US 12,085,685 B2
Systems and methods for seismic well tie domain conversion and neural network modeling
Philippe Georges Christophe Nivlet, AlKhobar (SA); Robert James Smith, Dhahran (SA); and Nasher Muqbel AlBinHassan, Dhahran (SA)
Assigned to Saudi Arabian Oil Company, Dhahran (SA)
Filed by Saudi Arabian Oil Company, Dhahran (SA)
Filed on Oct. 11, 2021, as Appl. No. 17/498,276.
Claims priority of provisional application 63/109,007, filed on Nov. 3, 2020.
Prior Publication US 2022/0137245 A1, May 5, 2022
Int. Cl. G01V 1/28 (2006.01); G06N 3/084 (2023.01)
CPC G01V 1/282 (2013.01) [G06N 3/084 (2013.01); G01V 2210/43 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A system for seismic well tie domain conversion, the system comprising:
one or more processors; and
a non-transitory computer-readable memory storing instructions that, when executed by the one or more processors, causes the one or more processors to:
receive input data for a field region, the input data including depth domain data and time domain data for at least one well in the field region;
preprocess the input data to generate training data for the field region;
train a well tie model to determine a length of an output sequence using the training data, wherein the well tie model is a neural network configured to determine a length of an output in a time domain for well data received in a depth domain, and wherein training the well tie model to determine the length of the output sequence includes selecting at least one hyper-parameter, generating a vector of output sequences in a time domain for a batch of data and modifying weights of the well tie model using a back-propagation algorithm to reduce error relative to expected output for the batch of data, wherein training further includes reducing a value of a validation loss function by performing a Bayesian hyper-parameter optimization;
train the well tie model to convert well data using the neural network, wherein the model is trained to convert a sequence of sonic log data in a depth domain to a sequence in a time domain;
transform input data in the depth domain to the time domain using the well tie model, wherein transforming is performed using the well tie model and determined length of output sequence; and
output the transformed data.