Deep Learning Geological Modeling with Sparse Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current geological modeling technologies face challenges in efficiently handling complex structural modeling, maintaining geological consistency, and accurately representing intricate geological features, especially in regions with sparse or unevenly distributed data.
Innovation Solution
The proposed method employs a convolutional neural network (CNN) for interactive implicit modeling, utilizing multi-source heterogeneous data to generate a full geological structure model. This approach involves acquiring and converting fault and horizon interpretation data, which are then input into a pre-trained neural network to produce a three-dimensional implicit structure model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional geological modeling methods are used to handle complex structural modeling, then the modeling process can be completed, but the efficiency is low and performance bottlenecks occur
Solution Approach 1:
The patent replaces traditional mechanical/mathematical modeling systems with a deep learning-based neural network system. The neural network learns geological patterns from training data and automatically generates consistent structural models, substituting manual or algorithmic geometric construction with intelligent pattern recognition and synthesis.
Solution Approach 2:
The patent performs preliminary training of the neural network using extensive geological data and expert knowledge before actual modeling. This pre-learning phase enables the model to internalize geological constraints and patterns, allowing it to rapidly generate consistent results during deployment without repeatedly solving complex mathematical equations.
2Measurement precision
If sparse or unevenly distributed data is used for modeling, then the modeling process can proceed, but the accuracy of representing intricate geological features deteriorates
Solution Approach 1:
The neural network learns from comprehensive training data that includes complete geological patterns and uses this learned knowledge to generate accurate representations even when input data is sparse. The model effectively copies geological patterns from the training distribution to the target region, inferring missing features based on learned relationships rather than direct observation.
Solution Approach 2:
The patent transforms the modeling approach from direct data interpolation to probabilistic pattern generation. By changing from deterministic geometric construction to stochastic sampling from learned distributions, the system can handle sparse data more effectively, generating multiple plausible configurations and selecting the most geologically reasonable one.
3Reliability
If conventional modeling approaches are used, then the process is straightforward, but the ability to maintain geological consistency in complex structures deteriorates
Solution Approach 1:
The patent merges multiple sources of geological knowledge and constraints into a unified neural network model. By combining structural geology principles, stratigraphic relationships, fault mechanics, and other domain knowledge into the training process, the system maintains geological consistency across all these aspects simultaneously rather than treating them as separate constraints.
Solution Approach 2:
The neural network model serves multiple functions: it performs structural interpretation, maintains geological consistency, generates three-dimensional models, and ensures kinematic feasibility all through a single unified system. This multi-functionality replaces multiple specialized tools and manual processes that would otherwise be needed to achieve the same goals.
4Measurement precision
If more data sources are integrated to improve modeling accuracy, then the model quality improves, but the data processing complexity increases
Solution Approach 1:
The neural network is designed to accept multiple types of heterogeneous data inputs (seismic data, well logs, geological maps, etc.) and process them through a unified architecture. This multi-functional input layer automatically adapts to different data types and integrates them in a coordinated manner, reducing the need for separate processing pipelines for each data source.
Solution Approach 2:
The patent uses a unified neural network architecture as an intermediary that mediates between multiple heterogeneous data sources and the final geological model. This intermediary layer automatically performs data harmonization, feature extraction, and integration, replacing complex manual data processing and correlation procedures.
Data Source
AI summary
A method of modeling interactive intelligent three-dimensional implicit structure based on deep learning is provided, including: generating a plurality of geological simulation structural models by using a data simulation technology, to construct a geological simulation structural model library, where the geological simulation structural model has a diversified fold and a fault feature; acquiring, for each, model, geological fault data and unevenly distributed geological horizon data to obtain a training sample data set; training a neural network by using the training sample data set; inputting multi-source heterogeneous data of a target region into a trained neural network, so as to output a geological structural model corresponding to the multi-source heterogeneous data, where the multi-source heterogeneous data includes fault interpretation data and horizon interpretation data, and the multi-source heterogeneous data includes at least one selected from: geological outcrop observation data, well logging data, various geophysical data or empirical knowledge data.