Transient Electromagnetic Inversion With Target-Dataset Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep learning-based electromagnetic inversion methods are data-driven and rely heavily on large-scale training datasets, leading to unsatisfactory performance when faced with unseen data, and they require significant computational resources and are prone to overfitting.
Innovation Solution
A method involving constructing an initial training set, training a convolutional neural network, performing forward simulation, extracting similar data, and iteratively learning on a target dataset to improve inversion accuracy and reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If data-driven deep learning methods are used for electromagnetic inversion, then inversion speed is improved, but inversion accuracy deteriorates when facing unseen data
Solution Approach 1:
The patent implements an iterative feedback mechanism where the network predictions are used to generate synthetic data that feeds back into retraining the network. This feedback loop allows the model to continuously improve its accuracy on unseen data while maintaining the speed advantages of deep learning methods.
Solution Approach 2:
The patent performs preliminary training on synthetic data generated from physical models before applying the network to real inversion problems. This preliminary action prepares the network to handle unseen data more effectively by pre-learning the underlying physical relationships.
2Reliability
If large-scale training datasets are constructed for deep learning inversion, then network performance is improved, but computational resource consumption increases
Solution Approach 1:
The patent employs a self-service approach where the network itself generates the training data through forward modeling. Instead of requiring external construction of large datasets, the network uses its own predictions to create synthetic training samples, significantly reducing computational resource requirements.
Solution Approach 2:
The patent merges the data generation process with the training process by integrating forward modeling capabilities into the deep learning framework. This combination eliminates the need for separate large-scale dataset construction and reduces overall computational consumption.
3Ease of manufacture
If pure data-driven deep learning methods are used, then training process is simplified, but handling of fine details between data and models deteriorates
Solution Approach 1:
The patent changes the approach by introducing physical model parameters into the training process. By retraining the network with synthetic data that includes ground truth models, the network learns to handle fine details and physical relationships that pure data-driven methods miss.
Solution Approach 2:
The patent introduces synthetic data as an intermediary between the simple training process and the complex inversion problem. This intermediary layer allows the network to learn fine details and physical relationships while maintaining training simplicity.
Data Source
AI summary
A method for constructing transient electromagnetic inversion model construction method driven by a target dataset is provided, including: constructing an initial training set, and training a first convolutional neural network using the initial training set to obtain an initial transient electromagnetic inversion network model; inputting electromagnetic response data to be measured into the initial transient electromagnetic inversion network model for prediction to obtain resistivity values; performing forward simulation on resistivity values to obtain forward-simulated electromagnetic response data, and constructing prediction data based on forward-simulated electromagnetic response data and resistivity values; extracting similar data to test electromagnetic response data from the initial training set, and constructing a target dataset based on similar data and prediction data; migrating parameters from the initial transient electromagnetic inversion network model to a second convolutional neural network to perform iterative learning on the target dataset, and obtaining a transient electromagnetic inversion model.