Multimodal data driven deep learning based seismic impedance inversion optimization

Area of Science:

• Geophysics
• Machine Learning
• Subsurface Characterization

Background:

• Seismic impedance inversion is crucial for subsurface property analysis.
• Conventional methods struggle with noise, low resolution, and complex geology.

Purpose of the Study:

• To improve seismic resolution and synthetic seismogram generation using deep learning.
• To address limitations of conventional seismic inversion techniques.

Main Methods:

• Deep learning models (LeNet, AlexNet, CNNs) were employed.
• Continuous Wavelet Transform (CWT) was used for feature extraction.
• Models were trained on synthetic data and validated on real seismic data.

Main Results:

• AlexNet achieved superior performance in seismic data reconstruction.
• AlexNet demonstrated the lowest Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE).
• AlexNet yielded the highest R2 score, indicating excellent predictive accuracy.

Conclusions:

• Deep learning, particularly AlexNet, offers a robust benchmark for advanced geophysical analysis.
• The proposed technique enhances subsurface characterization and reduces geological risks.
• Improved seismic data reconstruction leads to more informed drilling decisions.