Exxonmobil Upstream Research Company

Exxonmobil Technology and Engineering Company

Exxonmobil Research and Engineering Company

Kuang-Hung Liu

Huseyin Denli

Wei D Liu

Cody J Macdonald

Michael H Kovalski

Niranjan A Subrahmanya

Thomas A Badgwell

Diego A Hernandez

Victoria M Som De Cerff

Stijn De Waele

Shivakumar Kameswaran

Myun-Seok Cheon

Dimitri J Papageorgiou

Francisco Trespalacios

Mary Johns

Jacquelyn Daves

A computer-implemented method for detecting geological elements or fluid in a subsurface from seismic images is disclosed. Seismic data may be analyzed to identify one or both of fluid or geologic elements in the subsurface. As one example, the analysis may include unsupervised learning, such as variational machine learning, in order to learn relationships between different sets of seismic data. For example, variational machine learning may be used to learn relationships among partially-stack images or among pre-stack images in order to detect hydrocarbon presence. In this way, an unsupervised learning framework may be used for learning a Direct Hydrocarbon Indicator (DHI) from seismic images by learning relationships among partially-stack or pre-stack images.

Detection of hydrocarbon presence in subsurface from seismic images using relational learning

Aspects of the technology described herein comprise a surrogate model for a chemical production process. A surrogate model is a machine learned model that uses a collection of inputs and outputs from a simulation of the chemical production process and/or actual production data as training data. Once trained, the surrogate model can estimate an output of a chemical production process given an input to the process. Surrogate models are not directly constrained by physical conditions in a plant. This can cause them to suggest optimized outputs that the not possible to produce in the real world. It is a significant challenge to train a surrogate model to only produce outputs that are possible. The technology described herein improves upon previous surrogate models by constraining the output of the surrogate model to outputs that are possible in the real world.

Surrogate model for a chemical production process

A method and system of machine learning-augmented geophysical inversion includes obtaining measured data; obtaining prior subsurface data; (a) partially training a data autoencoder with the measured data to learn a fraction of data space representations and generate a data space encoder; (b) partially training a model autoencoder with the prior subsurface data to learn a fraction of model space representations and generate a model space decoder; (c) forming an augmented forward model with the model space decoder, the data space encoder, and a physics-based forward model; (d) solving an inversion problem with the augmented forward model to generate an inversion solution; and iteratively repeating (a)-(d) until convergence of the inversion solution, wherein, for each iteration: partially training the data and model autoencoders starts with learned weights from an immediately-previous iteration; and solving the inversion problem starts with super parameters from the previous iteration.

Machine learning-augmented geophysical inversion

A method and apparatus for automated seismic interpretation (ASI), including: obtaining trained models comprising a geologic scenario from a model repository, wherein the trained models comprise executable code; obtaining test data comprising geophysical data for a subsurface region; and performing an inference on the test data with the trained models to generate a feature probability map representative of subsurface features. A method and apparatus for machine learning, including: an ASI model; a training dataset comprising seismic images and a plurality of data portions; a plurality of memory locations, each comprising a replication of the ASI model and a different data portion of the training dataset; a plurality of data augmentation modules, each identified with one of the plurality of memory locations; a training module configured to receive output from the plurality of data augmentation modules; and a model repository configured to receive updated models from the training module.

Automated seismic interpretation systems and methods for continual learning and inference of geological features

A method and apparatus for machine learning for use with automated seismic interpretation include: obtaining input data; extracting patches from a pre-extraction dataset based on the input data; transforming data of a pre-transformation dataset based on the input data and geologic domain knowledge and/or geophysical domain knowledge; and generating augmented data from the extracted patches and the transformed data. A method and apparatus for machine learning for use with automated seismic interpretation include: a data input module configured to obtain input data; a patch extraction module configured to extract patches from a pre-extraction dataset that is based on the input data; a data transformation module configured to transform data from a pre-transformation dataset that is based on the input data and geologic domain knowledge and/or geophysical domain knowledge; and a data augmentation module configured to augment data from the extracted patches and the transformed data.

Data augmentation for seismic interpretation systems and methods

A method and apparatus for seismic interpretation including machine learning (ML). A method of training a ML system for seismic interpretation includes: preparing a collection of seismic images as training data; training an interpreter ML model to learn to interpret the training data, wherein: the interpreter ML model comprises a geologic objective function, and the learning is regularized by one or more geologic priors; and training a discriminator ML model to learn the one or more geologic priors from the training data. A method of hydrocarbon management includes: training the ML system for seismic interpretation; obtaining test data comprising a second collection of seismic images; applying the trained ML system to the test data to generate output; and managing hydrocarbons based on the output. A method includes performing an inference on test data with the interpreter and discriminator ML models to generate a feature probability map representative of subsurface features.

Training machine learning systems for seismic interpretation

Systems and methods are provided for using a Deep Reinforcement Learning (DRL) agent to provide adaptive tuning of process controllers, such as Proportional-Integral-Derivative (PID) controllers. The agent can monitor process controller performance, and if unsatisfactory, can attempt to improve it by making incremental changes to the tuning parameters for the process controller. The effect of a tuning change can then be observed by the agent and used to update the agent's process controller tuning policy. It has been unexpectedly discovered that providing adaptive tuning based on incremental changes in tuning parameters, as opposed to making changes independent of current values of the tuning parameters, can provide enhanced or improved control over a controlled variable of a process.

Growing community of inventors

Basking Ridge, NJ, United States of America

Kuang-Hung Liu