Landmark Graphics Corporation

Halliburton Energy Services, Inc.

Ajay Pratap Singh

Jeffrey Marc Yarus

Marko Maucec

Gustavo Carvajal

Dwight David Fulton

Seyed M Mirzadeh

Hasnain A Khan

Luigi Saputelli

Ashwani Dev

Satyam Priyadarshy

Steven Patton Knabe

Michael Nikolaou

Suresh Venugopal

Geetha Gopakumar Nair

Roxana Nielsen

Luigi A Saputelli

Sayed M Mirzadeh

Gustavo Adolfo Carvajal

Srimoyee Bhattacharya

The subject disclosure provides for a mechanism implemented with neural networks through machine learning to predict wear and relative performance metrics for performing repairs on drill bits in a next repair cycle, which can improve decision making by drill bit repair model engines, drill bit design, and help reduce the cost of drill bit repairs. The machine learning mechanism includes obtaining drill bit data from different data sources and integrating the drill bit data from each of the data sources into an integrated dataset. The integrated dataset is pre-processed to filter out outliers. The filtered dataset is applied to a neural network to build a machine learning based model and extract features that indicate significant parameters affecting wear. A repair type prediction is determined with the applied machine learning based model and is provided as a signal for facilitating a drill bit operation on a cutter of the drill bit.

Drill bit repair type prediction using machine learning

A hydrocarbon field including thief zones can be modeled based on production history data and supplemental data constraining a thief-zone distribution across the field. In various embodiments, a numerical optimization approach involves creating a plurality of model thief-zone distributions with varying parameter values, parameterizing permeability maps obtained for the thief-zone distributions and aggregating across the thief-zone distributions to obtain distributions of coefficients associated with the parameterization, and then iteratively sampling the coefficient distributions, determining a permeability map of the model based on the selected coefficients, and measuring a deviation between the measured production history data and simulated production history data derived from the computational model until a match is achieved.

History matching of hydrocarbon production from heterogenous reservoirs

Systems and methods for a hybrid approach to assisted history matching in large reservoirs based on a reservoir model built using connectivity between each production well and each corresponding injection well, aquifer or gas cap.

Hybrid approach to assisted history matching in large reservoirs

Systems and methods for creating virtual production logging tool profiles for improved history matching and proactive control of smart wells.

Creating virtual production logging tool profiles for improved history matching

The present disclosure describes systems and methods for performing multi-level reservoir history matching. At least some illustrative embodiments include a method that includes generating a first history-matched model using at least one updated model parameter derived from one or more existing model parameters, generating a set of second history-matched models by applying a probabilistic inversion to the first history-matched model, and deriving a set of third history-matched models from the set of second history-matched models. The method further includes generating dynamic simulation realization sets using each of the set of third history-matched models, ranking each of the set of third history-matched models based at least in part on the dynamic simulation realization sets, and presenting a production forecast to a user based on the highest ranked third history-matched models.

Multi-level reservoir history matching

Growing community of inventors

Houston, TX, United States of America

Ajay Pratap Singh