May-26-22 Thursday Daily Update - Mohammad Elkady

 

Daily Update - Mohammad Elkady - (May-26-2022)

1.  What did I do today?

Today, I did more research about the Machine Learning application in Chemical EOR and Surfactant simulation using CMG STARS, Also I read SPE-195492-MS. 

 

2.  What did I find interesting/Work on?

I gathered 6 more papers addressing Surfactant simulation using CMG STARS and Machine learning application in Chemical EOR.

In this paper (SPE 195493) they constructed 1,100 ANN training cases using CMG-STARS to capture the variation in reservoir petrophysical properties and the range of injected chemicals properties for a five-spot pattern.

we generated a total of 1,100 training cases by feeding the reservoir characteristics and design parameters into the model to obtain the reservoir response.

·      ANN is a type of artificial intelligence inspired by actual biological neural networks that constitute the human brain. Most common structure is most common of which is the multi-layer network with back propagation.

·      Reservoir of interest is Sandstone reservoir. (Conventional)

·      For the reservoir response, they recorded three parameters for ANN training:

o   Producer oil rate, Producer water cur and Injector bottomhole pressure.

·      Model #1: is a forward model that predicts "reservoir response" (oil flowrate, cumulative oil production, water cut, and injector BHP). The reservoir characteristics and design parameters are trained as inputs and provide production profile prediction.

·      Model #2: is an inverse model that predicts "reservoir characteristics" (thickness, permeability, residual oil saturations, and adsorption parameters). The reservoir response and design parameters were fed as inputs into this model. This model works as a history matching tool, assuming the reservoir response is known, which is usually after the field execution.

·      Model #3 is an inverse model that predicts "project design parameters" (pattern size, chemical slug size, chemical concentration, and chemical injection rate). The inputs for this model are reservoir response and reservoir characteristics. This model can assess the design parameters post-field execution for quality-check purposes.

·      Model #4 is another inverse model that predicts the same "project design parameters" as Model #3. However, the inputs are reservoir characteristics and target cumulative oil and project duration instead of the reservoir response. This model can support field expansion decisions prior to field execution.

·      Model #5 is also an inverse model that predicts not only the same "project design parameters" as Model #3 but also the project duration for a target cumulative oil. The required inputs for this model are only the reservoir characteristics and the target cumulative oil. This model is similar to Model #4, but can predict the project duration, which is beneficial for economical evaluation prior to field implementation.

The table below is the parameters ranges used to generate the data using CMG  and also used to construct the ANN Models.

 

 

3.  What will I do next? 

I will keep looking for other literature about Artificial Neural Network in surfactant EOR (Seems like there is more literature about it) and I will read more about Surfactant simulation using CMG STARS software to get familiar to the procedure. Also, to get familiar to software itself I will watch more tutorials about CMG STARS. Finally, ANN seems to be the most used deep learning model in chemical EOR so I will enroll in an online course about them to know more about it.