Saudi Arabian Oil Company

Robert Smith

Andrey Bakulin

Philippe Georges Christophe Nivlet

Nasher Muqbel AlBinHassan

Nasher Muqbel AlbinHassan

Abdullah Alramadhan

Rayan Kanfar

Stanislav Glubokovskikh

Philippe Nivlet

Systems and methods are provided for seismic well tie domain conversion. In one embodiment, a process is provided to integrate well and seismic data for reservoir characterization. System configurations and processes described herein use neural networks to predict sonic well logs in the two way time (TWT) domain from measured well logs in depth, rather than predicting drift function. Embodiments are also directed to systems for reservoir characterization. Domain conversion of data includes receiving input data, preprocessing the data, and training a model to determine a length of an output sequence. The method also includes training the model for conversion of data based on at least one neural network. A sequence length prediction may be output as part of training and to perform modeling/prediction operations. The method also includes outputting the sequence of converted data in a TWT domain and a length of the sequence.

Systems and methods for seismic well tie domain conversion and neural network modeling

A method is claimed for automatically transforming sonic well logs from a depth domain to a seismic two-way travel-time domain. The method includes obtaining a training well with a measured sonic well log in the depth domain and a borehole seismic dataset in the depth domain and obtaining an application well with only a measured sonic well log in the depth domain. The method further includes training a network to predict a transformed sonic well log for the training well based, at least in part, on the measured sonic well log and the borehole seismic dataset in the training well, and predicting with the network, the transformed sonic well log in the application well.

System and method for automated domain conversion for seismic well ties

Methods for determination of mechanical properties of geological formations using deep learning include receiving, by a computer system, data acquired during drilling a geological formation. The computer system generates features of the data acquired during drilling. The features are indicative of mechanical properties of the geological formation. The computer system segments the features of the data acquired during drilling into sequences readable by a trained temporal convolutional network (TCN). The computer system determines the mechanical properties of the geological formation using the TCN based on the sequences obtained from the features of the data. A display device of the computer system generates a graphical representation of the mechanical properties of the geological formation.

Determination of mechanical properties of a geological formation using deep learning applied to data acquired while drilling

Methods for determination of elastic properties of geological formations using machine learning include extracting a first feature vector from data acquired during drilling. The data includes at least drilling parameters. The first feature vector is indicative of a drilling environment classification. A machine learning classification algorithm determines the drilling environment classification based on the first feature vector. A machine learning regression algorithm is selected from multiple machine learning regression algorithms based on the drilling classification. A second feature vector is extracted from the data acquired during drilling based on the drilling classification and the selected machine learning regression algorithm. The second feature vector is indicative of elastic properties of a geological formation. The selected machine learning regression algorithm determines the elastic properties of the geological formation based on the second feature vector.

Determination of elastic properties of a geological formation using machine learning applied to data acquired while drilling

A method of evaluating processing imprint on seismic signals includes receiving a first and a second seismic dataset of a reservoir. A first and a second synthetic dataset are generated, where the second synthetic dataset is generated by multiplying at least a portion of data in the first synthetic dataset by a scaling factor. A first and a second combined dataset are generated by adding the respective seismic dataset and the respective synthetic dataset. A first and a second processed dataset are generated by applying a seismic processing step on the first and the second combined dataset, respectively. A difference factor between the first and the second processed dataset is calculated. Based on the difference factor and the scaling factor, it is determined whether the seismic processing step is able to preserve signal amplitude changes between the first and the second seismic dataset.

Growing community of inventors

Dhahran, Saudi Arabia

Robert Smith