Gas exploration is a complex and time-consuming task that requires a large investment of time and resources. Traditional seismic interpretation methods are often inaccurate and inefficient, especially in areas with inconspicuous targets or low signal-to-noise ratios.
The invention consists of an artificial intelligence algorithm that uses recurrent neural networks to identify seismic signatures of gas accumulation. The invention treats seismic data as a set of traces, which are one-dimensional signals. This allows the invention to use recurrent neural networks, which are more suitable for processing temporal data, as they are able to learn long-range patterns, which is essential for identifying seismic gas signatures, which can extend over several traces.
In addition, the proposed methodology does not require the existence of seismic anomalies, which makes it more accurate and efficient. These characteristics are advantageous compared to traditional methods.
The invention works in three stages: first, the seismic data is pre-processed to remove noise and normalize the amplitude values. Next, the neural network is trained with a labeled data set, which contains seismic data traces with and without gas signatures. Finally, the neural network is used to detect seismic gas signatures in new seismic data. This patent uses deep learning to perfect the process.
Patent title:
Method for detecting gas reservoir signatures in seismic surveys
Deposit Number:
BR 10 2020 010867 0
Pontifical Catholic University of Rio de Janeiro – PUC-Rio
Rua Marquês de São Vicente, 225, Gávea
Rio de Janeiro, RJ – Brasil
Zip code: 22451-900
Postal box: 38097
Phone:
+55 (21) 3527-2155