| CPC G01V 1/30 (2013.01) [G01V 1/282 (2013.01); G01V 1/345 (2013.01); G06N 3/091 (2023.01)] | 12 Claims |

|
1. A computer-implemented method for efficient seismic inversion that does not require a user-specified low frequency model (LFM) as input, comprising:
a. receiving, at one or more computer processors, well logs and seismic angle stacks representative of a subsurface volume of interest;
b. deriving, via the one or more computer processors, rock physics models from the well logs and the seismic angle stacks;
c. performing, via the one or more computer processors, low frequency Markov Chain Monte Carlo (MCMC) processes on the rock physics models to generate low frequency models (LFMs) of rock properties, wherein the low frequency models (LFMs) are lower than 15 Hz;
d. training a deep neural network using the LFMs, the rock physics models, and the seismic angle stacks as training pairs of {log properties, seismic data} to generate a trained neural network;
e. providing a seismic dataset to the trained neural network to generate a high frequency rock property model;
f. performing broad-band MCMC processes on the high frequency rock property model to generate an ensemble of high frequency realizations of rock properties; and
g. using the ensemble of high frequency realizations of rock properties to identify hydrocarbon deposits and recover hydrocarbons.
|