Seismic Inversion Using CMA-ES for Subsoil Elastic Parameters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current seismic inversion methods face challenges in accurately estimating elastic parameters of subsoil regions due to the presence of noise in seismic data, leading to convergence towards local minima of the cost function, and require assumptions about statistical distributions that do not reflect real data, resulting in inaccurate models.
Innovation Solution
The method employs a global minimization algorithm using the covariance matrix adaptation - evolution strategy (CMA-ES) to select elastic parameter values that minimize the cost function without assuming a specific statistical distribution, allowing for a more accurate representation of complex subsoil structures by probing a large set of candidate values and adapting uncertainty levels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional seismic inversion methods are used with statistical distribution assumptions, then the inversion process is simplified and converges faster, but the accuracy of elastic parameter estimation deteriorates due to convergence towards local minima and biased results
Solution Approach 1:
The subsoil region is divided into multiple columns of cells, with seismic inversion performed independently for each column. This segmentation allows the CMA-ES algorithm to be applied column-by-column, reducing the overall complexity while maintaining global optimization capabilities for each segment without requiring unrealistic statistical distribution assumptions
Solution Approach 2:
The patent changes the optimization approach from conventional gradient-based methods to CMA-ES (Covariance Matrix Adaptation Evolution Strategy), which adapts the covariance matrix of the search distribution based on successful iterations. This parameter adaptation enables the algorithm to efficiently explore the search space and converge to the global minimum without being trapped by local minima, thereby improving accuracy while maintaining reasonable computational efficiency
2Measurement precision
If global minimization algorithms like CMA-ES are used without statistical distribution assumptions, then the accuracy of elastic parameter estimation is improved, but the computational complexity and processing time increase
Solution Approach 1:
By dividing the subsoil region into multiple columns and performing inversion independently for each column, the patent reduces the dimensionality of the optimization problem. The CMA-ES algorithm then operates on smaller subsets of parameters simultaneously, which reduces computational complexity compared to applying global optimization to the entire subsoil model at once, while still achieving accurate results through unbiased global minimization
Solution Approach 2:
The patent applies CMA-ES with adaptive covariance matrix updates, which performs partial optimization actions in each iteration by adapting only the necessary parameters based on successful search steps. This partial adaptation approach balances computational effort with optimization effectiveness, avoiding the need to compute and update the full covariance matrix for all parameters at every iteration, thus reducing overall computational complexity
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
A subsoil region is represented using a grid having columns of cells arranged in an array (2) of horizontal positions. The cells of a column are delineated by geological layering of the subsoil region. The method, used for estimating elastic parameters in the subsoil region, comprises obtaining seismic traces and performing seismic inversion for the columns of cells. The seismic inversion for a column comprises selecting a set of elastic parameter values that minimizes a cost function regardless of statistical distribution of the elastic parameter values. The set of elastic parameter values is selected among a plurality of candidate sets of elastic parameter values each having values of elastic parameters for the cells of the column.