Seismic-to-well tie using neural network inversion and evolutionary algorithms
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional seismic imaging methods struggle with maintaining high vertical resolution when seismic wavelets propagate at high rates, especially in carbonated subsoils, leading to uncertainties in well location and seismic data correlation.
Innovation Solution
A method utilizing a trained neural network and evolutionary algorithms to determine a seismic-to-well tie by iteratively refining well location candidates and retraining the neural network, thereby enhancing the correlation between seismic data and well locations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional seismic imaging tools are used with high propagation rate, then processing speed is maintained, but vertical resolution of the seismic image deteriorates
Solution Approach 1:
The patent applies dynamic time warping to adaptively adjust the time scaling between seismic data and well data, allowing the system to dynamically compensate for propagation rate variations. This enables maintaining processing efficiency while achieving accurate vertical resolution by flexibly transforming time domains rather than being constrained by fixed propagation rates
Solution Approach 2:
The invention changes the time parameter through dynamic time warping transformations, modifying the time scaling factor to optimize the match between seismic and well data. By adjusting this parameter, the system resolves the contradiction between processing speed and vertical resolution without requiring recalibration of the entire imaging system
2Area of stationary object
If seismic wavelet is propagated at high rate in carbonated subsoil, then imaging coverage is maintained, but time domain accuracy deteriorates leading to significant spatial domain variations
Solution Approach 1:
The patent uses an iterative feedback mechanism where the dynamic time warping transformation is applied, evaluated against well data, and refined through multiple cycles. This feedback loop continuously improves time domain accuracy by adjusting the time scaling factor based on the mismatch between seismic and well data, thereby reducing spatial domain variations in carbonated subsoil
Solution Approach 2:
The system performs preliminary dynamic time warping transformation before final imaging to pre-correct for propagation rate uncertainties. This preliminary action prepares the seismic data with optimized time scaling, preventing the accumulation of time domain errors that would otherwise lead to significant spatial domain variations
3Ease of manufacture
If well location is not accurately related to seismic data, then data processing simplicity is maintained, but seismic image definition and information extraction deteriorate
Solution Approach 1:
The patent replaces the mechanical alignment process with an intelligent computational system using dynamic time warping and iterative optimization. Instead of manual or simple automated well-seismic tying, the system uses machine learning algorithms to automatically optimize the time scaling and alignment, maintaining processing simplicity while dramatically improving information extraction and image definition
Data Source
Figure 1
Figure 2a~2b
Figure 3a~3b
AI summary
Disclosed is a method for determining a seismic-to-well tie for a plurality of wells comprised within a reservoir zone. The method comprises generating a first candidate population of well locations for each of said plurality of wells and using a trained neural network to invert seismic data relating to said reservoir zone in accordance with said candidate population of well locations, to obtain inversion well data. The inversion well data is evaluated by comparing said inversion well data to validation well data relating to the reservoir zone. An evolutionary algorithm is applied to said candidate population of well locations to obtain an updated candidate population of well locations. A further training said trained neural network is performed in accordance with said updated candidate population of well locations, and the steps iteratively repeated until at least one stopping criterion is reached.