Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Feasibility of moment tensor inversion for a single-well microseismic data using neural networks

Slides for my talk at GEO 2018 in Manama, Bahrain. We approach the problem of full moment tensor reconstruction when given data from a single well. Only 5 of 6 moment tensor components are resolved in isotropic medium (in anisotropic all 6 might be resolved in theory), whereas the 6th one is only approximated. We propose a data-driven approach to build a representation of all 6 components from amplitudes of first arrivals at 3-component geophone using a vanilla feed-forward multilayer perceptron.

  • Login to see the comments

  • Be the first to like this

Feasibility of moment tensor inversion for a single-well microseismic data using neural networks

  1. 1. O. Ovcharenko, J. Akram, D. Peter March 6, 2018 Feasibility of moment tensor inversion for a single-well microseismic data using neural networks
  2. 2. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Outline Single-well data Artificial Neural Network Example in a homogeneous model
  3. 3. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Idea Reconstruct source mechanisms from single-well data using an artificial neural network https://www.geoexpro.com/articles/2017/07/
  4. 4. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANNhttp://www.halliburton.com/en-US/ps/pinnacle/microseismic-monitoring/
  5. 5. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Source mechanism P S “Beach ball” http://demonstrations.wolfram.com/author.html?author=Frank+Scherbaum%2C+Nicolas+Kuehn%2C+and+Bj%C3%B6rn+Zimmermann Radiation patterns
  6. 6. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Source mechanism P S 6 independent components Seismic moment tensor (Stein & Wysession, 2003) Radiation patterns “Beach ball”
  7. 7. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Source mechanism P S 6 independent components Seismic moment tensor (Stein & Wysession, 2003) Radiation patterns “Beach ball”
  8. 8. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Source mechanism P S 6 independent components Seismic moment tensor (Stein & Wysession, 2003) Radiation patterns “Beach ball”
  9. 9. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN (Rebetsky et al., 2018 in preparation) Stress state Seismic hazard assessment Estimates of fracturing direction Better reservoir characterization Used for
  10. 10. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Microseismic acquisition design SurfaceBorehole (Leo Eisner, MicroSeismic Inc.) (Geometrics Inc.)
  11. 11. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Multi-well borehole acquisition (Vavryčuk, 2007)
  12. 12. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Single-well borehole acquisition (Vavryčuk, 2007) Up to 5 moment tensor components could be resolved with conventional P+S arrival inversions
  13. 13. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Can we retrieve the full moment tensor from a single-well acquisition using an artificial neural network?
  14. 14. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Single-well data Artificial Neural Network Example in a homogeneous model
  15. 15. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Machine Learning
  16. 16. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Neural Networks pros and cons - Lots of parameters - Hard to interpret - Computational costs + Good for nonlinear problems + Good for large inputs + Data-driven + Easy to implement and parallelize + Ability to generalize Not a magic wand, use with care Pros Cons
  17. 17. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Neural network architecture Input Hidden OutputLayers: Neuron Bias Weight Training = tuning up weights w b w + = W x b
  18. 18. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Training Training inputs Training outputs Minimize L2 norm Neural networks require training with multiple inputs
  19. 19. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Data generation https://www.geoexpro.com/articles/2017/07/
  20. 20. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Input data Px Py Pz Sx Sy Sz Px Py Pz Sx Sy Sz First receiver Last receiver 3-component receivers 2 types of waves (P and S) 3 x 2 x NREC Maximum amplitudes of P and S waveforms multiplied by their first arrival polarity +Pz +Sz
  21. 21. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Output data Mxx Myy Mzz Myz Mxz Mxy 6 components of moment tensor describing source mechanism
  22. 22. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Single pair of training data INPUT OUTPUT 6 components of moment tensor P S INPUT OUTPUT
  23. 23. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Single-well data Artificial Neural Network Example in a homogeneous model
  24. 24. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Homogeneous medium N = 8000 events M = -3 … 0 Synthetic data Mostly shear mechanisms
  25. 25. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Before After INPUTS OUTPUTS Features Features Data normalization Train Test Important stage
  26. 26. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Computational facts 8000 300 6 NN: 5 hidden layers Batch size: 200 Learning rate: 0.005 Optimizer: Adam Weight regularization: 0.005 ObspyTensorFlow 1.3.0Python 3.6 Keras 2.0.5Matlab R2016b Initialization “xavier” (Glorot & Bengio, 2010) Training time ~ 5 min Prediction time < 1 sec Training data generation ~ 2 min on 24 cores
  27. 27. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Detected focal mechanisms Neural network never seen these data before
  28. 28. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Detected focal mechanisms
  29. 29. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Analysis Test data 1600 events 1294 DC 306 DC + ISO 17% moment tensors matched with 10% tolerance 80% eigenvalues matched with 5% tolerance Std 12O Deviation between detected and true P and T axes
  30. 30. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Conclusions • Used a neural network to extract radiation pattern from a single-well data • Able to retrieve full moment tensor • Difficult to match absolute values • Promising match of eigenvalues and eigenvectors Future work • Quantitative comparison • Seismic moment • Realistic cases
  31. 31. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Acknowledgements We are grateful to Leo Eisner, SMI and SWAG groups at KAUST for fruitful discussions.
  32. 32. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Conclusions • Used a neural network to extract radiation pattern from a single-well data • Able to retrieve full moment tensor • Difficult to match absolute values • Promising match of eigenvalues and eigenvectors Future work • Quantitative comparison • Seismic moment • Realistic cases
  33. 33. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN The End Automated fault detection (Araya-Polo et al., 2017) Salt body picking (Guillen et al., 2017) Mapping reservoirs on migrated seismic (Bougher, 2016) Facies classification and reservoir properties prediction (Hall, 2017; Ahmed et al., 2010) Event detection (Akram et al., 2017) Interpolation of missing data (Jia and Ma, 2017) Denoising (Zhang et al, 2017) Some NN applications in geophysics Inversion of seismic, DC data etc. (Röth, Tarantola, 1994; Neyamadpour et al., 2009)
  34. 34. oleg.ovcharenko@kaust.edu.saFeasibility of moment tensor inversion using ANN Bonus

×