Large-Scale Reservoir Simulation and Big Data Visualization Dr. Zhangxing John Chen NSERC/Alberta Innovates Energy Environment Solutions/Foundation CMG Chair Alberta Innovates Technology Future (icore) Chair Director, Foundation CMG/Frank-Sarah Meyer Collaboration Center, University of Calgary Slide 1
Outlines Motivations - Importance Simulation and Visualization Case Studies Summary Remarks Slide 2
Outlines Motivations - Importance Parallelization Means Case Studies Summary Remarks Slide 3
Motivations Accuracy Stability Robustness Speed capital savings Slide 4
Speed Issue 3D thermal models: with 500,000 cells geomodels and multiprocessor computers, runs take hours to weeks. For typical optimization run, 400-1,000 runs required. If done serially, hundreds to thousands of days for each optimization run required. With parallelization, optimization runtimes will drop. More geological descriptions can be added. Slide 5
Motivations Grid (scale) requirements Physics requirements Process requirements Slide 6
Grid Requirements Reservoir (rock and fluid) heterogeneity Thin moving thermal front Thin mobile solvent-rich layers ~ 1cm Discontinuity: presence of faults, fractures, and mud and shale layers Small dispersion/diffusion Local reaction (chemistry) zones Slide 7
Grid Requirements Rock heterogeneity Slide 8
Fluid heterogeneity Grid Requirements Slide 9
Grid Requirements Thin moving thermal and solvent fronts Slide 10
Grid Requirements Discontinuity: presence of faults, fractures, and mud and shale layers Slide 11
Grid Requirements Small dispersion/diffusion Slide 12
Grid Requirements Local reaction (chemistry) zones Slide 13
Physics Requirements Disparate data with different scales Thermal and solvent effects Mass and heat transfer Phase behavior Geomechanics Wellbore flow Slide 14
Process Requirements Thermal Recovery Processes - SAGD (steam assisted gravity drainage) - CSS (cyclic steam stimulation) - ISC (in situ combustion) Solvent Recovery Processes - VAPEX (vapor extraction) - SAP (solvent aided process) Slide 15
Business Competiveness Field management Equipment management Cost management SIMULATION TECHNOLOGY Strategy Safety Slide 16
Outlines Motivations - Importance Simulation and Visualization Case Studies Summary Remarks Slide 17
Petroleum Reservoir Simulators Black oil simulator (water flooding, fractured reservoirs) Compositional simulator (CO 2, N 2, CBM, tight and shale oil and gas) Thermal simulator (CSS, SAGD, ISC, THAI, Steam flooding) Chemical simulator (SAP+Foam) Wellbore module Geomechanical module Slide 18
Acceleration Means Software Acceleration Hardware Acceleration Slide 19
Software Acceleration Grid Management Discretization (Numerical) Methods System Solvers Parallelization (Matrices and Vectors) Software Design Slide 20
Grid Management The most critical modules - Type, choice of numerical methods - Partition, workload, communication - Data structure, info and data distribution - Numbering, bandwidth, and input Slide 21
Discretization Methods Finite difference methods Finite volume (control volume) methods Finite element methods Slide 22
System Solvers Preconditioners Linear Solvers A x = b Nonlinear Solvers Slide 23
Polynomial Approximate Inverse ILUT(p, tol) ILU(k) Preconditioners Domain Decomposition (Restricted Additive Schwarz) Algebraic Multigrid: classical and smoothed aggregation Slide 24
Linear Solvers GMRES, CG, BICGSTAB and GCR CGS, Orthomin and Orthodir Classical AMG Smoothed Aggregation AMG Slide 25
Nonlinear Solvers Newton Iterations Newton-Raphson Iterations Slide 26
Software Design OpenMP - Multi-core, easy to use, limited scalability MPI - communication - MPI-IO Programming languages - C, C++, Fortran, Slide 27
Hardware Acceleration CPUs (central processing units) GPUs (graph processing units) Slide 28
CPU VS GPU GPU (C2050) CPU (X5570) Cores 448 6 Memory 144 GB/s 10 GB/s Float Performance 1030G (s) / 515G (d) ~20G GPU is around 10 times faster than CPU! Slide 29
Outlines Motivations Simulation and Visualization Case Studies Summary Remarks Slide 30
AMG: Example 1 Two-dimensional elliptic problem Grid size: 1,000x1,000, non-zeros: 4,996,000 Tol: 1e-6, level: 8 36 iterations GPU: 1.30s CPU: 12.26s Speedup: 9.33 Slide 31
AMG: Example 2 Two-dimensional elliptic problem Grid size: 1,500x1,500, non-zeros: 11,244,000 Tol: 1e-6, level: 8 31 iterations GPU: 2.58s CPU: 24.50s Speedup: 9.42 Slide 32
AMG: Example 3 Three-dimensional elliptic problem Grid size: 100x100x100, non-zeros: 6,940,000 Tol: 1e-6, level: 8 21 iterations GPU: 1.13s CPU: 10.80s Speedup: 9.58 Slide 33
AMG: Example 4 Three-dimensional elliptic problem Grid size: 130x130x130, non-zeros: 15,277,000 Tol: 1e-6, level: 8 25 iterations GPU: 3.02s CPU: 37.86s Speedup: 12.47 Slide 34
Numerical Studies on CPUs Parallel (University of Calgary), Westgrid 528 standard nodes - 26-core Intel Xeon E5649 processors - 24 G memory InfiniBand 4X QDR, 40 Gbit/s Slide 35
15M Case: Example 5 GMRES + DDM Grid: 250x250x250 Unknowns: 15 million # processors 2 4 8 16 32 64 Grid time (s) 44.93 21.74 10.63 5.36 2.81 1.59 Overall time (s) 122.68 59.82 27.48 13.54 6.90 3.49 Slide 36
125M Case: Example 6 GMRES + DDM Grid: 500x500x500 Unknowns: 125 million # processors 16 32 64 128 Grid time (s) 49.19 24.18 11.84 5.45 Overall time (s) 1258.55 662.10 338.68 166.54 Slide 37
200M Case: Example 7 GMRES + DDM Grid: 585x585x585 Unknowns: 200 million # processors 16 32 64 128 Grid time (s) 80.54 38.93 19.82 9.52 Overall time (s) 2471.98 1286.08 670.83 346.07 Slide 38
Billion Case: Example 8 GMRES + DDM Grid size: billion (B) # processors 64 128 160 200 Grid size 1 B 1B 2 B 3 B Grid time (s) 115.76 57.56 94.74 123.62 Overall time (s) 4141.05 2029.93 3229.67 4060.92 Slide 39
Stereo 3D and Immersive Visualization Immersive visualization has been shown to enable interactions that are more intuitive and that let users focus on analysis. Slide 40
Stereo 3D and Immersive Visualization Slide 41
Stereo 3D and Immersive Visualization Slide 42
Case Study T.Q.C. Dang, L.X. Nghiem, Z. Chen, and T.B.N. Nguyen, CO 2 low-salinity water alternating gas: A promising new approach for EOR, Journal of Petroleum Technology, January 2015, 84-86 Slide 43
Case Study Geological Modeling Facies Modeling Clay Mapping Advanced Field Scale LSW Modeling Sensitivity Analysis History Matching Optimization Uncertainty Assessment Reservoir Simulation Flow, Ion Exchange, Geochemistry, Wettability Alteration CEC & Porosity Modeling Permeability Modeling Slide 44
Case Study Slide 45
Case Study Optimal Pattern Slide 46
Case Study Slide 47
Outlines Motivations Simulation and Visualization Case Studies Summary Remarks Slide 48
Summary Remarks Ultimate Goals Increasing reserves Reducing operating costs Enhancing petroleum recovery Capital savings Slide 49
Sponsors Slide 50