DETAILED ACTION
The following Non-Final Office Action is in response to Applicant communication dated 04/27/2026.
Status of Claims
Claims 1, 9, 12, 15 and 17 have been amended. Claims 1-20 are currently pending and have been rejected as follows.
Response to Amendments
The original 35 USC § 112(b) rejection is withdrawn based on the amendments that remedied the issues. A new 35 USC § 112(b) rejection is issued. The 35 USC § 101 rejection is maintained for claims 1 and 3-20. The 35 USC § 102(a)(1) rejection of claims 1-20 is withdrawn, and a new prior art rejection is made under 35 USC § 103.
Response to Arguments: 35 USC § 101
Applicant argues [p.2] that the subject matter of claim 1 cannot be performed in the human mind specifically citing to the “reconstructing” limitation arguing that the human mind is not capable of containing or executing a machine learning model to reconstruct log data.
This argument is not persuasive given that the “reconstructing” process can simply involve filling in missing log data when log data from wells are missing, incomplete or of bad quality [0037] according to a predefined model representing historically measured data. This claim limitation at least recites a mathematical concept in the form of a mathematical calculation given that the reconstructing process is based on a trained machine learning model which can be an ensemble-based regression model or an artificial neural network model [0037] (see Recentive Analytics, Inc. v. Fox Corp., Case No. 2023-2437 (Fed. Cir. Apr. 18, 2025) wherein the mere use of machine learning to complete a specific task does not automatically quality the claim as non-abstract under Step 2A Prong 1, stating “the claimed methods are not rendered patent eligible by the fact that (using existing machine learning technology) they perform a task previously undertaken by humans with greater speed and efficiency than could previously be achieved” (id. p. 15)).. Furthermore, MPEP 2106.04 (a)(2) states that a step such as “determining” or “performing” may still be considered a mathematical calculation given the broadest reasonable interpretation of the claim in light of the specification, even if the claim does not specifically recite “calculating” or a mathematical operation.
Applicant similarly argues [p.2] that the “calibrating” step is not directed to a mathematical concept.
This argument is not persuasive given that “calibrating” involves determining an effective change point detection model architecture, cost function, and hyper parameters (e.g., number of break points, minimum distance between change points, etc.) [0034]). The claim limitation thus recites a mathematical concept in the form of mathematical calculations for similar reasons as stated above. See the rejection for more details.
Regarding Applicant’s arguments [p. 2-3] pertaining to claim 1 not being directed to a mathematical concept per se and on its own based on USPTO Examples 39 and 47, while claim 1 does not refer to mathematical calculations by name and does not explicitly recite the mathematical formulas, symbols, calculations etc., given the broadest reasonable interpretation of the claim in light of the specification, MPEP 2106.04 (a)(2)(I) states that a step such as “determining” or “performing” may still be considered a mathematical calculation, even if the claim does not specifically recite “calculating” or a mathematical operation. In this case, claim 1 does recite a mathematical concept as explained for the argued limitations above. See the rejection for further details.
Applicant argues [p.3] that even if claim 1 were directed to an abstract idea, claim 1 provides a specific technological improvement because the features of claim 1 can determine formation boundaries for wells having missing or incomplete well logs by reconstructing the missing data using a machine learning model. By using a calibrated model to detect formation boundaries and using a machine learning model to reconstruct missing log data, the method of claim 1 can determine formation boundaries for target wells in near real- time. The method of claim 1 can reduce processing time for detecting formation boundaries by a factor of ten.
This argument is not persuasive because merely applying mathematical concepts to analyze information does not itself constitute an improvement of technology. Rather, the additional elements must meaningfully limit the judicial exception by improving how the technology operates.
In this case, the machine learning model is being used as a tool, which is not a technological improvement. The claim merely recites a generic machine learning model to produce reconstructed data. MPEP 2106.05(a) explains that an improvement exists where the claims recite how technology is improved and not merely that an abstract process is performed using conventional computing techniques.
In terms of “reducing processing time by a factor of ten”, MPEP 2106.05(f) has made it clear that “claiming the improved speed or efficiency inherent with applying the abstract idea on a computer" does not integrate a judicial exception into a practical application or provide an inventive concept. Therefore, this statement does not demonstrate a technological improvement.
As for the “reconstruction” of the logs, the improvement is to the quality of information and not the technology itself. Therefore, the improvement resides in the information derived from the data rather than in the operation of the technological system - an improvement to the abstract ideas of calibrating models, reconstructing log data, or determining log similarities does not amount to an improvement to technology or a technical field (see MPEP § 2106.05(a)(III) stating “it is important to keep in mind that an improvement in the abstract idea itself (e.g. a recited fundamental economic concept) is not an improvement in technology. For example, in Trading Technologies Int’l v. IBG, 921 F.3d 1084, 1093-94, 2019 USPQ2d 138290 (Fed. Cir. 2019), the court determined that the claimed user interface simply provided a trader with more information to facilitate market trades, which improved the business process of market trading but did not improve computers or technology.”).
Regarding “determining formation boundaries for target wells in near real- time”, the claim limitation recites the desired result or the functional objective rather than a particular technical solution that achieves those results. The use of the term “in real-time” does not disqualify the limitation from being categorized as an abstract idea. Language such as concurrently, automatically, instantly, or simultaneously to describe the automation of a manual process is not enough to overcome a subject matter eligibility rejection (MPEP § 2106.05(a)(I) Examples that the courts have indicated may not be sufficient to show an improvement in computer-functionality no. (iii) mere automation of manual processes). Examiner also notes that language such as this is not restricted to computer processes, humans can automatically/instantly/simultaneously/in real-time complete different tasks (see MPEP § 2106.04(a)(2)(III) stating that the mental processes may be completely by humans plural – not just a singular human mind).
For these reasons, the argued claim limitations and claim 1 as a whole does not recite a specific technological improvement.
Response to Arguments: 35 USC § 102(a)(1)
Applicant argues [p.4] that Akkurt (WO 2018208634 A1) does not describe or suggest “calibrating a model to detect formation boundaries” and “determining depths of formation boundaries based on the model” because a person of ordinary skill in the art would understand that decision boundaries for an SVM are not formation boundaries in a subsurface formation.
Applicant’s arguments are persuasive, and a new prior art rejection is made under 35 USC § 103.
Claim Rejections - 35 USC § 112(b)
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-20 are rejected under 35 U.S.C. 112(b) or pre-AIA 35 U.S.C. 112, second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or, for applications subject to pre-AIA 35 U.S.C. 112, the applicant) regards as the invention.
Claim 1, line 4 recites the limitation "based on the log data” in the calibrating step, and there is insufficient antecedent basis for this limitation in the claim because, in line 3, the claim recites “receiving first log data”, which constitutes two different types of data. For example, is “the log data” referring to “first log data” or another type of data entirely? For these reasons, claim 1 and its dependents are rejected under 35 U.S.C. 112(b). Independent claims 12 and 17 recite similar limitations and therefore claims 12-20 are rejected for the same reasons.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1 and 3-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception without significantly more. A subject matter eligibility analysis is set forth below. See MPEP 2106.
Representative Claim 1 recites:
A method for determining formation boundaries in a subsurface formation, the method comprising:
receiving first log data from a set of reference wells in a subsurface formation;
calibrating a first model to detect formation boundaries, the calibrating being based on the log data from the set of reference wells;
receiving second log data from a set of target wells in the subsurface formation, the set of target wells being different than the set of reference wells;
reconstructing the second log data from the set of target wells based on a machine learning model, the machine learning model being trained on the log data from the set of reference wells;
determining depths of formation boundaries based on the first model;
determining similarities between log data in intervals defined by the determined depths of formation boundaries from two or more wells, the two or more wells selected from the set of reference wells, the set of target wells, or both;
correlating formation boundaries between the two or more wells in the subsurface formation based on the determined similarities;
and generating a visual representation of the depth of formation boundaries in the subsurface formation based on the correlated formation boundaries.
The claim limitations in the abstract idea have been highlighted in bold above. They constitute a mental process and/or mathematical calculation, thus qualifies those elements as abstract ideas. The remaining limitations are “additional elements.”
Under Step 1 of the analysis, claim 1 does belong to a statutory category, namely it is a process claim. Likewise, claim 12 is an apparatus claim, and claim 17 is a machine claim.
Step 2A, Prong One: This part of the eligibility analysis evaluates whether the claim recites a judicial exception. As explained in MPEP 2106.04, subsection II, a claim “recites” a judicial exception when the judicial exception is “set forth” or “described” in the claim., Under Step 2A, Prong One, the broadest reasonable interpretation of the steps recited in Claim 1 include at least one judicial exception, that being a mathematical concept and/or mental process.
This can be seen in the claimed process steps of “calibrating a model to detect formation boundaries, the calibrating being based on the log data from the set of reference wells” (FIG. 2, [0027] of the instant specification), “reconstructing log data from the set of target wells based on a machine learning model, the machine learning model being trained on the log data from the set of reference wells” (FIG. 2 ( blocks 220, 221), [0037] of the instant specification), “determining depths of formation boundaries based on the model” (FIG. 3 (step 310), [0046] of the instant specification), “determining similarities between log data in intervals defined by the determined depths of formation boundaries from two or more wells of the set of reference wells and the set of target wells” (FIG. 3 (step 312), [0047] of the instant specification), “correlating formation boundaries between the two or more wells in the subsurface formation based on the determined similarities” (FIG. 3 (step 314), [0048] of the instant specification), each of which encompasses mathematical concepts requiring specific mathematical calculations and/or mental steps.
For example, (See paras. [0034-0035] of the instant specification) where “A change point detection model can determine break points in series data…The change point detection model can have several architectures depending on the how the architecture performs on a given set of data…a dynamic programming change point detection model, a change point detection model with linear computation cost, a multiple change-point detection model with a reproducing kernel, a binary segmentation change point detection model, a bottom-up segmentation change point detection model, or a sliding window change point detection algorithm. Calibration of the change point detection model can include determining an effective change point detection model architecture, cost function, and hyper parameters (e.g., number of break points, minimum distance between change points, etc.)…an experiment can include the data processing system executing a change point detection model having a specified cost function 215, architecture or search method 216, set of well log data 217, and values of hyper parameters 218 with the input to the change point detection model being the log data from the reference wells 204…a metric can be based on an average distance between predicted depths and measured depths of formation boundaries in the reference wells. Another example metric can be based on a difference between a number of known formation boundaries and a number of predicted formation boundaries. Another example metric is given by:”
PNG
media_image1.png
202
307
media_image1.png
Greyscale
to perform the process of determining the depths of formation boundaries based on the collected log data, and therefore encompasses mathematical concepts and/or mental steps. For example, when given the broadest reasonable interpretation in light of the specification, the steps of “receiving,” “calibrating,” “reconstructing,” “determining,” and “correlating” are performed using one or more training algorithms (machine learning model(s)) and/or mental steps. Claims 12 and 17 recite analogous judicial exceptions.
Step 2A, prong 2 of the eligibility analysis evaluates whether the claim as a whole integrates the recited judicial exception(s) into a practical application of the exception. This evaluation is performed by (a) identifying whether there are any additional elements recited in the claim beyond the judicial exception, and (b) evaluating those additional elements individually and in combination to determine whether the claim as a whole integrates the exception into a practical application.
In addition to the abstract ideas recited in claims 1, 12 and 17, the claimed method recites additional elements including “receiving log data from a set of reference wells in a subsurface formation” and “receiving log data from a set of target wells in the subsurface formation” and “generating a visual representation of the depth of formation boundaries in the subsurface formation based on the correlated formation boundaries,” however these elements are found to be data gathering and output steps, which are recited at a high level of generality, and thus merely amount to “insignificant extra-solution” activity(ies). See MPEP 2106.05(g) “Insignificant Extra-Solution Activity.” Machine claims 12 and 17 recite analogous additional elements.
The generic data gathering, processing, and output steps, are recited at such a high level of generality (e.g. using “at least one processor; and a memory storing instructions” and “one or more non-transitory machine-readable storage devices storing instructions”) that it represents no more than mere instructions to apply the judicial exceptions on a computer. It can also be viewed as nothing more than an attempt to generally link the use of the judicial exceptions to the technological environment of a computer. Noting MPEP 2106.04(d)(I): “It is notable that mere physicality or tangibility of an additional element or elements is not a relevant consideration in Step 2A Prong Two. As the Supreme Court explained in Alice Corp., mere physical or tangible implementation of an exception does not guarantee eligibility. Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 573 U.S. 208, 224, 110 USPQ2d 1976, 1983-84 (2014) ("The fact that a computer ‘necessarily exist[s] in the physical, rather than purely conceptual, realm,’ is beside the point")”.
Under Step 2B, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements, as described above with respect to Step 2A Prong 2, merely amount to a general purpose computer system that attempts to apply the abstract idea in a technological environment, limiting the abstract idea to a particular field of use, and/or merely performs insignificant extra-solution activit(ies) (claims 1, 12 and 17). Such insignificant extra-solution activity, e.g. data gathering and output, when re-evaluated under Step 2B is further found to be well-understood, routine, and conventional as evidenced by MPEP 2106.05(d)(II) (describing conventional activities that include transmitting and receiving data over a network, electronic recordkeeping, displaying an output, and storing and retrieving information from memory.)
Therefore, similarly the combination and arrangement of the above identified additional elements when analyzed under Step 2B also fails to necessitate a conclusion that claim 1, as well as claims 12 and 17, amount to significantly more than the abstract idea.
Regarding dependent Claims 3-5, 9, 11, 13, 15, and 18, these claims merely add more abstract idea limitations. Therefore, the recited judicial elements in the claim limitations do not have further additional elements that would integrate the recited subject matter into a particular practical application. Therefore, these claims are rejected under 35 U.S.C. 101.
Regarding dependent Claims 6-8, 10, 14, 16, 19 and 20, these claims merely recite further detailed use of the machine learning model. However, the details remain under the category of generic computer processing unit. Although the claims (e.g. claim 6) recite a specific type of generic ML model, e.g. ensemble-based regression or ANN, this merely attempts to limit the abstract idea to a particular field of use and/or is still found to be the use of generic computer technology as a tool to apply the abstract idea. The other mentions of “training” in the claims, e.g. claim 7, merely specify the type of data parameters used and does not specify any particular details on how the model itself is being trained. The claims merely characterize the data inputs and outputs, e.g. claim 10, without describing any details of the functioning of the ML model itself. Therefore, the recited judicial elements in the claim limitations do not have further additional elements that would integrate the recited subject matter into a particular practical application for the reasons mentioned in Step 2A/2B above. Therefore, these claims are rejected under 35 U.S.C. 101.
Regarding Claim 2, “drilling a well at the determined location” is an additional element which is sufficient to integrate the claim into a particular practical application since it reflects a real-world transformation. Therefore, Claim 2 is not rejected under 35 U.S.C. 101.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Akkurt International Publication WO 2018/208634 A1 (hereinafter “Akkurt”) in view of Al-Garni (US 20220066064 A1) further in view of Gillan (US 20180266245 A1).
Regarding representative claim 1, Akkurt teaches a method for making determinations in a subsurface formation, the method comprising: [Akkurt: Abstract]; receiving first log data from a set of reference wells in a subsurface formation; calibrating a first model to detect, the calibrating being based on the log data from the set of reference wells; receiving second log data from a set of target wells in the subsurface formation, the set of target wells being different than the set of reference wells; (Akkurt, paras. [0143], [0096], [0162], respectively; [0143]: [“Given an ML model, the user then uses it to predict the formation properties, for the test wells, as shown in the bottom row. Unlike the training-wells, test-wells do not have the ground truth. Given the ML model (bottom row, middle) 310, predictor data from any test- well (bottom row, left) 308 is fed into the model, and formation properties (Sw and PHIT in this case) are predicted (bottom row, right) 312.”]; [0096]: [“A candidate well is defined as one that has the appropriate input and ground data that can be used in building an ML model.”]; [0162]: [A model may be built using a given a number of candidate training wells. The candidate wells have both the input data to be used in the prediction (e.g., MGL+DD+GR) and the ground truth to be used in the training (Sw, PHIT). The candidate wells may also have the WL or LWD logs that are used in the determination of the ground truth.”]); reconstructing the second log data from the set of target wells based on a machine learning model, the machine learning model being trained on the log data from the set of reference wells; determining depths based on the first model; (Akkurt, paras. [0108], [0110], [0243], [0053], [0212, steps 1-2 and 6], respectively; [0108]: [“Consider a case where one of the logs on the WL run is bad, for example, the density log. Either the problem is discovered too late to repeat the measurement, or a rerun is not considered for operational reasons. A replacement "density" log can be created in a number of ways: (i) using from MGL+DD alone from adjacent wells, (ii) using WL or LWD logs from adjacent wells, (iii) using a combination of (i) and (ii). The caveat in the third case is that invasion physics may be taken into account when combining data acquired at different times during the drilling of a well.”]; [0110]: [“Another variation is a well where there is no log data, due to well collapse, stuck pipe, instability, etc. Replacement logs can be computed from MGL+DD, as they would be acquired as soon as the bit penetrated the formation.”]; [0243]: [“Jaccard similarity is not sensitive to situations where the footprint of well A is a subset of well B. In these situations, Jaccard similarity will be less than one, even if the footprint of well A is completely contained within the footprint of well B. A user may identify these situations because, in this case, well B would be a strong candidate to build a predictive model to reconstruct logs in well A. Overlap similarity provides a way to identify such overlaps.”]; [0053]: [“The systems and methods disclosed herein may predict formation properties that are normally interpreted or measured directly, using Machine Learning (ML) Algorithms. The systems and methods disclosed herein use of Mud Gas Logs (MGL) and Drilling Data (DD), rather than Wireline (WL) or Logging-While-Drilling (LWD) logs, in the prediction of formation properties such as water saturation or total porosity.”]; [0212, steps 1-2 and 6]: [“1. The user selects a well on which to predict the response variable of interest. 2. A pre-existing Quantile Regression Forest (QRF) model is selected from a library or a new QRF model is created from a training dataset. 6. For each measured depth sample in the well of interest”]; determining similarities between log data in intervals defined by the determined depths from two or more wells, the two or more wells selected from the set of reference wells, the set of target wells, or both; correlating between the two or more wells in the subsurface formation based on the determined similarities; (Akkurt, Claims: 13-14; paras. [0232], [0219], respectively; [Claim 13]: [“The computing system of claim 9, wherein the wells are sorted into the groups using a petrophysical similarity analysis that includes determining a similarity matrix using the well log data, the flag, or both.”]; [Claim 14]: [“The computing system of claim 9, wherein the wells are sorted into the groups on a well- by-well basis using one or more similarity metrics selected from the group consisting of Jaccard similarity, overlap similarity, overlap indicator, overlap similarity by row, one-way similarity by row, and a symmetric similarity matrix.”]; [0232]: [“For each possible pair of wells, compute the sample-wise Jaccard distance (equal to one minus the Jaccard similarity) by testing the footprint of the first well with the sample data of the second and vice versa.”]; [0219]: [“The term footprint is used to describe the decision boundary computed from a one-class support vector machine. In practice, a well may have multiple footprints if partitioning the logs into related groups MDL, DD, WL etc. A well can also have multiple footprints formed by splitting its logs by zone, facies or fluids.”]) and generating a visual representation of the depth in the subsurface formation based on the correlation; (Akkurt, para. [0270]; [“The method 200 may also include sorting the wells into groups, as at 1206. The wells may be sorted based on the well log data and/or the flag. The wells may be sorted into groups using a petrophysical similarity analysis. The petrophysical similarity analysis may include computing a similarity matrix using the well log data and/or the flag. The similarity metrics used in the similarity analysis may include Jaccard, overlap, etc., as described in greater detail above. The results from the similarity analysis may be visualized.”]).
Akkurt does not explicitly teach determining and detecting the boundaries of formation as well as their correlating depths and generating a visual representation of the depth of formation boundaries based on the correlated formation boundaries.
Al-Garni teaches determining and detecting the boundaries of formation as well as their correlating depths (Predicting formation top depths [0014] and predicting the location of formation tops from well log data [0024, Fig. 3] where the reservoir simulator (160) determines one or more formation top depths from seismic data and/or well log data [0028] where FIG. 7 shows an example of formation top depth predictions in comparison to geologist selections in accordance with one or more embodiments. As shown in FIG. 7, a formation top comparison table (700) illustrates a geologist selection (701) of a formation top depth for a respective layer alongside a machine-learning model prediction (702) of the formation top depth for the same layer. The formation top comparison table (700) also shows the difference (703) between a predicted depth by a machine-learning model as well as the geologist selections (701) of depths. Here, formation top depths are determined for various formation layers, i.e., layer A (711), layer B (712), layer C (713), layer D (714), and layer E (715). Accordingly, the output of this machine-learning model in this example highly resembles the geologist selections (701) [0040]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Akkurt with the teachings of Al-Garni to determine and detect the boundaries of formation as well as their correlating depths in order to optimize well placement and reduce human prediction error and subjectivity (Al-Garnin in [0001],.
Akkurt does not explicitly teach generating a visual representation of the depth of formation boundaries based on the correlated formation boundaries.
Gillan teaches generating a visual representation of the depth of formation boundaries based on the correlated formation boundaries (The comparison may include a simple visual comparison of the layers where the system may be configured to import an updated earth model to correlate with the formation boundaries detected by the down hole gamma probe [0066], and the lithology windows may be compared visually, such as comparing the placement and size of formations and formation boundaries. The comparison may highlight differences between the windows visually, such as shading areas of discrepancy red. Additionally, the comparison may include overlaying the lithology windows to create a combined image of the formations [0090]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Akkurt with the teachings of Al-Garni and Gillan to generate a visual representation of the depth of formation boundaries based on the correlated formation boundaries in order to improve subsurface visualization and better understand the exact geometry of the reservoir.
Regarding Claim 2, Akkurt in view of Al-Garni further in view of Gillan teach the the method of claim 1, Akkurt further teaches determining a location in the subsurface formation comprising hydrocarbons based on the correlation; and drilling a well at the determined location; (Akkurt, paras. [0104], [0261], [0037], respectively; [0104]: [“MGL+DD, in combination with cuttings shows and other non-traditional data can be used to flag zones containing hydrocarbons.”]; [0261]: [“The similarity metrics described above represent the similarity of petrophysical responses between wells. There is also information in the relative spatial location of the wells that may be used to weight the similarity of their petrophysical responses. Two wells that are relatively close spatially should have their petrophysical similarity weighted more highly than two wells that are further apart. A spatial proximity matrix is defined as a measure of proximity for each pair of wells.”]; [0037]: [For example, the management components 1 10 may allow for direct or indirect management of sensing, drilling, injecting, extracting, etc., with respect to the geologic environment 150. In turn, further information about the geologic environment 150.”]).
Akkurt does not explicitly teach determining a location based on the correlated formation boundaries.
Al-Garni teaches determining a location based on the correlated formation boundaries (Predicting formation top depths [0014] and predicting the location of formation tops from well log data [0024, Fig. 3] where the reservoir simulator (160) determines one or more formation top depths from seismic data and/or well log data [0028] where FIG. 7 shows an example of formation top depth predictions in comparison to geologist selections in accordance with one or more embodiments. As shown in FIG. 7, a formation top comparison table (700) illustrates a geologist selection (701) of a formation top depth for a respective layer alongside a machine-learning model prediction (702) of the formation top depth for the same layer. The formation top comparison table (700) also shows the difference (703) between a predicted depth by a machine-learning model as well as the geologist selections (701) of depths. Here, formation top depths are determined for various formation layers, i.e., layer A (711), layer B (712), layer C (713), layer D (714), and layer E (715). Accordingly, the output of this machine-learning model in this example highly resembles the geologist selections (701) [0040]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Akkurt with the teachings of Al-Garni and Gillan to determine a location in the subsurface formation comprising hydrocarbons based on the correlated formation boundaries in order to improve oil drilling efficiency.
Regarding Claim 3, Akkurt in view of Al-Garni further in view of Gillan teach the the method of claim 1, Akkurt further teaches wherein calibrating the first model comprises determining a set of log data for each well that minimizes an uncertainty of determined locations; (Akkurt, para. [0179]; [“One criterion for matching the new well data against the library of stored models is to minimize the amount of extrapolation that occurs when applying a model to the new well data. It is known that predictive models tend towards poor performance in extrapolation situations: those where the predictor variables for the new well fall outside the range of the training set for which the model was designed.”]).
Akkurt does not explicitly teach determined locations of formation boundaries.
Al-Garni teaches determined locations of formation boundaries (Predicting formation top depths [0014] and predicting the location of formation tops from well log data [0024, Fig. 3] where the reservoir simulator (160) determines one or more formation top depths from seismic data and/or well log data [0028] where FIG. 7 shows an example of formation top depth predictions in comparison to geologist selections in accordance with one or more embodiments. As shown in FIG. 7, a formation top comparison table (700) illustrates a geologist selection (701) of a formation top depth for a respective layer alongside a machine-learning model prediction (702) of the formation top depth for the same layer. The formation top comparison table (700) also shows the difference (703) between a predicted depth by a machine-learning model as well as the geologist selections (701) of depths. Here, formation top depths are determined for various formation layers, i.e., layer A (711), layer B (712), layer C (713), layer D (714), and layer E (715). Accordingly, the output of this machine-learning model in this example highly resembles the geologist selections (701) [0040]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Akkurt with the teachings of Al-Garni and Gillan to determine a set of log data for each well that minimizes an uncertainty of determined locations of formation boundaries in order to reduce costs, mitigate hazards, and improve oil drilling efficiency.
Regarding Claims 4-5, Akkurt in view of Al-Garni further in view of Gillan teach the method of claim 3, and Akkurt further teaches wherein the first model comprises a change point detection model; Akkurt in view of Al-Garni further in view of Gillan teach the method of claim 4, and Akkurt further teaches wherein the change point detection model comprises a dynamic programming change point detection model, a change point detection model with linear computation cost, a multiple change-point detection model with a reproducing kernel, a binary segmentation change point detection model, a bottom-up segmentation change point detection model, or a sliding window change point detection algorithm; (Akkurt, paras. [0215], [0217], [0218], respectively; [0215]: [“One-class Support Vector Machine [0216] A classification algorithm called the Support Vector Machine (SVM) may be used where the idea of the algorithm is to choose a small number of the training data samples (these are the so-called support vectors) to define a decision boundary which governs the classification process. SVMs have proven to be popular due to their flexibility in capturing complex decision boundaries. An extension to SVMs may allow the user to trace the boundary of a training data set, a problem they call domain description.”];
Specifically, wherein the change point detection model comprises a dynamic programming change point detection model;
Examiner note: Although not specifically named a dynamic programming change point detection model, the disclosure of SVMs in an analogous art in Akkurt read on the claimed limitation mentioned above considering that, for defining purposes, dynamic programming changepoint detection models are used to identify the locations of changepoints within a sequence, which rely on a penalty parameter to regulate the number of changepoints. To estimate this penalty parameter, a variety of simple models may be used such as linear models or decision trees (NPL: Nguyen, 2024).
Given the broadest reasonable interpretation, “creating a decision boundary from logs from one or more input wells” reads on “identifying the locations of changepoints in a sequence” and “parameter being used to create a decision boundary that is tight around the remaining bulk of the training set” reads on a “relying on a penalty parameter to regulate the number of changepoints” where the “penalty parameter is estimated” in steps 1-9 para. [0218].
[0217]: [“The parameters that control the algorithm are:
1. The outlier fraction, a small percentage (e.g., 5%) of the training samples can be treated as outside the decision boundary. This parameter may be used to create a decision boundary that is tight around the remaining bulk of the training set.
2. A parameter that controls the number of support vectors and hence the amount of detail in the decision boundary.”];
[0218]: [“The outlier fraction is a parameter in training the one-class SVM model for outlier detection. An automated procedure is used to pick the value of this parameter, as follows:
1. Select logs from one or more input wells;
2. Train a one-class support vector machine using a value of zero the outlier fraction;
3. Output trained model which defines the normal data footprint at zero percent outliers;
4. Test the entire training set against the footprint, compute the SVM score at each data point;
5. Compute the empirical cumulative distribution function of the SVM score at each data point;
6. Define a regular sampling of cumulative probability between 0 and 100% (e.g., 1%);
7. Resample the empirical cumulative distribution function of SVM scores to the regular sampling of cumulative probability defined in 6;
8. Compute the 2.sup.nd derivative of the resampled empirical cumulative distribution defined in 7;
9. Identify largest peak in the 2.sup.nd derivative subject to user specified limits on cumulative probability; and 10. Output the cumulative probability of the picked peak and the associated SVM score. The output cumulative probability is the value for the outlier fraction parameter.”]).
Regarding Claim 6, Akkurt in view of Al-Garni further in view of Gillan teach the method of claim 1, and Akkurt further teaches wherein the machine learning model comprises an ensemble-based regression model or an artificial neural network model; (Akkurt, paras. [0053], [0158], [0193], respectively; [0053]: [“The systems and methods disclosed herein may predict formation properties that are normally interpreted or measured directly, using Machine Learning (ML) Algorithms…The systems and methods disclosed herein may also use of two classes of ML algorithms, called Random Forest (RF) and Support Vector Machines (SVM)… RF is an ensemble method because it utilizes the output of many decision trees,” where the [See para. 0158] “RF algorithm” is a “variation” of the “Quantile Regression Forest (QRF) Algorithm.”]; [0193]: [“The idea of random forests is to construct a large number of regression trees from bootstrap samples of the training data.”]).
Regarding Claims 7-9, Akkurt in view of Al-Garni further in view of Gillan teach the method of claim 1, and Akkurt further teaches training the machine learning model based on well log data, cuttings-based lithology data, drilling data, and mud gas data from the set of reference wells; [0142]: [“The wells used in the training or Model Building are called the training wells. A well in the training-set has the corresponding data for both the predictor and the ground truth (MGL+DD+GR, and Sw or PHIT, respectively).”]; [0138]: [“The systems and methods disclosed herein may use non-traditional logs, such as Mud Gas Logs (MGL) and Drilling Data (DD), with the addition of a GR log obtained from cuttings or an MWD/LWD run, to predict interpreted formation properties, that may also be determined from WL or LWD logs.”]; [0104]: [“While the WL or LWD logs may not have the resolution to detect laminated pay, MGL+DD, in combination with cuttings shows and other non-traditional data can be used to flag zones containing hydrocarbons. Early knowledge of such a zone may then lead to the collection of additional petrophysical information (e.g., cores or MDT tests) to validate the predictions.”];
[0212]: [“For each measured depth sample in the well of interest:
1. predict the quantiles for the discretized CCDF using the QRF model.
2. interpolate the discretized CCDF to a regular sampling of the response variable.
3. compute the CPDF by differencing the CCDF derived above.
4. compute the Information Gain, a measure of the distance between the CPDF and a reference uniform prior distribution.
5. predict the quantiles for the prediction interval using the QRF model.
6. compute the prediction interval width.”]).
Akkurt in view of Al-Garni further in view of Gillan teach the method of claim 1, and Akkurt further teaches wherein the machine learning model is trained on well log data, cuttings-based lithology data, drilling data, and mud gas data from the set of reference wells;
[0142]: [“The wells used in the training or Model Building are called the training wells. A well in the training-set has the corresponding data for both the predictor and the ground truth (MGL+DD+GR, and Sw or PHIT, respectively).”]; [0138]: [“The systems and methods disclosed herein may use non-traditional logs, such as Mud Gas Logs (MGL) and Drilling Data (DD), with the addition of a GR log obtained from cuttings or an MWD/LWD run, to predict interpreted formation properties, that may also be determined from WL or LWD logs.”]; [0104]: [“While the WL or LWD logs may not have the resolution to detect laminated pay, MGL+DD, in combination with cuttings shows and other non-traditional data can be used to flag zones containing hydrocarbons. Early knowledge of such a zone may then lead to the collection of additional petrophysical information (e.g., cores or MDT tests) to validate the predictions.”];
[0212]: [“For each measured depth sample in the well of interest:
1. predict the quantiles for the discretized CCDF using the QRF model.
2. interpolate the discretized CCDF to a regular sampling of the response variable.
3. compute the CPDF by differencing the CCDF derived above.
4. compute the Information Gain, a measure of the distance between the CPDF and a reference uniform prior distribution.
5. predict the quantiles for the prediction interval using the QRF model.
6. compute the prediction interval width.”]).
Akkurt in view of Al-Garni further in view of Gillan teach the method of claim 1, and Akkurt further teaches wherein the first log data and the second log data comprise formation tops data, cuttings-based lithology data, and well logging data; [0138]: [“The systems and methods disclosed herein may use non-traditional logs, such as Mud Gas Logs (MGL) and Drilling Data (DD), with the addition of a GR log obtained from cuttings or an MWD/LWD run, to predict interpreted formation properties, that may also be determined from WL or LWD logs.”]; [0104]: [“While the WL or LWD logs may not have the resolution to detect laminated pay, MGL+DD, in combination with cuttings shows and other non-traditional data can be used to flag zones containing hydrocarbons. Early knowledge of such a zone may then lead to the collection of additional petrophysical information (e.g., cores or MDT tests) to validate the predictions.”];
Regarding Claim 10, Akkurt in view of Al-Garni further in view of Gillan teach the method of claim 1, and Akkurt further teaches wherein reconstructing log data comprises: providing cuttings-based lithology data, drilling data, and mud gas data from the set of target wells as input to the machine learning model; and receiving the reconstructed log data as output from the machine learning model; (Akkurt, paras. [0142], [0143], [0108], [0110], [0104], respectively; [0142]: [“The wells used in the training or Model Building are called the training wells. A well in the training-set has the corresponding data for both the predictor and the ground truth (MGL+DD+GR, and Sw or PHIT, respectively).”]; [0143]: [“Given an ML model, the user then uses it to predict the formation properties, for the test wells, as shown in the bottom row. Unlike the training-wells, test- wells do not have the ground truth. Given the ML model (bottom row, middle) 310, predictor data from any test- well (bottom row, left) 308 is fed into the model, and formation properties (Sw and PHIT in this case) are predicted (bottom row, right) 312.”]; [0108]: [“Consider a case where one of the logs on the WL run is bad, for example, the density log. Either the problem is discovered too late to repeat the measurement, or a rerun is not considered for operational reasons. A replacement "density" log can be created in a number of ways: (i) using from MGL+DD alone from adjacent wells, (ii) using WL or LWD logs from adjacent wells, (iii) using a combination of (i) and (ii). The caveat in the third case is that invasion physics may be taken into account when combining data acquired at different times during the drilling of a well.”]; [0110]: [“Another variation is a well where there is no log data, due to well collapse, stuck pipe, instability, etc. Replacement logs can be computed from MGL+DD, as they would be acquired as soon as the bit penetrated the formation.”]; [0104]: [“The systems and methods disclosed herein may use non-traditional logs, such as Mud Gas Logs (MGL) and Drilling Data (DD), with the addition of a GR log obtained from cuttings or an MWD/LWD run, to predict interpreted formation properties, that may also be determined from WL or LWD logs.”]).
Regarding Claim 11, Akkurt in view of Al-Garni further in view of Gillan teach the method of claim 1, and Akkurt further teaches wherein determining similarities between identified intervals comprises determining similarities based on an adjusted Rand index, an adjusted mutual information metric, an area under a receiver operating characteristic curve, or an area under a precision-recall curve;
Specifically, an adjusted mutual information metric;
Examiner note: Although not specifically named an adjusted mutual information metric, the process described in paras. [0144] and [0156] in an analogous art in Akkurt read on the claimed limitation mentioned above considering that, given the broadest reasonable interpretation, the process below involves quantitative mutual information values from a set of wells, and this data is adjusted in the way that 2 out of the 6 wells are used to build the RF model and the corrected data is applied to the remaining 4 wells, decisions being based on expected accuracy/precision where in order to decide whether the predictions made by the ML model are reliable enough for decision-making, the end-user may use some metric (uncertainty) to assess the quality of the answers provided.
(Akkurt, paras. [0144 and 0156], respectively; [0144]: [“An example for Sw prediction is shown in Figure 3B, which shows three tracks 320, 322, 324. Given a set of 6 wells, with MGL, DD, WL and core data, a RF model for Sw may be built from two wells, and applied to the remaining four wells. The comparison of the ground truth vs. the prediction, from one of the wells is shown in the middle track 322 of Figure 3B. The agreement between the predicted Sw and ELAN based SW is strong, supporting the case that the predictions are quantitative and have the accuracy/precision expected from logs.”]; [0156]: [“To decide whether the predictions made by the ML model are reliable enough for decision-making, the end-user may use some metric to assess the quality of the answers provided. Such metrics are referred to as measures of "uncertainty," and there are many different approaches and algorithms to produce them.”];
Regarding representative claim 12, Akkurt in view of Al-Garni further in view of Gillan teach a system for making determinations in a subsurface formation, the system comprising: at least one processor; and a memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
receiving first log data from a set of reference wells in a subsurface formation; calibrating a first model to detect, the calibrating being based on the log data from the set of reference wells; receiving second log data from a set of target wells in the subsurface formation, the set of target wells being different than the set of reference wells; reconstructing the second log data from the set of target wells based on a machine learning model, the machine learning model being trained on the log data from the set of reference wells; determining depths based on the calibrated first model; determining similarities between log data in intervals defined by the determined depths from two or more wells, the two or more wells selected from the set of reference wells, the set of target wells, or both; correlating between the two or more wells in the subsurface formation based on the determined similarities; and generating a visual representation of the depth in the subsurface formation based on the correlation; (Akkurt, para. [0004]; [“A computing system includes a processor and a memory system. The memory system includes a non-transitory computer-readable medium storing instructions that, when executed by the processor, cause the computing system to perform operations. The operations include…”]; (Akkurt, paras. [0143], [0096], [0162], respectively; [0143]: [“Given an ML model, the user then uses it to predict the formation properties, for the test wells, as shown in the bottom row. Unlike the training-wells, test-wells do not have the ground truth. Given the ML model (bottom row, middle) 310, predictor data from any test- well (bottom row, left) 308 is fed into the model, and formation properties (Sw and PHIT in this case) are predicted (bottom row, right) 312.”]; [0096]: [“A candidate well is defined as one that has the appropriate input and ground data that can be used in building an ML model.”]; [0162]: [A model may be built using a given a number of candidate training wells. The candidate wells have both the input data to be used in the prediction (e.g., MGL+DD+GR) and the ground truth to be used in the training (Sw, PHIT). The candidate wells may also have the WL or LWD logs that are used in the determination of the ground truth.”]); (Akkurt, paras. [0108], [0110], [0243], [0053], [0212, steps 1-2 and 6], respectively; [0108]: [“Consider a case where one of the logs on the WL run is bad, for example, the density log. Either the problem is discovered too late to repeat the measurement, or a rerun is not considered for operational reasons. A replacement "density" log can be created in a number of ways: (i) using from MGL+DD alone from adjacent wells, (ii) using WL or LWD logs from adjacent wells, (iii) using a combination of (i) and (ii). The caveat in the third case is that invasion physics may be taken into account when combining data acquired at different times during the drilling of a well.”]; [0110]: [“Another variation is a well where there is no log data, due to well collapse, stuck pipe, instability, etc. Replacement logs can be computed from MGL+DD, as they would be acquired as soon as the bit penetrated the formation.”]; [0243]: [“Jaccard similarity is not sensitive to situations where the footprint of well A is a subset of well B. In these situations, Jaccard similarity will be less than one, even if the footprint of well A is completely contained within the footprint of well B. A user may identify these situations because, in this case, well B would be a strong candidate to build a predictive model to reconstruct logs in well A. Overlap similarity provides a way to identify such overlaps.”]; [0053]: [“The systems and methods disclosed herein may predict formation properties that are normally interpreted or measured directly, using Machine Learning (ML) Algorithms. The systems and methods disclosed herein use of Mud Gas Logs (MGL) and Drilling Data (DD), rather than Wireline (WL) or Logging-While-Drilling (LWD) logs, in the prediction of formation properties such as water saturation or total porosity.”]; [0212, steps 1-2 and 6]: [“1. The user selects a well on which to predict the response variable of interest. 2. A pre-existing Quantile Regression Forest (QRF) model is selected from a library or a new QRF model is created from a training dataset. 6. For each measured depth sample in the well of interest”]).
Akkurt does not explicitly teach determining and detecting the boundaries of formation as well as their correlating depths and generating a visual representation of the depth of formation boundaries based on the correlated formation boundaries.
Al-Garni teaches determining and detecting the boundaries of formation as well as their correlating depths (Predicting formation top depths [0014] and predicting the location of formation tops from well log data [0024, Fig. 3] where the reservoir simulator (160) determines one or more formation top depths from seismic data and/or well log data [0028] where FIG. 7 shows an example of formation top depth predictions in comparison to geologist selections in accordance with one or more embodiments. As shown in FIG. 7, a formation top comparison table (700) illustrates a geologist selection (701) of a formation top depth for a respective layer alongside a machine-learning model prediction (702) of the formation top depth for the same layer. The formation top comparison table (700) also shows the difference (703) between a predicted depth by a machine-learning model as well as the geologist selections (701) of depths. Here, formation top depths are determined for various formation layers, i.e., layer A (711), layer B (712), layer C (713), layer D (714), and layer E (715). Accordingly, the output of this machine-learning model in this example highly resembles the geologist selections (701) [0040]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Akkurt with the teachings of Al-Garni to determine and detect the boundaries of formation as well as their correlating depths in order to reduce human prediction error and subjectivity (Al-Garnin in [0001].
Al-Garni does not explicitly teach generating a visual representation of the depth of formation boundaries based on the correlated formation boundaries.
Gillan teaches generating a visual representation of the depth of formation boundaries based on the correlated formation boundaries (The comparison may include a simple visual comparison of the layers where the system may be configured to import an updated earth model to correlate with the formation boundaries detected by the down hole gamma probe [0066], and the lithology windows may be compared visually, such as comparing the placement and size of formations and formation boundaries. The comparison may highlight differences between the windows visually, such as shading areas of discrepancy red. Additionally, the comparison may include overlaying the lithology windows to create a combined image of the formations [0090]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Akkurt with the teachings of Al-Garni and Gillan to generate a visual representation of the depth of formation boundaries based on the correlated formation boundaries in order to improve subsurface visualization and better understand the exact geometry of the reservoir.
Regarding Claim 13, Akkurt in view of Al-Garni further in view of Gillan teach the system of claim 12, and Akkurt further teaches wherein the model comprises a dynamic programming change point detection model, a change point detection model with linear computation cost, a multiple change-point detection model with a reproducing kernel, a binary segmentation change point detection model, a bottom-up segmentation change point detection model, or a sliding window change point detection algorithm; (Akkurt, paras. [0215], [0217], [0218], respectively; [0215]: [“One-class Support Vector Machine [0216] A classification algorithm called the Support Vector Machine (SVM) may be used where the idea of the algorithm is to choose a small number of the training data samples (these are the so-called support vectors) to define a decision boundary which governs the classification process. SVMs have proven to be popular due to their flexibility in capturing complex decision boundaries. An extension to SVMs may allow the user to trace the boundary of a training data set, a problem they call domain description.”];
Specifically, wherein the change point detection model comprises a dynamic programming change point detection model, considering that dynamic programming changepoint detection models are used to identify the locations of changepoints within a sequence, which rely on a penalty parameter to regulate the number of changepoints. To estimate this penalty parameter, a variety of simple models may be used such as linear models or decision trees (NPL: Nguyen, 2024).
[0217]: [“The parameters that control the algorithm are:
1. The outlier fraction, a small percentage (e.g., 5%) of the training samples can be treated as outside the decision boundary. This parameter may be used to create a decision boundary that is tight around the remaining bulk of the training set.
2. A parameter that controls the number of support vectors and hence the amount of detail in the decision boundary.”];
[0218]: [“The outlier fraction is a parameter in training the one-class SVM model for outlier detection. An automated procedure is used to pick the value of this parameter, as follows:
1. Select logs from one or more input wells;
2. Train a one-class support vector machine using a value of zero the outlier fraction;
3. Output trained model which defines the normal data footprint at zero percent outliers;
4. Test the entire training set against the footprint, compute the SVM score at each data point;
5. Compute the empirical cumulative distribution function of the SVM score at each data point;
6. Define a regular sampling of cumulative probability between 0 and 100% (e.g., 1%);
7. Resample the empirical cumulative distribution function of SVM scores to the regular sampling of cumulative probability defined in 6;
8. Compute the 2.sup.nd derivative of the resampled empirical cumulative distribution defined in 7;
9. Identify largest peak in the 2.sup.nd derivative subject to user specified limits on cumulative probability; and 10. Output the cumulative probability of the picked peak and the associated SVM score. The output cumulative probability is the value for the outlier fraction parameter.”]).
Regarding Claim 14, Akkurt in view of Al-Garni further in view of Gillan teach the system of claim 12, and Akkurt further teaches wherein the machine learning model comprises an ensemble-based regression model or an artificial neural network model; (Akkurt, paras. [0053], [0158], [0193], respectively; [0053]: [“The systems and methods disclosed herein may predict formation properties that are normally interpreted or measured directly, using Machine Learning (ML) Algorithms…The systems and methods disclosed herein may also use of two classes of ML algorithms, called Random Forest (RF) and Support Vector Machines (SVM)… RF is an ensemble method because it utilizes the output of many decision trees,” where (See [para. 0158]) the “RF algorithm” is a “variation” of the “Quantile Regression Forest (QRF) Algorithm.”]; [0193]: [“The idea of random forests is to construct a large number of regression trees from bootstrap samples of the training data.”]).
Regarding Claim 15, Akkurt in view of Al-Garni further in view of Gillan teach the system of claim 12, and Akkurt further teaches wherein the first log data and the second log data comprise formation tops data, cuttings-based lithology data, and well logging data; (Akkurt, paras. [0104], [0138], respectively; [0138]: [“The systems and methods disclosed herein may use non-traditional logs, such as Mud Gas Logs (MGL) and Drilling Data (DD), with the addition of a GR log obtained from cuttings or an MWD/LWD run, to predict interpreted formation properties, that may also be determined from WL or LWD logs.”]; [0104]: [“While the WL or LWD logs may not have the resolution to detect laminated pay, MGL+DD, in combination with cuttings shows and other non-traditional data can be used to flag zones containing hydrocarbons. Early knowledge of such a zone may then lead to the collection of additional petrophysical information (e.g., cores or MDT tests) to validate the predictions.”];
Regarding Claim 16, Akkurt in view of Al-Garni further in view of Gillan teach the system of claim 12, and Akkurt further teaches wherein reconstructing log data comprises: providing cuttings-based lithology data, drilling data, and mud gas data from the set of target wells as input to the machine learning model; and receiving the reconstructed log data as output from the machine learning model; (Akkurt, paras. [0142], [0143], [0108], [0110], [0104], respectively; [0142]: [“The wells used in the training or Model Building are called the training wells. A well in the training-set has the corresponding data for both the predictor and the ground truth (MGL+DD+GR, and Sw or PHIT, respectively).”]; [0143]: [“Given an ML model, the user then uses it to predict the formation properties, for the test wells, as shown in the bottom row. Unlike the training-wells, test- wells do not have the ground truth. Given the ML model (bottom row, middle) 310, predictor data from any test- well (bottom row, left) 308 is fed into the model, and formation properties (Sw and PHIT in this case) are predicted (bottom row, right) 312.”]; [0108]: [“Consider a case where one of the logs on the WL run is bad, for example, the density log. Either the problem is discovered too late to repeat the measurement, or a rerun is not considered for operational reasons. A replacement "density" log can be created in a number of ways: (i) using from MGL+DD alone from adjacent wells, (ii) using WL or LWD logs from adjacent wells, (iii) using a combination of (i) and (ii). The caveat in the third case is that invasion physics may be taken into account when combining data acquired at different times during the drilling of a well.”]; [0110]: [“Another variation is a well where there is no log data, due to well collapse, stuck pipe, instability, etc. Replacement logs can be computed from MGL+DD, as they would be acquired as soon as the bit penetrated the formation.”]; [0104]: [“The systems and methods disclosed herein may use non-traditional logs, such as Mud Gas Logs (MGL) and Drilling Data (DD), with the addition of a GR log obtained from cuttings or an MWD/LWD run, to predict interpreted formation properties, that may also be determined from WL or LWD logs.”]).
Regarding representative claim 17, Akkurt in view of Al-Garni further in view of Gillan teach one or more non-transitory machine-readable storage devices storing instructions for making determinations in a subsurface formation, the instructions being executable by one or more processors, to cause performance of operations comprising:
Receiving first log data from a set of reference wells in a subsurface formation; calibrating a first model to detect, the calibrating being based on the log data from the set of reference wells; receiving second log data from a set of target wells in the subsurface formation, the set of target wells being different than the set of reference wells; reconstructing the second log data from the set of target wells based on a machine learning model, the machine learning model being trained on the log data from the set of reference wells; determining depths based on the first model; determining similarities between log data in intervals defined by the determined depths from two or more wells, the two or more wells selected from the set of reference wells, the set of target wells, or both; correlating between the two or more wells in the subsurface formation based on the determined similarities; and generating a visual representation of the depth in the subsurface formation based on the correlation; (Akkurt, para. [0005]; [“A non-transitory computer-readable medium is also disclosed. The medium stores instructions that, when executed by a processor of a computing system, cause the computing system to perform operations. The operations include…”]; (Akkurt, paras. [0143], [0096], [0162], respectively; [0143]: [“Given an ML model, the user then uses it to predict the formation properties, for the test wells, as shown in the bottom row. Unlike the training-wells, test-wells do not have the ground truth. Given the ML model (bottom row, middle) 310, predictor data from any test- well (bottom row, left) 308 is fed into the model, and formation properties (Sw and PHIT in this case) are predicted (bottom row, right) 312.”];
[0096]: [“A candidate well is defined as one that has the appropriate input and ground data that can be used in building an ML model.”]; [0162]: [A model may be built using a given a number of candidate training wells. The candidate wells have both the input data to be used in the prediction (e.g., MGL+DD+GR) and the ground truth to be used in the training (Sw, PHIT). The candidate wells may also have the WL or LWD logs that are used in the determination of the ground truth.”]); (Akkurt, paras. [0108], [0110], [0243], [0053], [0212, steps 1-2 and 6], respectively; [0108]: [“Consider a case where one of the logs on the WL run is bad, for example, the density log. Either the problem is discovered too late to repeat the measurement, or a rerun is not considered for operational reasons. A replacement "density" log can be created in a number of ways: (i) using from MGL+DD alone from adjacent wells, (ii) using WL or LWD logs from adjacent wells, (iii) using a combination of (i) and (ii). The caveat in the third case is that invasion physics may be taken into account when combining data acquired at different times during the drilling of a well.”]; [0110]: [“Another variation is a well where there is no log data, due to well collapse, stuck pipe, instability, etc. Replacement logs can be computed from MGL+DD, as they would be acquired as soon as the bit penetrated the formation.”]; [0243]: [“Jaccard similarity is not sensitive to situations where the footprint of well A is a subset of well B. In these situations, Jaccard similarity will be less than one, even if the footprint of well A is completely contained within the footprint of well B. A user may identify these situations because, in this case, well B would be a strong candidate to build a predictive model to reconstruct logs in well A. Overlap similarity provides a way to identify such overlaps.”]; [0053]: [“The systems and methods disclosed herein may predict formation properties that are normally interpreted or measured directly, using Machine Learning (ML) Algorithms. The systems and methods disclosed herein use of Mud Gas Logs (MGL) and Drilling Data (DD), rather than Wireline (WL) or Logging-While-Drilling (LWD) logs, in the prediction of formation properties such as water saturation or total porosity.”]; [0212, steps 1-2 and 6]: [“1. The user selects a well on which to predict the response variable of interest. 2. A pre-existing Quantile Regression Forest (QRF) model is selected from a library or a new QRF model is created from a training dataset. 6. For each measured depth sample in the well of interest”]).
Akkurt does not explicitly teach determining and detecting the boundaries of formation as well as their correlating depths and generating a visual representation of the depth of formation boundaries based on the correlated formation boundaries.
Al-Garni teaches determining and detecting the boundaries of formation as well as their correlating depths (Predicting formation top depths [0014] and predicting the location of formation tops from well log data [0024, Fig. 3] where the reservoir simulator (160) determines one or more formation top depths from seismic data and/or well log data [0028] where FIG. 7 shows an example of formation top depth predictions in comparison to geologist selections in accordance with one or more embodiments. As shown in FIG. 7, a formation top comparison table (700) illustrates a geologist selection (701) of a formation top depth for a respective layer alongside a machine-learning model prediction (702) of the formation top depth for the same layer. The formation top comparison table (700) also shows the difference (703) between a predicted depth by a machine-learning model as well as the geologist selections (701) of depths. Here, formation top depths are determined for various formation layers, i.e., layer A (711), layer B (712), layer C (713), layer D (714), and layer E (715). Accordingly, the output of this machine-learning model in this example highly resembles the geologist selections (701) [0040]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Akkurt with the teachings of Al-Garni to determine and detect the boundaries of formation as well as their correlating depths in order to reduce human prediction error and subjectivity (Al-Garnin in [0001]
Al-Garni does not explicitly teach generating a visual representation of the depth of formation boundaries based on the correlated formation boundaries.
Gillan teaches generating a visual representation of the depth of formation boundaries based on the correlated formation boundaries (The comparison may include a simple visual comparison of the layers where the system may be configured to import an updated earth model to correlate with the formation boundaries detected by the down hole gamma probe [0066], and the lithology windows may be compared visually, such as comparing the placement and size of formations and formation boundaries. The comparison may highlight differences between the windows visually, such as shading areas of discrepancy red. Additionally, the comparison may include overlaying the lithology windows to create a combined image of the formations [0090]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Akkurt with the teachings of Al-Garni and Gillan to generate a visual representation of the depth of formation boundaries based on the correlated formation boundaries in order to improve subsurface visualization and better understand the exact geometry of the reservoir.
Regarding Claim 18, Akkurt in view of Al-Garni further in view of Gillan teach the non-transitory, machine-readable storage devices of claim 17, and Akkurt further teaches wherein the model comprises a dynamic programming change point detection model, a change point detection model with linear computation cost, a multiple change-point detection model with a reproducing kernel, a binary segmentation change point detection model, a bottom-up segmentation change point detection model, or a sliding window change point detection algorithm; (Akkurt, para. [0005]; [“A non-transitory computer-readable medium is also disclosed. The medium stores instructions that, when executed by a processor of a computing system, cause the computing system to perform operations. The operations include…”]; (Akkurt, paras. [0215], [0217], [0218], respectively; [0215]: [“One-class Support Vector Machine [0216] A classification algorithm called the Support Vector Machine (SVM) may be used where the idea of the algorithm is to choose a small number of the training data samples (these are the so-called support vectors) to define a decision boundary which governs the classification process. SVMs have proven to be popular due to their flexibility in capturing complex decision boundaries. An extension to SVMs may allow the user to trace the boundary of a training data set, a problem they call domain description.”];
Specifically, wherein the change point detection model comprises a dynamic programming change point detection model, considering that dynamic programming changepoint detection models are used to identify the locations of changepoints within a sequence, which rely on a penalty parameter to regulate the number of changepoints. To estimate this penalty parameter, a variety of simple models may be used such as linear models or decision trees (NPL: Nguyen, 2024).
[0217]: [“The parameters that control the algorithm are:
1. The outlier fraction, a small percentage (e.g., 5%) of the training samples can be treated as outside the decision boundary. This parameter may be used to create a decision boundary that is tight around the remaining bulk of the training set.
2. A parameter that controls the number of support vectors and hence the amount of detail in the decision boundary.”];
[0218]: [“The outlier fraction is a parameter in training the one-class SVM model for outlier detection. An automated procedure is used to pick the value of this parameter, as follows:
1. Select logs from one or more input wells;
2. Train a one-class support vector machine using a value of zero the outlier fraction;
3. Output trained model which defines the normal data footprint at zero percent outliers;
4. Test the entire training set against the footprint, compute the SVM score at each data point;
5. Compute the empirical cumulative distribution function of the SVM score at each data point;
6. Define a regular sampling of cumulative probability between 0 and 100% (e.g., 1%);
7. Resample the empirical cumulative distribution function of SVM scores to the regular sampling of cumulative probability defined in 6;
8. Compute the 2.sup.nd derivative of the resampled empirical cumulative distribution defined in 7;
9. Identify largest peak in the 2.sup.nd derivative subject to user specified limits on cumulative probability; and 10. Output the cumulative probability of the picked peak and the associated SVM score. The output cumulative probability is the value for the outlier fraction parameter.”]).
Regarding Claim 19, Akkurt in view of Al-Garni further in view of Gillan teach the non-transitory, machine-readable storage devices of claim 17, and Akkurt further teaches wherein the machine learning model comprises an ensemble-based regression model or an artificial neural network model; (Akkurt, para. [0005]; [“A non-transitory computer-readable medium is also disclosed. The medium stores instructions that, when executed by a processor of a computing system, cause the computing system to perform operations. The operations include…”]; (Akkurt, paras. [0053], [0158], [0193], respectively;
[0053]: [“The systems and methods disclosed herein may predict formation properties that are normally interpreted or measured directly, using Machine Learning (ML) Algorithms…The systems and methods disclosed herein may also use of two classes of ML algorithms, called Random Forest (RF) and Support Vector Machines (SVM)… RF is an ensemble method because it utilizes the output of many decision trees,” where (See para. [0158]) the “RF algorithm” is a “variation” of the “Quantile Regression Forest (QRF) Algorithm.”]; [0193]: [“The idea of random forests is to construct a large number of regression trees from bootstrap samples of the training data.”]).
Regarding Claim 20, Akkurt in view of Al-Garni further in view of Gillan teach the non-transitory, machine-readable storage devices of claim 17, and Akkurt further teaches wherein reconstructing log data comprises: providing cuttings-based lithology data, drilling data, and mud gas data from the set of target wells as input to the machine learning model; and receiving the reconstructed log data as output from the machine learning model; (Akkurt, para. [0005]; [“A non-transitory computer-readable medium is also disclosed. The medium stores instructions that, when executed by a processor of a computing system, cause the computing system to perform operations. The operations include…”];
(Akkurt, paras. [0142], [0143], [0108], [0110], [0104], respectively; [0142]: [“The wells used in the training or Model Building are called the training wells. A well in the training-set has the corresponding data for both the predictor and the ground truth (MGL+DD+GR, and Sw or PHIT, respectively).”]; [0143]: [“Given an ML model, the user then uses it to predict the formation properties, for the test wells, as shown in the bottom row. Unlike the training-wells, test- wells do not have the ground truth. Given the ML model (bottom row, middle) 310, predictor data from any test- well (bottom row, left) 308 is fed into the model, and formation properties (Sw and PHIT in this case) are predicted (bottom row, right) 312.”]; [0108]: [“Consider a case where one of the logs on the WL run is bad, for example, the density log. Either the problem is discovered too late to repeat the measurement, or a rerun is not considered for operational reasons. A replacement "density" log can be created in a number of ways: (i) using from MGL+DD alone from adjacent wells, (ii) using WL or LWD logs from adjacent wells, (iii) using a combination of (i) and (ii). The caveat in the third case is that invasion physics may be taken into account when combining data acquired at different times during the drilling of a well.”];
[0110]: [“Another variation is a well where there is no log data, due to well collapse, stuck pipe, instability, etc. Replacement logs can be computed from MGL+DD, as they would be acquired as soon as the bit penetrated the formation.”]; [0104]: [“The systems and methods disclosed herein may use non-traditional logs, such as Mud Gas Logs (MGL) and Drilling Data (DD), with the addition of a GR log obtained from cuttings or an MWD/LWD run, to predict interpreted formation properties, that may also be determined from WL or LWD logs.”]).
Pertinent Prior Art
US 20210340861 A1: A method may include obtaining a request to determine automatically a depth of a formation top for a well in a geological region of interest. The method may include obtaining various well logs regarding the well and various wells in the geological region of interest. The method may include determining various depth values using the various well logs and a statistical interpolation method. The method may further include determining a final depth of the well using the various depth values and a searching method.
Conclusion
An inquiry concerning this communication or earlier communication from the examiner should be directed to LOGAN D COONS whose telephone number is (571) 272-2698. (via email: logan.coons@uspto.gov “without a written authorization by applicant in place, the USPTO will not respond via internet e-mail to an internet correspondence” MPEP 502.02 II). The examiner can normally be reached on M-F 9:30am – 6pm ET.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, SPE SHELBY A TURNER, can be reached at (571) 272-6334. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/LOGAN D COONS/Examiner, Art Unit 2857
/SHELBY A TURNER/Supervisory Patent Examiner, Art Unit 2857
Prosecution Insights