Q: How do you scale for lateral length?
A: As our clients are moving away from the mile and a half lateral length to longer horizontal distances that are at least 2 miles, sometimes 3, they’re interested in understanding the extra degradation that they might see in the greater lateral length. “The trend toward long laterals began in the Marcellus and is spreading in the Permian where operators are taking advantage of major gains in drilling and completions operations,” JPT reported earlier this year. Pioneer, the biggest operator in the Permian, is a strong advocate of longer laterals with about 19% of wells drilled between 2020-2021 having a length greater than 12,500 ft.
“Lateral length is an important feature in the Petro.ai Drilling Spacing Unit Design Service (DSUDS),” Charles Connell, COO explained.“ We’ll start with a very simple model that maybe has five, seven, eight features and lateral length will be one of them. The model will look at all of the wells in the training set and will treat lateral length like the other features that we’re using. As we sophisticate the model we’ll add and remove different features. Usually, lateral length remains included.
“The lateral length is scaled as the machine learning algorithm assigns a relationship between lateral length and production in the context of all the other features that we’re training the model on.
“It's not necessarily a one-to-one scaling of production with the lateral. And we’re not predicting a 12 month cum per foot and training the model on a per footage basis and then just multiplying it by the lateral length. It is a feature in the model. It’s also a feature in the time series model. As we’re predicting parameters to go from our 12 month cum value to a time series the lateral length is also in there. If there’s some nonlinear relationship between lateral length and EUR (Estimated Ultimate Recovery) or 12 month cum then the model is seeing that.”
“The technology has changed dramatically in the industry with faster drilling and completion methods,” Dr. Brendon Hall emphasizes. "There’s an important balance between length of the lateral, economic constraints, and lease boundaries. Although there is a trend towards longer laterals, it must be acknowledged that the per foot production from a longer well goes down.
“For Petro.ai DSUDS, we do know the lateral length for all of the wells that we put into the system and we know how much they produce with all the training data. What we do is look for relationships between the lateral length and the amount of oil and gas produced. That is a part of the modeling effort. We look at how the lateral length influences production, total production, EUR, the decline, all of those metrics. We develop a model to understand those relationships by teasing them out of the training data set that we have.
“Petro.ai understands this emerging shift towards the longer laterals and is working with our clients to address the data sparsity that can affect predicting around these newer, longer lateral lengths and the associated productivity. Even when we get to the edges of the data set we’re able to determine with confidence around the numbers and accuracy we provide.”