GNN-Based TOC Prediction in Shale Using Multi-Log Graph Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting total organic carbon (TOC) in shale reservoirs are limited by their inability to fully analyze the complex nonlinear relationships between logging curves and TOC, often relying on incomplete parameter sets and linear modeling approaches.
Innovation Solution
A graph neural network (GNN)-based prediction system that processes multiple logging curves, including radioactive uranium, thorium, potassium, acoustic velocity, and resistivity logs, to accurately predict TOC by employing a trained GNN model with a weight matrix and spatial-temporal convolution blocks, enabling the analysis of nonlinear relationships and sensitivity analysis to improve prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional wireline logging parameters are used to predict TOC, then the prediction process can be performed, but the complex nonlinear relationship between logging parameters and TOC cannot be fully characterized
Solution Approach 1:
The patent replaces traditional statistical equations and linear modeling methods with a graph neural network (GNN) model. The GNN model uses graph convolutional layers to automatically learn and characterize the complex nonlinear relationships between logging parameters and TOC, eliminating the need for manual model construction and achieving superior prediction accuracy without requiring complex mathematical derivations.
Solution Approach 2:
The patent transforms the prediction approach by changing from fixed statistical parameters (binomial fitting, monomial fitting) to adaptive neural network parameters. The GNN model dynamically adjusts its internal parameters during training to optimize the characterization of nonlinear relationships, allowing the model to adapt to different logging parameter combinations and achieve high prediction accuracy across varying geological conditions.
2Device complexity
If only P-wave impedance data is used for TOC prediction, then the prediction process is simplified, but the characterization ability to describe complex nonlinear relationships is insufficient
Solution Approach 1:
The patent creates a universal prediction framework that can handle multiple logging parameters (acoustic, gamma-ray, resistivity, density, etc.) within a single GNN model architecture. The model universally processes different types of logging data through standardized graph convolutional layers, enabling it to capture complex nonlinear relationships across all logging parameters simultaneously, rather than requiring separate models for each parameter type.
Solution Approach 2:
The patent transitions from one-dimensional single-parameter prediction (using only P-wave impedance) to multi-dimensional multi-parameter prediction. By constructing a graph structure that incorporates multiple logging parameters as nodes and their relationships as edges, the GNN model operates in a higher-dimensional feature space, enabling simultaneous analysis of interactions between acoustic, gamma-ray, resistivity, and density logs to achieve comprehensive TOC characterization.
3Ease of manufacture
If simple linear modeling methods are used, then the model is easier to construct, but it cannot characterize the complex nonlinear relationship between TOC and logging parameters
Solution Approach 1:
The patent replaces manual linear model construction with automated GNN training. The graph neural network automatically learns optimal feature representations and nonlinear relationships through backpropagation and gradient descent, eliminating the need for manual model specification. This automated approach maintains ease of use while achieving superior accuracy in characterizing complex nonlinear relationships between logging parameters and TOC.
4Device complexity
If monomial fitting is used to model the relationship between TOC and logging parameters, then the modeling process is simplified, but the characterization ability for complex nonlinear relationships remains insufficient
Solution Approach 1:
The patent substitutes explicit mathematical fitting functions (monomial, binomial) with implicit neural network representations. The GNN model captures complex nonlinear relationships through composed transformations of neural network layers rather than closed-form mathematical equations. This substitution maintains modeling simplicity while enabling the model to capture arbitrary nonlinear patterns in the data through learned feature representations and interactions.
Data Source
AI summary
A graph neural network (GNN)-based prediction system for total organic carbon (TOC) in shale solves the problem that the existing shale TOC prediction method cannot fully analyze the complex nonlinear relationship between all logging curves and TOC. The prediction system adopts a method including: acquiring and preprocessing a plurality of logging curves of a target well location in a target shale bed to acquire a plurality of standardized logging curves, windowing the plurality of standardized logging curves, and inputting the windowed logging curves and weight matrix into a trained GNN-based TOC prediction network to acquire TOC of the target well location. The prediction system inputs the plurality of logging curves as correlative multi-dimensional dynamic graph data for analysis and can acquire the complex nonlinear relationship between the logging curves and TOC, thus improving the prediction accuracy of TOC.