Hybrid Deep Learning Approach to Speed up Certain Numerical Simulations
Hybrid Deep Learning Approach to Speed up Certain Numerical Simulations


Two kinds of equations play an important role in the upstream of oil and gas industry, one is the Darcy law (fluid through a porous medium), which is critical in conducting reservoir simulations for production forecasting; the other is wave equations that could illuminate the subsurface structure using seismic data.
In this presentation, we will demonstrate how to leverage deep learning to speed up production forecasting, as well as seismic imaging.
In order to accelerate reservoir production simulations, we build a recurrent neural network model using production forecasting results. The model can be viewed as a proxy and allow us to understand the reservoir much quicker. Once the model is released, we can rely on it to predict reservoir performance and make economic decisions . The model will be monitored and updated if necessary.
One difficulty is the amount of data sample, as deep learning usually requires a considerable amount of training data. However, we might not have enough data available, as the numerical simulations are computationally intense, and a large number of simulation scenarios could also become the bottleneck for the acceleration.
Another challenge is how to integrate statics features into dynamic features (inputs), as they exhibit varied impacts on production performance (outputs). Dynamic inputs, e.g., wellhead Pressure (WHP), would change the day-to-day production outlook, and static features such as permeability, oil-water contact (OWC) and gas oil contact (GOC) would affect the cumulative production. In our deep learning approach, we will demonstrate how to integrate those features into one model.
When it comes to seismic imaging, we attempt to decode the wave equations and inversions in the ML framework. First, we started with some randomly generated depth velocity models, and perform forward modeling to generate shot gathers. Next, we run reverse time migration (RTM) to generate depth images. Once the training data set is ready, we leverage the state-of-the-art of Machine Learning to extract features from shots and velocity model to generate seismic images. Once we have a satisfied ML model, the future work could be focusing on applying reinforcement learning to uplift velocity model building.


Cheng Zhan is a senior data scientist at Microsoft's Cloud AI group. Prior to his current role, he worked in the upstream oil and gas, mainly focusing on the exploration (geoscience) and production (engineering) domains, like physics informed machine learning, and production forecast for unconventional reservoirs. He has a Ph.D. in mathematics from University of Houston, and a B.S. in mathematics from Sun Yat-sen University in Guangzhou, China.

Open Data Science




Open Data Science
One Broadway
Cambridge, MA 02142

Privacy Settings
We use cookies to enhance your experience while using our website. If you are using our Services via a browser you can restrict, block or remove cookies through your web browser settings. We also use content and scripts from third parties that may use tracking technologies. You can selectively provide your consent below to allow such third party embeds. For complete information about the cookies we use, data we collect and how we process them, please check our Privacy Policy
Consent to display content from - Youtube
Consent to display content from - Vimeo
Google Maps
Consent to display content from - Google