simulation.
The program is supported by an excellent and easy to follow manual.

An Excel/Crystal Ball spreadsheet which performs the risk analysis of `


the example cited in this note is included below.

Вам также может понравиться