Obviousness

Obviousness Analysis of US Patent Application US20180240021A1 under 35 U.S.C. § 103

This analysis evaluates whether the invention claimed in U.S. Patent Application Publication No. US20180240021A1 (the '021 application) would have been obvious to a Person of Ordinary Skill in the Art (POSITA) at the time the invention was made. The analysis is based on the prior art references identified in the preceding section.

Under 35 U.S.C. § 103, a patent claim is invalid if the differences between the claimed invention and the prior art are such that the subject matter as a whole would have been obvious at the time the invention was made to a person having ordinary skill in the art.

Definition of a Person of Ordinary Skill in the Art (POSITA)

For the purposes of this analysis, a POSITA would be an individual with a degree in petroleum engineering, geological sciences, or a related field, coupled with several years of experience in reservoir engineering. This individual would be familiar with numerical reservoir simulation, conventional well planning techniques, and the application of statistical and computational modeling, including early-stage machine learning techniques, to analyze reservoir data and forecast production.

Analysis of Independent Claims (1, 8, and 15)

The independent claims of the '021 application cover a method, system, and data storage device for the same core process:

1. Receiving proposed parameters for a target well (location, configuration, production rates).
2. Forming a classification model by processing reservoir simulation results, with the model indicating fluid production rates, flows, and pressures.
3. Forming a probabilistic estimate of the target well's production rates using the classification model.
4. Making a decision to form (drill) the well if the estimate is acceptable.

The central inventive concept is the use of a machine learning-based "classification model," trained on simulation data, to generate a "probabilistic estimate" of a new well's performance to guide a drilling decision. The prior art, when combined, suggests this concept would have been obvious.

---

Combination 1: US 10,138,717 B1 ('717 patent) in view of US 9,043,188 B2 ('188 patent)

This combination of prior art would render the claims of the '021 application obvious.

• Primary Reference: US 10,138,717 B1 ('717 patent)
  The '717 patent discloses the core of the claimed method: predicting the performance of a new, un-drilled "target well" using a machine learning model. It teaches receiving the proposed well's features (location, completion, geology) and using a model based on "feature similarity" to find analogous existing wells. The performance of these analogous wells is then used to predict the target well's performance. This directly teaches steps 1, 3, and 4 of the '021 claims. The prediction based on a group of similar wells is inherently a "probabilistic estimate," and the entire purpose is to inform the decision of whether to drill the well.

• Secondary Reference: US 9,043,188 B2 ('188 patent)
  The primary element missing from the '717 patent is the explicit step of building the predictive model from a large set of computerized reservoir simulation results. The '717 patent builds its model from a database of existing wells. The '188 patent directly addresses this by teaching the creation of a "proxy model" (such as a neural network) by running a limited number of detailed, time-consuming reservoir simulations. This proxy model is then used to rapidly forecast production for many different scenarios.

• Motivation to Combine:
  A POSITA, familiar with the '717 patent's approach of using a data-driven model to predict target well performance, would naturally seek the best and most comprehensive data sources to train such a model. While the '717 patent uses historical data from existing wells, a POSITA would recognize the limitations of this approach—namely, that the historical data may not cover the full range of desired operational or geological scenarios for a new well.

  The '188 patent provides a well-known solution to this problem: using detailed numerical simulations to generate a rich dataset that can be used to train a faster "proxy model." A POSITA would have been motivated to combine these teachings for a predictable result. They would replace the historical well data used for training in the '717 patent with the synthetic, but more comprehensive, simulation data taught by the '188 patent. This would allow the "feature similarity" model of the '717 patent to be trained on a wider and more controlled set of data, leading to a more robust predictive tool. This straightforward combination of a known machine learning framework ('717) with a known method for generating training data ('188) arrives directly at the invention claimed in the '021 application.

---

Combination 2: US 9,043,188 B2 ('188 patent) in view of US 9,910,938 B2 ('938 patent)

This combination also provides a strong basis for an obviousness rejection.

• Primary Reference: US 9,043,188 B2 ('188 patent)
  The '188 patent teaches nearly the entire claimed invention. It describes a system that uses a "proxy model" (which can be a neural network, a type of classification model) built from reservoir simulations to forecast production. This covers receiving well parameters (as part of defining an "operational scenario") and forming a predictive model from simulation results to estimate production rates. The goal is to optimize a field development plan, which inherently includes making decisions about which wells to drill.

• Secondary Reference: US 9,910,938 B2 ('938 patent)
  The '188 patent describes its output as a "forecast." The '021 application specifically claims a "probabilistic estimate" and a "classification model." The '938 patent explicitly addresses this by teaching a method to generate a probabilistic forecast of production. It does this by running simulations on many geological "realizations" and then using statistical analysis and clustering to analyze the results. Clustering is a form of classification—it groups wells into categories of similar performance. This directly teaches the "classification" and "probabilistic" aspects of the '021 claims.

• Motivation to Combine:
  A POSITA starting with the proxy model concept from the '188 patent would understand that a single-point forecast has limitations and that a probabilistic output would be more valuable for risk assessment and decision-making. The '938 patent provides a known method for achieving this, teaching the use of statistical analysis and clustering on simulation results to create a probabilistic forecast.

  A POSITA would be motivated to apply the probabilistic and classification techniques from the '938 patent to the proxy model framework of the '188 patent. This would enhance the '188 system by allowing it to not just provide a single production forecast, but to classify the likely outcome of a proposed well (e.g., "good" or "bad," as described in the '021 specification) and provide a probabilistic confidence level. This is a simple application of a known data analysis technique ('938) to a known modeling system ('188) to achieve the predictable improvement of a more robust, risk-aware output. The result of this combination is precisely the system claimed in the '021 application.

Conclusion

The independent claims of US20180240021A1 appear to be obvious over at least two different combinations of the cited prior art. The core concept of using a computer-generated model, trained on reservoir simulation data, to produce a probabilistic estimate of a target well's performance was a known and evolving practice in the field before the '021 application's priority date. The prior art teaches the use of machine learning models for this purpose, the generation of training data from simulations, and the framing of the output in probabilistic and classificatory terms. Combining these known elements to create the claimed system would have been a predictable and logical step for a Person of Ordinary Skill in the Art seeking to improve the efficiency and accuracy of well planning.