Shale gas geological "sweet spot" parameter prediction method and its application based on convolutional neural...

Area of Science:

• Geophysics
• Petroleum Geoscience
• Artificial Intelligence in Earth Sciences

Background:

• Shale gas reservoir exploration relies on key parameters like gas content (GAS), porosity (PHI), and total organic carbon (TOC) to identify geological "sweet spots".
• Current prediction methods using well logging data offer high accuracy but lack the three-dimensional spatial distribution crucial for large-scale exploration.
• Prestack seismic inversion provides a basis for large-scale spatial predictions, but integrating it with well data for high-precision "sweet spot" identification remains a challenge.

Purpose of the Study:

• To develop a high-precision, large-scale "sweet spot" prediction method for shale gas reservoirs.
• To leverage deep learning, specifically 1D-CNN, to integrate well logging and seismic data for improved spatial prediction.
• To establish a robust model for identifying geological "sweet spots" in shale formations.

Main Methods:

• Intersection analysis of well logging data to identify sensitive "sweet spot" parameters.
• Development of a novel standardized data preprocessing technique tailored for well logging data.
• Design of a 1D-CNN framework capable of processing both depth-domain (well logging) and time-domain (seismic) data.
• Training a 1D-CNN model using well logging data to predict "sweet spot" parameters.

Main Results:

• A high-precision and robust geological "sweet spot" prediction model was successfully trained.
• The proposed 1D-CNN method demonstrated capability in predicting "sweet spot" parameters from both well logging and seismic data.
• Successful application of the method in the WeiRong shale gas field, Sichuan Basin, for high-precision "sweet spot" prediction in the Wufeng-Longmaxi shale reservoir.

Conclusions:

• The 1D-CNN approach offers a viable solution for high-precision, large-scale "sweet spot" prediction in shale gas exploration.
• Integrating well logging and seismic data through deep learning enhances the spatial understanding of reservoir potential.
• This method advances the capability for efficient and targeted exploration of shale gas resources.