Top-Down, Full Field Reservoir Simulation & Modeling (TDM ... · PDF file1 Top-Down, Full...

Post on 08-Mar-2018

226 views 4 download

transcript

1

Top-Down, Full Field Reservoir Simulation & Modeling (TDM) of Shale Assets09122-04Shahab D. MohagheghWest Virginia University

Summary of Results from Completed GTI Marcellus Shale R&D ProjectMay 7, 2013Meeting Location

rpsea.org

2

Contacts

PI: Shahab D. MohagheghWest Virginia UniversityShahab@wvu.edu304 293 3984

RPSEA PM: Kent Perrykperry@rpsea.org

3

Outline

o Conventional Modeling of Shale Assetso Top-Down Modeling of Shale Assets

• Definition• Technology• Workflow

o Results & Discussions• History Matching• Sensitivity Analysis• Forecasting Production

Existing Wells New Wells (pads)

o Conclusions

4

CONVENTIONAL MODELING OF

SHALE ASSETS

5

Conventional Modeling of Shale Assets

o Material Balance: ultra-low matrix permeability makes obtaining average reservoir pressure impractical (Strickland-Blasingame-2011)

o Volumetric: Matrix flow properties can vary more widely than it is practical to measure and ultimate recovery also depends heavily on how the rock responds to hydraulic fracturing (Cox-Gilbert-2002)

o Decline Curve Analysis : Hyperbolic Decline behavior commonly observed in the shale wells but using a traditional Arp’s decline curve methodology during either transient flow or the transitional period between the end of transient flow and onset of boundary dominated flow may cause a significant error for reserve evaluation (Rushing and et al-2007)

6

Conventional Modeling of Shale Assets

o Rate Transient Analysis: It can better represents the actual physics involved in reservoir dynamics but early applications face steep problems. It is a good tool for production analysis which gives some initial estimation of fracture properties for reservoir simulation (Nobakht, Mattar-2010)

7

Conventional Modeling of Shale Assets

o Numerical Simulation : So far, the most flexible tool for shale gas assessments ... It is undergoing further development to include better understandings of the basic physics controlling gas flow as the industry learns more (Lee, Sidle-2010)

Dual Porosity/Single Permeability Models: Industry Standard Dual Porosity/Dual Permeability Models: Important in some shale plays in which

unstimulated fracture permeability may be significant (Cipola 2010) Quad Porosity Models (Micro-Scale Modeling): Consideration of four porosity

systems as Organic, Inorganic, Natural Fractures and Hydraulic Fractures Porosity- (Civan et al-2010, 2012)

Reduced Physics Surrogate Model: It provides a single-porosity model without desorption or grid refinement but with tuned values for permeability and porosity in the stimulated zone. (Wilson- Durlofsky-2012)

8

TOP – DOWN MODELING OF

SHALE ASSETS

9

Top-Down Modeling - Characteristics

o The only technology currently available for full field reservoir simulation and modeling of shale assets.

o Makes no assumptions, approximations, or simplifications about the state of our understanding of the underlying physics (flow within and across the matrix, natural fracture, and induced fracture).

o Incorporates only “Hard Data”• Type and Amount of Fluid and Proppant, Clean and Slurry Volumes, Proppant

Concentration, Injection Rates, Injection & Breakdown Pressure, ISIP, Well Logs, Well Types, Offset Well Performance, Frac Hit, etc.

o Avoids using “Soft Data”• Fracture Half Length, Fracture Width, Fracture Height, Fracture Conductivity,

Stimulated Reservoir Volume (SRV), etc.

10

Top-Down Modeling - Definition

o A formalized and comprehensive, empirical, multi-variant, reservoir simulation model, developed solely based on field measurements (logs, core, well test, seismic, etc.) and historical production /injection data.

o Top-Down Modeling uses the pattern recognition capabilities of Artificial Intelligence & Data Mining in order to deduce physics from the available data, instead of imposing our current understanding of the physics on the modeling process.

11

Top-Down Modeling – Technology (Paradigm Shift)

o FIRST PARADIGM: Thousands of year ago:• Science was Empirical (describing the natural phenomena)

o SECOND PARADIGM: Last few hundred years :• Theoretical branch (using models, generalizations)

Kepler’s Law, Newton’s Law of Motion, Maxwell’s Equation

o THIRD PARADIGM: Last few decades :• Computational branch (simulating complex phenomena)

Theoretical models grew too complicated to solve analytically

Jim Gray: (1944-2007) Legendary American computer scientist , received the Turing Award for seminal contributions to computer science.

12

Top-Down Modeling – Technology (Paradigm Shift)

o FOURTH PARADIGM: Today:• Data-Intensive, Data Exploration, eScience• Unify theory, experiment, and simulation• Data captured by instruments or generated by simulator• Processed by Software• Information/Knowledge stored in computer• Scientists analyze database/files using data management tools

and statistics

Jim Gray: (1944-2007) Legendary American computer scientist , received the Turing Award for seminal contributions to computer science.

13

Artificial Intelligence & Data Mining

o Inspired by functionalities of human brain.o Human brain has remarkable pattern recognition capabilities.o Using Machine Learning, we build computer software that can learn

from experience.o This is called “Data-Driven Solutions”

14

Artificial Intelligence & Data Mining

15

Artificial Intelligence & Data Mining

16

Machine Learning in Action

17

Top-Down Modeling - Workflow

Model Validation

Model Validation

Prediction of Reservoir Performance

Forecasting and Sensitivity Analysis

Dynamic InformationDays of 

ProductionWellhead Pressures

Water Production

Rich Gas Production

Well Information

Static InformationReservoir

CharacteristicsCompletion 

DataStimulation 

DataGeomechanical 

Reservoir Model Base Model Flow Regimes Effect

Well Types Effect

Well/Pads Distances Effect

Offset Wells Effect

Final Model

History Matching 

Key Performance Indicator ‐ KPI

DATA

Integration into aSpatio‐Temporal Database

18

RESULTS & DISCUSSION

19

Results & Discussions

20

Results & Discussions

History matching result of the entire asset.

21

Well 10081 located in a two‐wells pad. Start of production: March 2009.

Good Results in History Matched Model

Results & Discussions

22

Good Results in History Matched Model

Results & Discussions

SPE 161184• Modeling and History Matching of Hydrocarbon Production from Marcellus Shale using Data Mining and Pattern Recognition Technologies •  Soodabeh Esmaili

100061000610062100621012510125

1009610096

23

Well 10072 located in a three‐wells pad

Start of production: August 2006.

Poor Results in History Matched Model

Results & Discussions

24

Poor Results in History Matched Model

Results & Discussions

1003210032

1002610026

10120101201003410034

SPE 161184• Modeling and History Matching of Hydrocarbon Production from Marcellus Shale using Data Mining and Pattern Recognition Technologies •  Soodabeh Esmaili

25

The error percentage of monthly gas production rate for all 135 wells was calculated using the following equation:

Results & Discussions

SPE 161184• Modeling and History Matching of Hydrocarbon Production from Marcellus Shale using Data Mining and Pattern Recognition Technologies •  Soodabeh Esmaili

26

Results & Discussions

o Results from the final history matched model:• Excellent Match (error < 10%):

101 wells (75% of wells)• Good Match (error < 20%):

22 wells (16% of wells)• Average Match (error < 30%):

6 wells (5% of wells)• Poor Match (error > 30%):

6 wells (4% of wells)

27

Results & Discussions

Input Parameters used in developing the Top-Down Model

28

Results & Discussions – Sensitivity Analysis

Degree of influence for each 8 

parameters for the production of a pad as a function of 

time.

29

Results & Discussions – Sensitivity Analysis

Type curves can be generated for individual wells, group of wells, and for entire field.

30

Results & Discussions - Forecast

Blind Ve

rification

Well 10105 located in a pad with 6 other laterals. This well came to production on November 2010, so it has 16 months history. The 20% of data for blind verification includes last 4 months.

Blin

d Ve

rific

atio

n

Well 10054 located in a two‐wells pad. This well came to production on February 2009, so it has 37 months history. The 20% of data for blind verification includes last 8 months.

31

Results & Discussions - Forecast

32

Results & Discussions - Forecast

33

Results & Discussions - Forecast

34

Results & Discussions - Forecast

35

Conclusions

o Conventional modeling of production from shale (pre-Shale technology), over-simplifies a complex problem.• It uses “Soft Data” and cannot be used as inverse problem to help design new

wells and completions.• It lacks predictive capabilities for new wells/pads in new locations• Computationally expensive• Impractical for full field analysis

36

Conclusions

o Top-Down Modeling may be used for full field modeling.o It has small computational foot print.o It demonstrates impressive predictive capabilities.o It can be set up as inverse problem to assist in identifying optimum

drilling locations, and to optimize frac design.o It uses only “Hard Data” and avoids “Soft Data” (interpretations)