Fast variogram modeling driven by artificial intelligence
A method for variogram modeling is disclosed. The method includes obtaining a synthetic well data and a well data for facies or petrophysical properties of interest in a targeted reservoir zone, training machine learning models using the synthetic well data as inputs and outputting a plurality of final variogram parameters predicted by using a plurality of machine learning models for the facies or petrophysical properties of interest in the targeted reservoir zone, wherein the well data for the facies or petrophysical properties of interest in the targeted reservoir zone is used as input.
1. A method for variogram modeling, the method comprising:
obtaining, by a computer processor, a synthetic well data and a well data for subsurface rock properties of interest in a targeted reservoir zone;
training, by the computer processor, a first machine learning model using the synthetic well data as inputs;
outputting, by the computer processor, a plurality of variogram parameters predicted by using the pre-trained first model for the subsurface rock properties of interest in the targeted reservoir zone, wherein the well data for the subsurface rock properties of interest in the targeted reservoir zone is used as input;
training, by the computer processor, a second machine learning model using the variogram parameters derived from the first machine learning model as starting parameters, wherein the well data for the subsurface rock properties of interest in the targeted reservoir zone is used as input;
outputting, by the computer processor, a plurality of variogram parameters predicted by using the trained second model for the subsurface rock properties of interest in the targeted reservoir zone, wherein the well data for the subsurface rock properties of interest in the targeted reservoir zone is used as input;
training, by the computer processor, a third machine learning model using the variogram parameters derived from the second machine learning model as starting parameters, wherein the well data for the subsurface rock properties of interest in the targeted reservoir zone is used as input; and
outputting, by the computer processor, a plurality of final variogram parameters predicted by using the trained third model for the subsurface rock properties of interest in the targeted reservoir zone, wherein the well data for the subsurface rock properties of interest in the targeted reservoir zone is used as input.
2. The method of claim 1 , wherein the synthetic well data represents spatial and vertical subsurface rock properties and petrophysical properties distribution along with a labeled data of the best-fit variogram parameters.
3. The method of claim 1 , wherein the synthetic well data includes a plurality of varying zone anisotropy, geometric anisotropy, cyclicity and other forms represented in nested variogram driven by secondary or tertiary geological settings and depositional systems.
4. The method of claim 1 , wherein the first machine learning algorithm may be selected from a group consisting of a convolutional neural network, a deep neural networks, and a recurrent neural networks.
5. The method of claim 1 , wherein the second machine learning algorithm may be selected from a group consisting of gradient descent, a simulated annealing, a random forest, a support vector machine, and a multi-layer perceptron.
6. The method of claim 1 , wherein the third machine learning algorithm may be a genetic algorithm.
7. The method of claim 1 , wherein the three machine learning algorithms are combined sequentially to achieve a fast model convergence.
8. The method of claim 1 , wherein a fitness function is computed for each parameter, the fitness function determining a probability that the parameter will be selected for a next generation.
9. The method of claim 1 , wherein the variogram parameters may be selected from a group consisting of a nugget, a model type, a sill, an azimuth, a range 1 for major direction, a range 2 for minor direction, and an anisotropy.
10. The method of claim 1 , wherein the final variogram parameters may be applied with a stochastic interpolation algorithm to build a model of petrophysical properties in earth modeling.
11. A system for variogram modeling, comprising:
a processor comprising a computer processing unit capable of executing instructions stored in a non-transitory computer storage medium which causes the processor to perform functions comprising:
obtaining a synthetic well data and a well data for subsurface rock properties of interest in a targeted reservoir zone;
training a first machine learning model using the synthetic well data as inputs;
outputting a plurality of variogram parameters predicted by using the pre-trained first model for the subsurface rock properties of interest in the targeted reservoir zone, wherein the well data for the subsurface rock properties of interest in the targeted reservoir zone is used as input;
training a second machine learning model using the variogram parameters derived from the first machine learning model as starting parameters, wherein the well data for the subsurface rock properties of interest in the targeted reservoir zone is used as input;
outputting a plurality of variogram parameters predicted by using the trained second model for the subsurface rock properties of interest in the targeted reservoir zone, wherein the well data for the subsurface rock properties of interest in the targeted reservoir zone is used as input;
training a third machine learning model using the variogram parameters derived from the second machine learning model as starting parameters, wherein the well data for the subsurface rock properties of interest in the targeted reservoir zone is used as input; and
outputting a plurality of final variogram parameters predicted by using the trained third model for the subsurface rock properties of interest in the targeted reservoir zone, wherein the well data for the subsurface rock properties of interest in the targeted reservoir zone is used as input.
12. The system of claim 11 , wherein the synthetic well data represents spatial and vertical subsurface rock properties and petrophysical properties distribution along with a labeled data of the best-fit variogram parameters.
13. The system of claim 11 , wherein the synthetic well data includes a plurality of varying zone anisotropy, geometric anisotropy, cyclicity and other forms represented in nested variogram driven by secondary or tertiary geological settings and depositional systems.
14. The system of claim 11 , wherein the first machine learning algorithm may be selected from a group consisting of a convolutional neural network, deep neural networks, recurrent neural networks.
15. The system of claim 11 , wherein the second machine learning algorithm may be selected from a group consisting of a gradient descent, a simulated annealing, a random forest, a support vector machine, and a multi-layer perceptron.
16. The system of claim 11 , wherein the third machine learning algorithm may be a genetic algorithm.
17. The system of claim 11 , wherein the three machine learning algorithms are combined sequentially to achieve a fast model convergence.
18. The system of claim 11 , wherein a fitness function is computed for each parameter, the fitness function determining a probability that the parameter will be selected for a next generation.
19. The system of claim 11 , wherein the variogram parameters may be selected from a group consisting of a nugget, a model type, a sill, an azimuth, a range 1 for major direction, a range 2 for minor direction, and an anisotropy.
20. The system of claim 11 , wherein the final variogram parameters may be applied with a stochastic interpolation algorithm to build a model of petrophysical properties in earth modeling.