Smart Predictive Analytics to Forecast Unconventional Well Performance More Accurately
Efforts to optimize unconventional play development face a big hurdle. Standard modeling and simulation tools offer only ballpark estimates, since they empirically assess future performance based on past production history. That makes it tough to optimize unconventional resource development. And that's why we turned to data science to improve production forecasting for our wells in the United States using existing data.
The gamble has paid off with the Smart Predictive Analytics (SPA) application developed by our teams, which can predict actual production with 90% accuracy, a major competitive edge for our developments.
Best Innovators 2017 - SPA: Smart Predictive Analytics
Why We Need to Forecast Well Performance More Accurately
The shale gas revolution in the United States has involved intensive drilling and massive fracturing campaigns. Hundreds of wells have been drilled, sharply boosting domestic production, but only 20% were profitable. This wasted enormous amounts of capital while also having a strong environmental impact.
The downturn in oil and gas prices and our ROACE goals have made it critical to be more selective about investment decisions. And therefore to identify the best wells to drill and predict their fields' future production more accurately.
The standard methods used to predict unconventional well performance (type curves) give a decent ranking of the zones to target, but fail to identify the best wells among the hundreds of possible locations. This led to the idea of testing a data analytics method to improve predictions for Utica, a shale gas play in Ohio in which we hold a 25% interest.
A Multidisciplinary Team Success
SPA's workflow drew on a multidisciplinary understanding of the challenge and two strong areas of expertise for Total.
In-Depth Information Processing and Analytics Capabilities
Unconventional developments are data-rich by nature. A massive amount of information from Utica's 600 producing wells and 12,000 hydraulic fractures was available. We consolidated our multiple sources — subsurface syntheses, a well completion database and continuously updated production data on the play — into a single database.
High-Power Computing Platforms
We then used machine learning algorithms to predict gas, condensate and water production, as well as the gas's composition. Integrating this workflow into our high-power computing resources generates thousands of profiles of future locations in less than a minute, compared to months using standard methods.
A Major Competitive Advantage
After cross-validating the result on existing Utica wells, the SPA application enables us to predict a field's actual production with 90% accuracy. That compares to just 50% accuracy for type curves. Plus, our massive computing power will enable us to assess even more complex future applications or integrate even more data.
The application, developed initially in our U.S. operations, is in the process of being patented. We expect to expand it to other affiliates in 2018, such as in China for the Sulige gas field, after scale-up.
Smart Predictive Analytics (SPA) analyzes millions of field data points in an instant to learn typical well behavior and forecast future reservoir decline.
This innovation allows forecasting with great accuracy the future performance of thousands of wells in a matter of minutes by using Predictive Analytics through High Performance Computing.
Comparison of actual production (Y) versus pre-drill estimates of production using traditional methods (in Y) – Utica example. CGR stands for Condensate-Gas-Ratio
Crossplot comparing actual production versus pre-drill estimates of production using traditional forecast (in blue) or predictive analytics (in black) – Utica example.