Many are using machine learning as a buzzword, we use it to find similar wells. Read on to learn how features known from social media help you improve your well designs and cut planning times!

Posted 03.06.2020 07:50 by Magnus Tvedt

Many are using machine learning as a buzzword, we use it to find similar wells. Read on to learn how features known from social media help you improve your well designs and cut planning times!


Now before we dive into more details, I want to let you know that the purpose of this article is to help you plan wells, not brief you before a presentation on machine learning. I am simplifying and I avoid details to make this readable. Artificial Intelligence, AI, Neural Networks, that’s what this is all about. But I use the term Machine learning, because that’s the word I’m used to in my vocabulary.

Now, to well planning and efficiency gains. Imagine you are designing a well, and you want to find references from historic wells. As an experienced engineer, you pick a few significant features from your new well, and start reading up on reports from wells where you think relevant experiences are stored. And after a couple of weeks, dozens of meetings and spreadsheets later, you have an overview of which of your selected wells have similar traits.

That took you about three weeks, and you have found 3-4 similar features from your selection of reference (or offset if you like that term better) wells. Examples could be presence of swelling shales; angles held through formations; depth of casing shoes; horizontal reach of well path; fluid systems and overbalance, or something very field specific. And your documentation is of high value for you as you continue the planning of a well.

What if you could do all that work in a few seconds, effortless, with an almost unlimited number of variables?

Our unique tool let you scroll through wells to learn experiences like you browse search results in Google. In Google searches, all the results are listed, and they are prioritized based on a myriad of variables. Machine learning present you with a list based on the words you wrote, the order of your words, what anyone else clicked on after writing the same words, and many more features. You browse through to find the best match. You could make this list of search results with filters, bu they would be very complex, and you would have to know the all the details.

With our well planning, it works the same. You find your reference wells, and machine learning finds similar wells, based on all the available parameters. Take our stratigraphy model for instance, which can be seen in the video above. It is trained on all the stratigraphy from all your wells in your database and they are grouped together based on similarity from all the layers.

You just couldn’t write the filters to do that without machine learning. Now, you just click a button, and start browsing results.

Did we mess up the Axis labels?

What are the axis on the plots? Our machine learning models are represented in many ways but often in x and y axis scatter plots, because that’s something we manage to visualize, and us users are capable of understanding. But the models show similarity, and between two wells, it can be the thickness of a given cretaceous rock. Or it can be the presence of a salt, or the thickness of four young formations on top of each other. And that similarity is shown with distance between two dots in a x-y scatter plot.


We have had so much fun learning about similarities, how the machine recognizes patterns and make them significant. And we have learned endless traits of how an area, field or region groups together. But maybe the most inspiring moments is when we find correlations we struggle with explaining, because that means we are learning something new. And that happens almost daily for us.

Keep expanding your knowledge

We know that you also love to develop your skills and become a more advanced engineer, and that is why we continue developing more machine learning models in our products.

Share this