ANADARKO KO PETROLEUM CORPORATION KINETICA CA INC FROM M DEEP LE P LEAR ARNI NING NG TO NE NEXT-GEN N VI VISU SUAL ALIZA IZATION ION: : A G A GPU PU-PO POWE WERED RED DIG IGIT ITAL AL TRAN ANSF SFORMAT RMATION ION Ingrid Tobar Amit Vij Senior nior Data Scien entist tist Preside sident nt & C Co-Founder nder Anadar An darko Kinetica tica San Jose, California March 18, 2019
OB OBJE JECTIVE: CTIVE: AT ATTE TEMPT MPT TH THE IMP E IMPOS OSSIBLE SIBLE Visualize and interact with a very high fidelity 3D representation of the Del elaware ware Ba Basin in for hydrocarbon exploration
9M to 90B 9M 90B Point Reservoir Model The e Delawar aware e Basin sin is s ro roughl ghly th the e si size of Mass ssachu achuse setts tts … 120 mi 80 0 mi mi 4,000 00 ft and d 3x th the e heigh ght t of th the Empi pire re Sta tate te Buildi ding ng 3
Age genda nda • About Anadarko • GPU-Enabled Tech and Projects • About Kinetica • GPU-Accelerated Visualization
About out Anada adarko rko 5
Cor orporate orate St Stra rategy tegy LEGEN END FOCUS US AREA EAS ANADARKO STRATEGIC FOCUS AREAS DJ Basin, Delaware Basin, DWGOM D J B ASIN CASH H GENERA NERATION ION DWGOM & International Oil D ELAWARE B ASIN FUTU TURE RE VALUE UE D EEPWATER GOM Exploration & LNG U.S. ONSHORE ALGERI RIA TOTAL L VOLUME UME GULF OF MEXICO 700 700 COLOMBIA BIA GHANA MBOE/D MOZAMBIQUE CAPEX EX $4.8 .8B ENHANCING SUCCESS EXPANDING LOWER 48 ENABLING DIGITA TAL IN DWGOM FOOTPRINT OPERATIONS Advanced geophysical analytics to enable High density Lower 48 subsurface “Intelligent” control and edge computing in exploration with tiebacks to existing infrastructure characterization to provide optionality Drilling, Completions and Production 6 About t Anadark rko GPU-Enabled Tech and Projects About Kinetica GPU-Accelerated Visualization Close / Q&A
AA AAET: T: A Adva vanced ced An Analy lytic tics s and Emer ergin ging g Te Tech chnology ology TEAM DEMOG OGRAPHI APHIC # TEAM MEMBE BERS PROJ OJECT CT PORTFOL OLIO PLATFOR ORM DEPLOYME YMENT NT 5 5 Data ta 50 50 30 30 25+ 25+ Scien enti tist sts 4 Dom omai ain 40 40 Pla latf tfor orms ms Proj rojects Geosc scientis tists ts rms # Platforms 50 50 3 20 20 30 30 rojects ts Person son # Pro 2 20 20 Engine ineer ers Core re Team 10 10 1 10 10 Data taOp Ops s 0 0 0 & Dev evOp Ops 2016 2019 2016 2019 2016 2019 INCEPTION OF DATA PRODUCTIZATION STAKEHOLDER ENGAGEMENT DEPLOYMENT AT SCALE SCIENCE SKILLS IN APC STRATEGY DEVELOPED FOCUS THROUGH PLATFORMS 2016 2017 2018 18 2019 Dr. Sean Gourley Strategic Alliance Kinetica Visualization APC Announces NVIDIA GPU appointed to Board Team Formed with RE Energy Group Project Kick-Off Google Partnership Technology Conference 7 About t Anadark rko GPU-Enabled Tech and Projects About Kinetica GPU-Accelerated Visualization Close / Q&A
Op Oper erationa ationalizing lizing Di Digi gital tal Exploration oration Development lopment Opera rations tions Identifying swe weet et spots ts where well Selecting the optimal mal we well design gn – Monitoring and understanding performance is high and land which involves choices in numerous asset et behavior or through the life- entry costs are low can generate areas such as completion size and well cycle of well construction (drilling) significant value to the company spacing – requires predicting the to extraction of underground performance for each candidate design resources (production) ENHANCING SUCCESS IN DWGOM EXPANDING LOWER 48 FOOTPRINT ENABLING DIGITAL OPERATIONS Seism smic c Str Strati tigra graphi phic c Real-Time Time Inter terpret retati tion on Top Corr rrelati tion on Dri rilling ling 8 About t Anadark rko GPU-Enabled Tech and Projects About Kinetica GPU-Accelerated Visualization Close / Q&A
GP GPU-Ena Enabled bled Te Tech ch and nd Proj rojects ects
Seis Se ismic ic Interpretation terpretation • Proje ject ct Sco cope • Envi vironment ment Seismic ismic int nterpre erpretation ation deep ne neur ural al ne network work 1.5 yr. ago: DGX GX-1 1 8x Tesla la P-100 100 GP GPUs model el for image ge process essing ing Today: y: □ DGX-1 8x Tesla V-100 GPUs & • Data Volu lume □ DGX-2 16x Tesla V-100 GPUs 100s GB GB – several eral TB 1000s images/attr ages/attribu ibutes tes Training ining on 1% data Inf nferen rence e across oss full l image ge • Fr Framew ewor ork TensorF sorFlow/Py low/PyTor Torch 2 c concurr rren ent fault lt predicti iction on models els 10 About Anadarko GPU-Enabled Tech and Projects cts About Kinetica GPU-Accelerated Visualization Close / Q&A
Se Seis ismic ic Interpretation terpretation 11 About Anadarko GPU-Enabled Tech and Projects cts About Kinetica GPU-Accelerated Visualization Close / Q&A
Se Seis ismic ic Interpretation terpretation • Benefits efits Training ining and nd inf nfere erenc nce 1.5 yr. ago: ~20 20 hours Today: <10 10 hours • Challenge llenges Time me intensiv ensive e training aining process ess Loadi ding g data a into o GPU U memo mory ry • Next xt Steps Netwo work rk enhancements ncements Workf kflow low impro provemen vements ts New DGX-2 2 box: Currently, the models only use 1% of the data □ 16 16x x V-100 00 GPUs Us + 512 GB GPU U Memor ory for training, however, inference is performed on the entire image. Future ure Environm ironment: ent: This means that we need to dedicate □ Google Cloud Platform (GCP) significant amounts of time to training in order to deliver good inferences. 12 About Anadarko GPU-Enabled Tech and Projects cts About Kinetica GPU-Accelerated Visualization Close / Q&A
St Stra ratigraph tigraphic ic To Top Cor orrelation relation • Proje ject ct Scope Seed Seed Seed Seed Well Well Well Well Learn rn from m identified ntified top tops propa pagate gate at basin in scale le • Data Volu lume 000 ft 18,000 Trainin ining g ~25GB GB Inferen rence: e: Size ze varies ies, on the fly • Fr Framew ewor ork CNN in TensorF sorFlow low • Envi vironmen ment De Dev/Train /Train (on n prem.) m.): □ DGX-1 8x Tesla V-100 GPUs Inferen rence/U e/UI I (on cloud) d): □ GCP V-100 and T4 GPUs 13 About Anadarko GPU-Enabled Tech and Projects cts About Kinetica GPU-Accelerated Visualization Close / Q&A
St Stra ratigraph tigraphic ic To Top Cor orrelation relation • Massiv ive e data vo volu lumes mes • Faster er train ining ing wit ith new w GPU chip ips Rap apid idly gr growin ing g ge geo da data GPU PU Qua uadr dro o P6 P6000: 000: co coup uple e weeks Pic icked ed and d in infer eren ence ce wel ells DGX DGX-1 1 8x Tes esla P-100 00: : 1.5 5 – 2 days Expe pert t pi pickin ing g (label elin ing) DGX DGX-1 1 8x Tes esla V-100 00: : < < 24 24 hours Benefits • Acc ccele elerat ated ed basin in eval valuation tion proce cess ss Next Steps • Short term Challenges Bet etter er net etworks ks as CNN runs ‘Self -tuni tuning ’ mechanism • Tim ime e in intensiv sive e train ining ing pr proce cess ss • Lo Long term CNN train inin ing Move e workfl rkflow w to cloud oud GPU tec ech h New T4 GPU for r in infer eren ence ce ad advan ances es TGS, Wood Mackenzie PetroView □ In n GC GCP since nce Jan n 2019 14 About Anadarko GPU-Enabled Tech and Projects cts About Kinetica GPU-Accelerated Visualization Close / Q&A
Real-Time Re Time Dr Dril illing ling • Proje ject ct Sco cope Real-Time Drilling ROP Footage Time Dr Drilling illing Ops: : $M d decisions isions Operati rationa nal Analytics lytics and DL models ls process ess real-time time Footage % On Bottom Time Off Bottom Time KPIs s at high streaming reaming log data a & & other er non-stre streaming aming data resolut ution on Rig states → De Deriv ive e opera rational tional KPIs of of Footage, Time, ROP, Connection Time Connection Time drill llin ing g ops at very high resolution solution connection statistics Real-Time Drilling Day Crew Night Crew Average Connection Depth Target Line Landing Projection Real-Time Drilling 3D Plot – Plan vs. Actual Current Deviation Side View Directi ction onal al Guidan ance ce Plan vs actual charts, directional efficiency and guidance, projections 3D Traject ector ory Plan vs actual and offset trajectories Plan Actual Upper Window Lower Window Projection Actual Plan Offset Well 15 About Anadarko GPU-Enabled Tech and Projects cts About Kinetica GPU-Accelerated Visualization Close / Q&A
Recommend
More recommend
