I’ve spent a lot of time this year talking about the process of Petro.ai, trying to give some background for how shale must be understood, analyzed and responded to— for getting the most out of your reservoir. We’ve worked with oil and gas companies around the world and learned in basin after basin that shale requires a different approach. Without being too dramatic, a revolution.
Petro.ai has a deep level of commitment to the industry and to our global resources-— to get the most out of every unconventional pad. And we know you do too. For the next few months, I’m going to be answering the questions we’ve had come in from clients and experts in the field. If you have one you’d like to add, send it to support@petro.ai. Let’s do this together.
Q: I have a frac simulator, can I use the output in Petro.ai?
A: Charles Connell, VP of Product, responds, “The short answer is yes, as an input in the Drainage Model.
“As part of our multi-step process for calibrating and validating the Digital Twin, one of the key steps is the Drainage Model. We use that as a feature when we’re predicting the well productivity.
“And consistently it’s one of, if not the highest in importance of all the features. We have our own way of creating this drainage area that uses geomechanics and some machine learning processes. That’s heavily built off diagnostic data collected in the field. Microseismic data is also used to inform the shape.
“There are other ways that we can calibrate that shape. For some of our clients that have reservoir or frack simulators, we can use that simulation to calibrate the inputs. If you don’t have microseismic, but you do have a simulation that can be a good way of validating and calibrating the drainage area that you have.
“Frequently companies, especially service providers, will do a frac simulation to understand what the fracture half-length is supposed to look like, what the fracture vertical growth is going to look like. Previously that could be used to inform some well spacing decisions, but the simulator requires a lot of technical inputs, sometimes a lot of assumptions around the make-up of the rock, the geology, and all these different variables.
“You might do one simulation for a well and even then, you’re not getting a forward prediction to production. You’re simulating the fracture but it’s not telling you how that’s impacting production. You might change some of those variables and can see how the shape of that fracture might change, but still, you can’t understand what the economic impact is of pumping more or less sand.
“Our model does allow for those forward predictions and economic evaluations.
“The value of using our model over just a simulation is that we can tell you what the well spacing tradeoffs are going to be in real dollars.
“We have some baseline models that we can start quickly with, then we want to fine tune those. One of the ways we can do that is by changing the size and see how that impacts the model accuracy on nearby pads. Another way is, if you have microseismic data that can be used to form the shape of those drainage images.
“Another way is with the oil fingerprint data which is like RevoChem. Some of our clients use RevoChem which is a diagnostic data type that they collect in the field that informs the vertical growth of the fracture. We can overlay that with our image to see if we’re predicting the vertical fracture growth, how that compares with RevoChem.
“Simulation data is another way of doing that. If you have a fracture simulation, then we can overlay that with our predicted drainage and see what the fracture height and width are going to be. We can use fracture simulations to calibrate and refine the model that we’re using to predict all the way through production and economic payouts.
“If you don’t have a simulator, and we’ve worked with a lot of clients that don’t have fracture simulations we’ll just use other techniques to tune in the model.
“If we don’t have simulator inputs, we can certainly proceed without it. If you have these other data types you can also use that to better inform the model. We’ll keep going through this iterative process of tuning the model and ingesting more data to improve the accuracy from one pad to the next.”
