Trans-Dimensional Geophysical Data Mining for Noise Discrimination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing geophysical data processing methods struggle with accurately distinguishing and removing noise, leading to distorted seismic signals and obscuring reservoir changes, necessitating improved statistical analysis for better seismic interpretation and reservoir management.
Innovation Solution
A computer-implemented method using a trans-dimensional sampler in a Bayesian framework to automatically discriminate noise sources and partition geophysical data, employing a Markov Chain Monte Carlo algorithm to estimate error distributions and generate probability weight maps, allowing for continuous probability evaluation of data points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual expert evaluation is used to discriminate noise in geophysical data, then subjective judgment can be applied, but the process is time-consuming and does not provide statistical information
Solution Approach 1:
The patent replaces manual expert evaluation with an automated computational system using Markov Chain Monte Carlo algorithms. The system automatically computes probability distributions for different noise models and selects the optimal model, eliminating the need for time-consuming manual assessment while providing objective statistical information about noise characteristics in the geophysical data.
Solution Approach 2:
The system performs self-evaluation by automatically computing error distributions, comparing different noise models using statistical criteria, and selecting the optimal model without external human intervention. The algorithm independently assesses the quality of different partitioning schemes and noise models, making the entire noise discrimination process autonomous and efficient.
2Ease of operation
If binary keep/discard data evaluation is used, then simple decision-making is achieved, but full range probability estimates are not provided
Solution Approach 1:
The patent transforms the binary keep/discard evaluation into a continuous probability-based assessment. Instead of simple yes/no decisions, the system computes full probability distributions for different data points under various noise models, providing nuanced probability estimates that retain information about uncertainty and confidence levels while maintaining operational simplicity through automated computation.
3Device complexity
If fixed-dimensional error models are used, then model simplicity is maintained, but the ability to capture complex noise patterns is limited
Solution Approach 1:
The patent implements dynamic model selection by using Markov Chain Monte Carlo methods to explore different error model configurations and partitioning schemes. The system adaptively adjusts the complexity of the error model based on the data characteristics, selecting the optimal number of partitions and error model parameters that best fit the observed geophysical data while avoiding overfitting through probabilistic model comparison.
Data Source
Figure 1
Figure 2
Figure 3A~3B
AI summary
The present invention relates to a computer-implemented method and a computing device for geophysical data mining. The computer-implemented method comprising: (a) receiving, by a processor, a first survey data representative of a subsurface volume of interest and a second survey data representative of the subsurface volume of interest, wherein the first survey data and the second survey data are acquired at different survey times; wherein each of the first survey data and the second survey data comprise a plurality of data-points, wherein the plurality of data-points are arranged in a data matrix with Nw columns and Ns rows, each data-point of the plurality of data-points is associated to a j-th column of the data matrix representing a position xj along a x-axis and to an i-th row of the data matrix representing a position ti along a t-axis; (b) estimating, by a processor, error distribution data by estimating an error distribution of each of the plurality of data-points of the first survey data; (c) generating, by the processor, a current error model by partitioning the error distribution data in a number of rectangular cells, wherein the current model is defined by error model parameters, and wherein the error model parameters comprise the number of rectangular cells and, for each rectangular cell of the number of rectangular cells, a position on the x-axis of the center of the each rectangular cell, a size in the x-axis of the each rectangular cell, a position on the t-axis of the center of the each rectangular cell, a size in the t-axis of the each cell, and a weight of each cell; (d) modifying, by the processor, the error model parameters of the current error model to generate a candidate error model based on a probability condition; (e) storing, in a memory, the error model parameters of the candidate error model as one element of a Markov Chain based on a result of comparing a likelihood value of the current error model and a likelihood value of the candidate error model; (f) repeating, by the processor, steps (d) and (e) as the candidate error model being the current error model, until convergence of the Markov Chain; and (g) generating, by the processor, a probability weight map using a plurality of partition models generated by assigning a weight to each of the plurality of data-points based on the error model parameters of the candidate error models which form the Markov Chain.