Innovation in Oil & Gas: Impacts of Digitalization on ... · AI frameworks to solve problems in specialized domain that typically requires human expertise AI solutions will draw upon

© Quantum Reservoir Impact, LLC | Confidential

(1/31/2018)

David Castiñeira (Chief Technology Officer at QRI)

Innovation in Oil & Gas: Impacts of Digitalization on Operations


1. Modeling Strategies for O&G fields

Simulation, Analytical and Data-driven Models

• Heuristics, Correlations and Analogues (PVT models, Recovery Factor,)

• Statistical & Signal Processing methods (Regressions, Wavelets, Laplace, …)

• Machine Learning and Neural-networks (Linear, recurrent, deep learning, …)

• Simple analytical models (Material balance, Decline Curves, Buckley-Leverett, ..)

• Simple numerical models (Capacitance-Resistance, Parametric, Streamlines, … )

• Full-physics reservoir simulation (Black-Oil, Compositional, Dual-Porosity, …)

Data-driven models vs Physics (?)

)(

75.0ln

2wf

w

eoo

roo PP

sr

rB

hKKQ

Example: Modeling oil production rate for pseudo-steady state flow circular reservoir

▪ K and h can be (partially) characterized by the location of a well

▪ Bo and o are function of the average reservoir pressure

o Average reservoir pressure is a strong function of the cumulative production, reservoir properties and drainage area

o The drainage area of a well is determined by reservoir characteristics, cumulative production and number of wells in the field.

▪ Kro is a function of the oil saturation, which is strongly related to the cumulative production.

▪ Pwf and skin factors are the most likely time-dependent and are usually hard to quantify

• Let data speak too!• Build from physics…• .. but don’t restrict

to textbook models and assumptions

• Data-driven models can capture underlying physics under right framework.

• AI + engineering context can provide optimum solution

Darcy’s law, integration in cylindrical coordinates)

Example: Determining OWC

Generalized Material Balance Equation

Determine Reservoir Drive Mechanisms & )(tSavg

Creating WOC & GOC Matching and Including , Historical BT & Coning)(tSavg

Example: Reduce the Risks of Excessive Water Production in Infill Drilling Campaign

Frac Oriented

Simple Cylindrical(SR=1000m)

Simple Cylindrical(SR=500m)

A’AOpen PerforationClosed Perforation

• Data-driven models must be combined with fundamental, engineering understanding of reservoir behavior (!)

Considerations to Select Right Modeling Strategies

i. Do we really know our reservoir?

ii. Do we have data?

iii. What is the time frame to solve the problem?

iv. Context: reservoir management?


2. Automation Opportunities in RM

1. Clearly understanding the goals of the problem we want to model

2. Data pre-processing (data gathering, exploration/visualization, transformation/reduction)

3. Determine the machine learning task (i.e., translate step 1 into a more specific statistical question).

4. Apply machine learning algorithm (e.g., ANN, Random Forests, SVM…)

5. Interpret results of the machine learning algorithm

6. Deploy the model (integrate model into operational system).

A Systematic Approach to Machine Learning Modeling

© Quantum Reservoir Impact, LLC | Confidential Slide 9

I/O view of the modeling problem

Run Model/sInput Output

This is easy to automate!


Modeling Execution Process

Data Collection Data Preprocessing

Run Models

Expert Vetting

Client/Asset team workshop

Value Creation

Playbook

Reservoir Management• Recovery Design (e.g., D&C

design)• KROs• Pressure Maintenance• Depletion• Reserves• Surveillance• Workovers• Economics• ….

Can we automate the whole thing?

Organizational Capabilities for Automation

i. Good problem framing (Mgmt&Engineering&Quants)

ii. Allow lateral thinking when it comes to automation

iii. Agile/Lean Development

iv. Emphasize knowledge mgmt


3. Machine Learning and AI applications

Application 1: Eagle Ford

-100.5 -100 -99.5 -99 -98.5 -98 -97.5 -97 -96.527.5

28

28.5

29

29.5

30

30.5

All well locations Eagle Ford

• Client owns land in Eagle Ford; first wells show disappointing results.

• Client is considering 3 options:I. Change operatorII. Sell entire positionIII. Be patient and wait for technology to improve

• Want quantitative answers

0 0.5 1 1.5 2 2.5 3 3.5 4

x 105

0

0.5

1

1.5

2

2.5

3

3.5

4

x 105 VALIDATION SET: 12-Month Cum BOE (BOE) (field data vs. predictions) ; Certainty = 83.3 %

True values

Pre

dic

ted v

alu

es

Eagle Ford

Treasure Map

Application 2: FDOs in non-economical Mexico field

• Objective: Opportunity identification in very large tight-oil field

• Timeframe:

▪ Preliminary data collection

▪ 3 weeks of focused work

▪ 1 week of meetings, reviews and workshops

• Approach:

▪ Top-down workflow focused on value creation

▪ Speed provided by fast Quantitative Analysis

▪ Guidance provided by experienced engineers & geoscientists

▪ Analysis accelerated by proprietary technologies

▪ Diversified modeling approaches

▪ Strong Knowledge Management foundation

▪ Thousands of opportunities identified using QRI AI and Machine Learning Algorithms and Workflows.

Application 3: Well Target Identification▪ Problem: To identify new well targets using AI in complex fractured carbonate

1994 1999 2004 2009 20140

1000

2000

3000

4000

5000

Date

Rate

[M

ST

B/D

] or

[MS

CF

/D]

Production Rates (CN-0010A)

Oil

Gas

Water

1974 1979 1984 1989 1994 1999 2004 2009 20140

2

4

6

8

10x 10

6

Cum Oil (STB)

Date

Cum Oil (CN-0010A)

4.85 4.86 4.87 4.88 4.89 4.9 4.91 4.92 4.93 4.94

x 105

1.992

1.993

1.994

1.995

1.996

1.997

1.998

1.999

2

x 106 CN-0010A

CN-0010ACN-0010

CN-0020ACN-0020B CN-0021

CN-1011

CN-1021

CN-0001

CN-0011 CN-0012 CN-0012A

CN-0020 CN-0022 CN-0023

CN-0030 CN-0030ACN030AR1 CN-0033

CN-0221

CN-1025

CN-1027

SA-0155

SA-0163 SA-0165 SA-0167 SA-0169

CN-0013 CN-0014

CN-0024

CN-0025 CN-0029

CN0029R1

CN-0032 CN-0035

CN-0040 CN-0041 CN0041RECN-0043 CN-0045

CN-0301 CN-0321

CN-0322

CN-0532

CN-1013

CN-1023

CN-1113

CN-5013

CN-5023 CN5023RE

IR-0168

IR-1168

SA-0143A SA-0145

SA-0151 SA-0151ASA-0153

SA-0157A SA-0159 SA-0159A

SA-0161

CN-0141

Ring 2

Ring 3

Ring 4

Ring 1CN-0010A

OFFSET WELL DATA

PRODUCTION DATA

0 2000 4000 60000

500

1000

1500

2000

2500Rs (SCF/STB) vs. P (psi)

0 2000 4000 60001

1.2

1.4

1.6

1.8

2

2.2

2.4

2.6Bo (RB/STB) vs. P (psi)

0 2000 4000 60000

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

x 10-3Bg (RB/SCF) vs. P (psi)

PVT DATA

PRESSURE DATAGEOLOGICAL DATA

1979 1984 1989 1994 1999 2004 2009 20140

1000

2000

3000

4000

5000

6000

7000

Date

Pres

sure

s [P

SI]

Field Pressures

SBHP

SBHP (smooth)

X=489610.9; Y= 1995589; Z=3784.7; Length = 12m

BASIC WELL DATA

KS

Potential good locations for new

producers derived from AI

Application 3: Well Target Identification▪ Problem: To identify new well targets using AI in complex fractured carbonate

0.5 1 1.5 2 2.5 3

x 106

0.5

1

1.5

2

2.5

3

x 106

True values

Pre

dic

ted v

alu

es

VALIDATION SET: True values vs predictions

CN-0010ACN-0015

CN-0059

CN-0063

CN-1013

CN-1023

CN-5013

IR-0128A

IR-0156A

IR-0164

IR1126R1

IR-1128

IR-1158

IR-2144

IR-5166

OX-0002

OX-0004

PL-0221

SA-0107

SA1187R1

SA1196R1

SA-1197

SA-1199

0

20

40

60

80

100

120

PL-0

221

IR-0

164

IR-0

156A

IR-2

144

CN

-1023

CN

-5013

IR1126R

1

IR-1

128

CN

-1013

SA

-1199

SA

1196R

1

CN

-0015

CN

-0010A

SA

1187R

1

CN

-0063

SA

-0107

IR-0

128A

CN

-0059

IR-5

166

OX

-0002

IR-1

158

SA

-1197

OX

-0004

PL-0

221

IR-0

164

IR-0

156A

IR-2

144

CN

-1023

CN

-5013

IR1126R

1

IR-1

128

CN

-1013

SA

-1199

SA

1196R

1

CN

-0015

CN

-0010A

SA

1187R

1

CN

-0063

SA

-0107

IR-0

128A

CN

-0059

IR-5

166

OX

-0002

IR-1

158

SA

-1197

OX

-0004

VALIDATION SET: Absolute percentage errors in predictions (sorted in descending order);

Absolu

te p

erc

enta

ge e

rror

(%)

Mean Error = 34.2437%

Std Error = 35.0927%

Mean Error (97% per)= 30.173%

Std Error (97% per)= 29.8473%

MAPE validation

set = 34 %

MAPE for 97%

percentile = 30%

MAPE analysis (MAPE = Mean

Absolute Percentage Error):

2-year Cum Oil Prediction

True value

Pre

dic

ted

val

ue


Application 4: Drilling

SDS - SpeedWise Drilling Solutions

▪ Problem: Analyzing drilling performance (and opportunities) for field with > 200 wells

SDS - SpeedWise Drilling Solutions

Optimize rig scheduling and resource allocation for different scenarios

Automatically extract hole size and depth for the daily description

Crucial drilling and reservoir management metrics & automated insights

• Overview of for each field/asset• Well book for each well• Online interactive visualization

Predict productive/non-productive time

• Major activities during D&C: Drilling (76.4%), Casing (7.8%), Completion (6.9%)

• Major NPT causes: LostCirculation (10.8%), Logistics (10.4%), WellControl (10.1%)

• Highest completion captital was spent in 2015: 46.4 MMUSD

• Highest completion capital was allocated to FA field during the past 7 years: 107 MMUSD

• Average completion cost per 12 month cum. increased from 2.4to 9.4 usd/bbl in the past 6 years

• Lowest recent completion cost per 12 month cum.: field FA

• Advanced technology to improve drilling efficiency.• Rapid and intelligent analysis of drilling data.• SDS automates metrics (NPT, DEI, ROP) and analytics• Value creation (NPV) by optimizing rig schedule, drilling practices, etc


▪ AI frameworks to solve problems in specialized domain that typically requires human expertise ▪ AI solutions will draw upon the our worldwide knowledge of reservoirs + technology applications.▪ Knowledge Base Systems, Expert Systems, Machine Learning and Heuristics to generate:

▪ Insights ▪ Business optimization▪ Process automation

Reservoir Analytics

Knowledge Base System

Global Reservoir Mgmt Experience Machine Learning Workflows

+ +

Load Inputs

Neural Network Architecture Samples

Call

ANN Performance testing or evolving

Return blind testing error

ANN performance objective function:

E=f(ANN architecture)

• Genetic algorithm• Monte Carlo experiments

Neural Network Optimization Model

Generate report and save required

parameters

The input parameters and required output parameters for report generation are saved as .mat data structure

• Input & output data samples

• ANN training setup parameters

• ANN optimization parameters

Ultimate goal: Augmented AI

Documents

Innovation in Oil & Gas: Impacts of Digitalization on ... · AI frameworks to solve problems in specialized domain that typically requires human expertise AI solutions will draw upon