It all started as an accident. We passed early time production to a “pre-drill” Novi model. When we saw the results, we were stunned — they were that accurate. Machine learning for PDP oil and gas forecasting was the next frontier.
What we had discovered is that existing methods were leaving a whole lot of valuable information on the table — the geology, the completions, the spacing. If you could leverage that data along with the existing production, you could see around corners: generating better early-time forecasts, handling noisy production data, anticipating evolving reservoir conditions. And all of that with a method that is faster and less biased than hand-fitting curves.
Now that our patent is pending, we are excited to discuss our PDP oil and gas forecasting method, which is already in use by some of the top operators in unconventionals.
In this Post
Issues with Traditional PDP Oil & Gas Forecasting
Traditional methods for forecasting production are slow, inaccurate, and subject to bias. Hand-fitting decline curves takes valuable reservoir engineering time that could be better spent on actual engineering projects. And with another year of layoffs and reorgs in the books, nearly everyone is short-staffed. Something has to break.
Furthermore, traditional DCA is subject to bias. We don’t blame engineers — they’re human, just like us! But with the wrong tools and wrong direction from management, they can end up producing forecasts that wildly miss actual production.
Forecast inaccuracy is driven by a number of factors:
- Difficulty handling early-time forecasts before a clear trend has been established
- Trouble fitting curves to noisy production data
- Complexity of unconventional reservoirs (e.g., complex matrix-fracture flow dynamics, nonuniform depletion)
- Parent-child relationships and inter-well interference
- Sub-optimal decline curve methods adapted from conventionals
- Management or cultural influence to maximize forecasts… or never miss them
- Individual engineer bias
But, not to worry: the Novi method offers solutions to many of these problems.
Novi PDP Oil & Gas Forecast Methodology
Novi uses a PDP machine learning-based autoregressive forecasting methodology that incorporates more than just production data.
- Machine Learning-Based means it uses actual well behavior to generate its forecasts, rather than an assumptions-laden parametric equation (hyperbolic, exponential, etc.)
- Autoregressive means the model uses data from previous time steps for its next-month forecast, uses that for the following, etc. Autoregressive modeling is a powerful technique for time series forecasting used in other disciplines.
- Incorporates more than just production data means we use geology, completions, spacing, well operations, AND prior production to make our forecasts.
Benefits of Novi Approach: Increased Accuracy, Decreased Bias, Better Confidence
Novi’s PDP forecasting approach offers several principal benefits:
- Improved accuracy: we will often see accuracy 2x improved compared to operator engineer forecasts.
- Decreased bias: because our forecasts are grounded in actuals, we will usually be within the low single digits on average — which means you’re less likely to miss guidance.
- Better confidence: our machine learning algorithms can accurately estimate the uncertainty in any forecast — more on this in a later post!
Let’s take a look at several example wells where curve fitting would struggle but including geology, completions, and spacing to identify similar wells helps:
By now you must be eager to see how this works in our oil and gas software. But first, we have one last topic to cover: estimating EURs in PDP oil and gas forecasting. Of course, machine learning predictions are data-driven, and the oldest unconventional plays are on the order of 10 years old. Right now, we can extend our forecasts out to eight years, but going beyond that still requires making assumptions. Just like with our pre-drill forecasts, we let you configure how that’s done:
Finally, let’s take a look at generating PDP forecasts in our software:
Novi algorithms provide several benefits over traditional forecasting methods:
- Speed – so you can save valuable engineering time
- Improved accuracy – make sure you know what you will be producing!
- Minimized bias – reduce human-driven error to the high- or low-side
- Part of a full data & forecasting platform – to integrate the entire forecasting process
Our next posts will cover The Benefits of Automation and PDP Risk & Uncertainty. Check back soon!
Click here to read more about our PDP product offering here: PDP in Oil and Gas Forecasting Press Release. To learn more, reach out below!