Machine Learning/AI in Oil & Gas
This guide provides information about how machine learning works and how It helps oil and gas companies & Investors assess the value of specific reservoirs, customize drilling and completion plans according to the geology of the area, and assess risks of each individual well.
Machine Learning in Oil and Gas
Machine learning (ML) focuses on the use and development of computer systems that learn from data and past experiences while identifying patterns to make predictions with minimal human intervention.
Why is machine learning so important?
Machine learning provides businesses with more insights and new approaches to customers based on their behaviors and operational patterns. It can be used in a big range of applications such as speech recognition, face recognition, image recognition, email filtering, computer vision, and medicine.
For many of today’s leading companies, machine learning has emerged as a key competitive differentiation, such as Facebook, Google, Uber, Apple, Netflix, and Amazon.
How has machine learning evolved?
Machine learning algorithms have been around for a long time, but their use has increased as artificial intelligence has become more prevalent. Modern AI applications are primarily powered by deep learning models.
Machine learning platforms are among the most competitive areas of enterprise technology, with the majority of big companies vying for customers by offering platform services that include data collection, data preparation, data classification, process data, data analytics, predictive analytics, model building, training, and application deployment.
What is the difference between AI and machine learning?
Although the terms artificial intelligence (AI) and machine learning are frequently used synonymously, machine learning is actually a subset of AI.
In this context, machine learning refers to the technologies and algorithms that enable systems to identify patterns, make decisions, and improve themselves based on experience and data, whereas artificial intelligence refers to the general ability of computers to emulate human thought and perform tasks in real-world environments.
Computer programmers and software developers use machine learning to solve problems in domains such as computer vision, natural language processing and time series forecasting with tools like linear regression, random-forest, and neural networks.
What is the difference between AI and machine learning?
The result of a machine learning algorithm applied to data is referred to as a “model”. The model can be operated as a program that uses the previously saved functionality of the algorithm to generate new predictions.
Algorithms come in a wide variety of forms with a wide range of uses. For instance, we have algorithms for:
- Classification: refers to the algorithms used to differentiate between categorical values, such as k-nearest neighbors
- Regression: used to create predictions when the output is a continuous value, such as linear regression
- Clustering: is a technique for gathering related objects or clustered data points, such as k-means.
A model is a representation of what a machine learning algorithm has learned. For example:
- The model produced by the linear regression algorithm is made up of a vector of coefficients applied to the training features.
- The model produced by the decision tree algorithm is made up of a tree of if-then statements with specific values.
- The model produced by a neural network is composed of layers made up of nodes and functions.
What are the different types of machine learning?
A common method to categorize conventional machine learning is the process by which a prediction-making algorithm learns to increase its accuracy. Reinforcement learning, semi-supervised learning, unsupervised learning, and supervised learning are the four main approaches. The kind of algorithm that data scientists use depends on the type of data that they want to analyze.
Supervised learning
Supervised learning is when the model is getting trained on a labeled dataset. This machine learning approach includes the inputs and outputs created by supervised learning techniques. The information is a collection of training examples and is referred to as training data.
Through repeated optimization of an objective function, supervised learning techniques create a function that may be used to predict the output linked to new inputs. An algorithm is considered to have learned to do a task when, over time, the accuracy of its outputs or predictions increases. There are two main categories of supervised learning algorithms
- Classification
- Regression
Novi’s production forecasting models are all regression models in the supervised learning category.
Unsupervised learning
Unsupervised learning algorithms analyze a collection of input-only data to identify patterns, such as grouping or clustering of data points. This approach trains on unlabeled data and locates commonalities in the data and acts based on the existence or absence of such commonalities in each new piece of data, as opposed to reacting to feedback. The model has to learn by itself.
Unsupervised learning algorithms are usually used for clustering, anomaly detection, association mining and dimensionality reduction. Unsupervised learning algorithms are usually used for clustering, anomaly detection, association mining and dimensionality reduction.
Semi-supervised learning
This application is a mixture of supervised learning and unsupervised learning. Data scientists input a limited quantity of labeled training data into an algorithm to perform semi-supervised learning. The algorithm gains knowledge of the data set and is allowed to explore the data on its own to come to its own conclusions.
Semi-supervised learning is applied in a number of fields, such as:
- Machine translation: Teaching algorithms to translate language based on less than a full dictionary of words.
- Fraud detection: Finding instances of fraud when there aren’t many successful examples.
- Labeling data: Algorithms can learn to automatically apply data labels to larger sets after being trained on small data sets.
Reinforcement learning
Reinforcement learning is a learning method used by data scientists to teach machines how to complete a task in a possibly complicated and uncertain environment. Artificial intelligence overlooks a game-like situation, and, in order to solve the problem, the computer uses trial and error. Artificial intelligence is rewarded or punished for the steps it takes to make the machine do what the programmer desires. Its goal is to maximize the total reward.
Reinforcement learning algorithms are used in autonomous vehicles or when trying to learn how to play a game against a human opponent. Reinforcement learning is frequently used in fields like:
- Robotics: Using this method, robots can learn to carry out tasks in the real world.
- Video game play: A variety of video games on how to play a game against a human opponent.
- Resource management: reinforcement learning can assist organizations to plan out how to allocate resources.
Want to learn why machine learning is so important in the oil and gas industry and where it can be applied?
Why use of machine learning in the oil & gas Industry?
Across oil exploration, production, transportation, and refinement, the oil and gas sector has long been at the cutting edge of technology, pioneering incredible feats of engineering. In recent years, the oil and gas industry has been catching up to other sectors in the machine learning space, thanks in part to better data handling and processing.
Low commodity prices and negative investor sentiment have also pushed operators to modernize operations. Machine learning enables the industry to learn from the vast amounts of data generated during oil and gas operations by improving operational efficiency and decision-making. Machine learning helps oil and gas businesses save time, reduce risk, and improve return.
But how exactly does machine learning benefit the oil and gas companies?
Of course, it varies depending on the workflow.
At the far upstream end, machine learning can speed up well log or seismic interpretation. In this area, geoscientists work with reinforcement learning algorithms or supervised learning algorithms, providing stratigraphic picks that are then propagated around a dataset to generate widespread interpretations, quickly. This can save a geologist hundreds of hours of work. Less time, fewer errors, and consistent outcomes translate into lower expenses. Unsupervised algorithms can also be used to classify log or seismic facies, potentially helping identify previously unrecognized rock groups.
Within the unconventional oil and gas space, many basins now have tens of thousands of producing wells, a treasure trove of data to analyze with statistical algorithms. Across these datasets, operators have tried many different combinations of completions designs and well spacing configurations, implemented across many different geologic settings. Supervised machine learning models can use these data types (completions, geology, spacing) as the training variables, with the “labels” being production data (what the model is trying to predict).
From all this data, the models learn how the input variables impact the well production. After the model is trained, it can be used to forecast hypothetical designs, such as forecasting production at proppant intensities of 2,000, 2250, 2500, 2750, and 3000 lbs/ft. Combining that with a cost model and price deck will give you a data-driven optimization on proppant intensity. This type of use case is core at Novi Labs, where our machine learning platform generates these forecasts using mostly decision-tree based methods. Because humans struggle to analyze the data across many dimensions, the machine learning models can provide a more accurate answer, potentially unlocking hundreds of thousands or millions of dollars of value at each well location by improving production or saving costs.
Other field development use cases include identifying the optimum number of wells, locations, or sequence of drilling. Machine learning can help identify geologic sweet spots and determine the profitability of each zone at a given location. These models also improve baseline forecast accuracy over type curve methods, because they minimize bias and effectively analyze well performance across many dimensions.
Machine learning in oil and gas is also used for production engineering and midstream applications. One promising technology in the space is virtual flow metering, which estimates flow rates based on pressure, temperature, and choke data.
Midstream companies can use machine learning for leak detection, preventative maintenance, and transportation optimization.
Novi products turn raw data into scenario-driven well forecasts for oil well drilling optimization
Analytics-Ready Data
Get to better answers, faster.
Oil & Gas Focused AI
Algorithms tuned for your business problems.
Efficiency & Automation
Teams using data-driven approaches outperform traditional methods.
Trusted by some of the world's top oil & gas and financial institutions
Want to discover why E&P operators are turning to Novi Labs to gain a competitive edge?
We provide the most accurate, timely, and complete upstream data, driven by Novi’s industry-leading data science expertise.
Machine Learning in the Oil and Gas Industry
earn from seasoned, experienced industry veterans as they go into technical topics related to machine learning, artificial intelligence, risk, and digital oilfield.
- by Dillon Niederhut
- by Kiran Sathaye
- by Kiran Sathaye