Best practices are not static; they evolve alongside advancements that redefine what is achievable.
In constructing type well profiles (TWP) for oil and gas production, the traditional approach has relied on decline curve parameters to create a "best fit" for average historical production. While this method served the industry well, it inherently simplifies the nonlinear behavior of unconventional reservoirs and creates limitations (SPE 158867).
With advancements in the use of artificial intelligence (AI) models to predict future production, this best-fit simplification is no longer necessary. As the industry shifts to tighter well spacing and the development of Tier 1a and Tier 2 rock, leveraging more data to create production forecasts becomes increasingly critical.
AI-based forecasting, coupled with percentile-based methods, such as P10/P50/P90 analysis and distribution plots, enables a more comprehensive view of production variability and uncertainty. Together, AI models and probabilistic outputs provide a reliable alternative workflow to traditional best-fit decline curve approaches for type well construction.
Traditional Curve Fitting and Its Limitations in Unconventional Reservoirs
Historically, TWPs have been constructed by aggregating production histories from relevant analog wells and creating a decline curve that best fits the combined production data. This curve, expressed through Arps or multi-segment Arps formulas, is based on a physical description of the reservoir and represents the final output used in economic evaluations and forecasts. Industry norms have long relied on this workflow, which provides a streamlined and standardized approach to forecasting.
However, the accuracy of this best fit can be subjective. Decline curve analysis (DCA) inputs of initial production point, and b-factor and decline percent, both which change several times over the early life of the well, can be estimated as very different values based on the interpretation of the historical and analogue data.
Percentile-Based Analysis: A Better Alternative
One evolution of best practices has been the use of percentile-based approaches, previously not traditionally embedded in unconventional workflows due to computational demands. Percentile-based approaches, such as the Modified Time Slice method (SPEE Monograph 5), address some of the limitations of traditional DCA by incorporating probabilities associated with higher or lower production outcomes.
Tangible Impacts of Using AI To Improve TWP Development
A few of the concrete advantages of using AI models to do forecasting include:
- Accurate Short-Term Forecasting—DCA struggles in the first 1 to 24 months when production is highly nonlinear and b-values are changing, obscuring critical variations during the most volatile phase of a well's life.
- Reproducible Answers—Traditional curve fitting introduces variability based on who is performing the analysis. Pure data-driven forecasts ensure consistent, repeatable results.
- Better Analog Well Selection—AI systematically analyzes vast datasets using statistical methods, offering a more objective and consistently applied approach to identifying relevant analogs.
- Full Probabilistic Forecasting—Unlike DCA's single best fit curve, AI models generate forecasts across a full probability distribution, explicitly quantifying uncertainty akin to Monte Carlo simulations.
- Adaptive Learning through Periodic Updates—AI models are periodically updated with new operational data and benefit from advancements in AI technology.
Note that AI-driven forecasting does not replace the expertise of engineers and asset teams. Instead, it serves as a decision-support tool, enhancing traditional workflows by providing a more data driven and probabilistic view of well performance.