EZ Ops: Using machine learning to get the most out of oil and gas wells

Oil and gas companies employ many frontline operators to maintain their field operations every day. Asset management and maintenance can be time-consuming and expensive but are crucial to their production. EZ Ops, an Edmonton-based software company, is using machine learning to solve some of the most pressing challenges in the industry.

EZ Ops was founded by a group of oil and gas operators working to build solutions to manage those challenges in the field. Their aim is to create an advanced software-as-a-service (Saas)-based operations platform that draws on machine learning and artificial intelligence (AI) to help reduce operating expenses and emissions.

The platform is built around a central information hub for operation teams to share field data and communicate in real-time. It also helps them track and optimize production levels and fluid management, as well as automate scheduling and maintenance activities. The platform also aids in monitoring for leaks and equipment failure, while streamlining regulatory compliance reporting. These changes can have a big impact – to date, the company estimates it has saved its customers around $48 million in time saved and operating costs, while reducing carbon dioxide emissions by 4,880 tonnes.

The company is now using machine learning to tackle two key issues in the industry: task prioritization and cost/production prediction.

The challenge of dimensionality

When it comes to getting the most production out of a well, operators have a tricky task. With so much information to sort and so many systems to deal with, it can be difficult to determine what tasks should take priority – everything from running simple diagnostics to injecting chemicals or repairs. Traditionally, these choices have been made based on established routines and operator experience. However, this can be a time-consuming process. And as the industry becomes more data-driven and the amount of information that needs to be analyzed grows, the chance of error also rises. EZ Ops’ platform generates a list of high-priority tasks that provide more information to the operators and takes much of the guesswork out of the process.

EZ Ops is now using artificial intelligence to make that task prioritization even more efficient. Their machine learning model uses data recorded both by frontline operators working on the wells with information collected by the wells themselves plotted over time. By sifting through this data, the model finds trends and patterns that give a better understanding of what tasks to move to the top of the list.

However, using that data is a challenge in itself – partially because of the sheer amount of it. It is commonly said that ML models work better with more data. However, these methods can face some difficulties if the data has a very high dimensionality – meaning it is too detailed, with too many features recorded. This problem is so consequential that ML specialists have a name for it: “the curse of dimensionality.” If a machine learning (ML) model is presented with too many variables, it might not be able to tease out which factors are actually important and which ones are noise.

Fortunately, most problems come with some solutions in ML. EZ Ops turned to a method called Principal Component Analysis (PCA) to address the dimensionality problem.

PCA involves calculating the relationships between the many variables in a data set, to find which ones are closely related to one another. If several variables are highly correlated with another, they can be eliminated from the dataset without losing much information. The end result is a streamlined dataset can then be used to train a machine-learning model on the best actions a well operator can take. The model was trained using a hybrid reinforcement learning approach, starting with some base heuristics and then running a continuous feedback loop with real operators to tune the model.

Predicting the future

By using the model, EZ Ops’ platform is now even better at determining the most important tasks to prioritize, passing them along to well operators, allowing them to increase production. PCA is also being used to with another one of an operator’s key concerns: predicting production and possible events.

Using a statistical method called regression, EZ Ops is working on a model to analyze the historical data from its wells to determine to the relationship between one variable – say, how much oil a well will produce – to other factors. It will allow them to forecast the production rate or any other valuable information about the well. This model can used both by the company and the operators in so many ways. For example, by having a precise prediction of future production, the company can adjust its prices more efficiently. And on the operators’ side, they can have insight on upcoming events such as an increase in well pressure or the exhaustion rate of the well’s parts.

EZ Ops has estimated that they have saved up to $48 million dollars for their customers using frontline business intelligence. ML methods such as PCA and regression models have helped the company create AI-driven models that show operators the top priority activities to that yield the most production and safeguards compliance, allowing them unlock the potential in their databases to improve their frontline operations.

Artificial intelligence can help companies grow, improve and expand. Learn more about how Amii's team of researchers, product managers and experts can help on our Industry Solutions page

This case study was developed in collaboration with the AI Pathways Partnership, made possible with funding from Prairies Canada.