statistical techniques for analyzing production impacts of
TRANSCRIPT
Statistical Techniques for Analyzing
Production Impacts of Completion
Designs
REU Calgary May 26th, 2015
Introduction
Thank you REU Canada & Hanson Wade
About VISAGE
Presentation
Questions (jot down the slide title if possible)
VISAGE: Interactive Visual Analytics Software (designed and built for the Oil and Gas industry)
Today’s presentation
will use IHS data
with WCFD from CDL
Why are Completions Important to Evaluations?
Completion designs introduce complexity, uncertainty and
opportunity to reserve evaluations
Completions impact value in terms of EUR and the rate at
which volume is recovered (NPV, ROR, ROI …)
Important to consider when creating analogue type wells
for bids
But …there are challenges to measuring the production
impacts of completion design parameters
Presentation Overview
Demonstrate changes in completions over time and the
associated changes in production profiles (Montney Study
Area)
Techniques for quantifying production impacts of
completion designs
What “production measure” should I use?
Discuss statistical analysis techniques using distributions
and maps
Completion Design Trends in the Montney
Wells from Canadian Discovery’s Well Completion & Frac
Database designated as Montney Distal Gas Resource Play.
Montney Sample Area
BC Wells from Canadian Discovery’s Well Completion & Frac
Database designated as Montney Distal Gas Resource Play.
Montney Sample Area
Montney Sample Area (Open vs Cased)
Montney Sample Area (Technology Group)
.
Montney Sample Area (Frac Fluid System)
.
Cased Wells: Production/$K Completion Cost
Open Wells: Production/$K Completion Cost
No dramatic change since 2012
Compare Forecasts of Open and Cased
Consider the time-value of money, time to drill and complete, cost to drill and
complete, number of wells I can get on production in a year….
**Forecast uses Modified Duong to Arps. All post 2011 wells in Montney Sample Area have >80% correlation on both Duong fits.
How Do We Quantify/Compare Production Impacts of
Completion Design Parameters?
Dimensional Normalization of Production Data
establish comparative measures (e.g. prod/stage, prod/100m …)
Type Curves (Rate vs Time, Cum vs Time, Rate vs Cum)
a single value represents a broad range of values (i.e. over simplification)
issues with dropping well count and survivor bias
Cumulative Probability Distributions (Percentile)
choose a production measure at a point in time
use binning of various completion parameters
Linear Correlations
very weak on large data sets…weak to moderate on smaller data sets
there are many other uncertainties involved
linearity assumption
Weak Linear Correlation
Wells from Canadian Discovery’s Well Completion & Frac
Database designated as Montney Distal Gas with IP Gas and
Avg Proppant Placed per Stage values.
Would a smaller, focused data set help?
Would the correlation
improve if we focused in
on one township in the
Montney?
Limited Data Set (one township)
Township = 080-18-W6
Colour by Avg Proppant/Stage
Bubble Size by IP Gas/Stage
Correlation Improves to Moderate
Township = 080-18-W6
Distribution Theory
Central Limit Theorem suggests that:
Additive Uncertainties• tend towards a Normal Distribution
Multiplicative Uncertainties***• tend towards a Lognormal Distribution
*** reservoir properties and completion design parameters
How to Translate XY Plots to a Distribution …
Plot the
production
measure
Bin the data using
a completion parameter
Same Information Displayed as a Distribution
Frac Analysis in VISAGE: Using Distributions as an Alternative to Linear Regressions
Frac Analysis in VISAGE: How to Refine Your Insights Using Distributions
Distributions may reveal a “Correlation Window”
Frac Analysis in VISAGE: Using Distributions as an Alternative to Linear Regressions
Frac Analysis in VISAGE: How to Refine Your Insights Using Distributions
“Correlation Window” = the range of values
where the strongest relationship exists
between two variables
Why are Distributions so Useful?
Ability to compare different sample sizes
Communicate the range of uncertainty (P10/P90)
Relative alignment of values for quantitative
comparison (e.g. low values relative to low values)
May illustrate a “Correlation Window” (the range of
values where the strongest relationship exists between two
variables)
What should I use distributions for?
1. Reality Check• completion costs, drilling costs, expected peak rates etc.
2. Vintage Trending• industry learnings (avg frac spacing, # of stages, proppant
loading, costs etc.)
• competitor analysis, play analysis
3. Testing Assumptions & Culling Data
4. Analyze how completions impact production…
What Production Measure Should I Use?
EUR
• new wells have insufficient data to forecast
• time consuming (if not automated)
• production profile is important to economic evaluation
IP90 (avg daily rate of the first 2160 hours = 90 days)
• includes ramp up time (i.e. pre-peak production)
• may not be indicative of elapsed time (e.g. soaking)
Peak Rate
• moderate to strong correlation to EUR (depends on the play)
Cum Prod in first N months (more months is better, but limits well count)
• normalized to first prod (dominated by ramp up time if too short a time frame)
• normalized to peak (caution: misses pre-peak production, it is better suited for
decline shape analysis)
Caution When Normalizing on Peak Date
How much pre-peak production could you be excluding?
When to Normalize on Peak (Type Curve Shapes)
Useful for comparing decline shapes… group by play,
technology, township etc.
What Production Measure Should I Use?
EUR
• new wells have insufficient data to forecast
• time consuming (if not automated)
IP90 (avg daily rate of the first 2160 hours = 90 days)
• includes ramp up time (i.e. pre-peak production)
• may not be indicative of elapsed time (e.g. soaking)
Peak Rate
• moderate to strong correlation to EUR (depends on the play)
Cum Prod in first N months (more months is better, but limits well count)
• normalized to first prod (dominated by ramp up time if too short a time frame)
• condense data (zero months excluded, makes analysis more consistent)
• normalized to peak (caution: misses pre-peak production, better for decline
shape analysis)
Correlations to EUR: More History is Better
EUR using Modified Duong to Arps. All wells in Montney Sample Area
have >80% correlation on both Duong fits. This analysis took 10
minutes to generate in VISAGE.
Correlation % Well Count
Peak Gas 60.0 585
IP90 Gas 49.2 579
6 Month Cum 49.3 585
12 Month Cum 67.1 523
18 Month Cum 75.3 473
24 Month Cum 79.7 377
30 Month Cum 83.4 227
36 Month Cum 87.5 227
Don’t rely on just one Production Measure
Different production measures can yield different results. Use multiple
perspectives in your analyses.
**Forecast uses Modified Duong to Arps. All wells in Montney Sample Area have >80% correlation on both Duong fits.
Peak EUR
Cumulative
Use Multiple Perspectives of Analysis (Dimensionally Normalized Variables)
Frac Analysis in VISAGE: Using Distributions as an Alternative to Linear Regressions
Frac Analysis in VISAGE: How to Refine Your Insights Using Distributions
What’s good in one
context, may not be
good in another.
Use Multiple Perspectives of Analysis(Dimensionally Normalized Variables)
Frac Analysis in VISAGE: Using Distributions as an Alternative to Linear Regressions
Frac Analysis in VISAGE: How to Refine Your Insights Using Distributions
What’s good in one
context, may not be
good in another.
Bubble Maps: Challenging to See Patterns
Use Quartiles or Other Colour Categories
Final Considerations…
Distributions require a significant sample size
Use multiple visual analysis techniques when possible (distributions, type curves, linear correlations … get to know what data
is available)
Use multiple production measures (Peak, Cumulative
Production, EUR … and apply multiple dimensional normalizations)
Validate your assumptions and conclusions with maps
and geological information when possible
Invest in data and analysis tools (when wells cost millions of
dollars to drill and complete …. What’s the value of a better decision?)
Presentation Recap
Demonstrated that completions are changing over time
with associated changes in production profiles
Reviewed techniques for quantifying and comparing
production impacts of completion designs
Considered what “production measure” to use
Discussed statistical analysis techniques using
distributions and maps
Data Sources used in VISAGE charts:
Information Hub
VISAGE Contact Information