A computer-implemented device, method, and computer-readable non-transitary storage medium.

Description

36 A device, a method implemented by a computer, and a computer-readable non-transient storage medium. COMPUTER INTRODUCTION 1. RELATED REQUEST

[001] This request claims benefit to the Non-Provisional Request of U.S. Application No. 18 / 386,469, entitled “GENERATIVE MACHINE LEARNING BASED PETROPHYSICS INTERPRETATION”, filed November 2, 2023, which claims the benefit of U.S. Provisional Application No. 63 / 469,314, entitled “GENERATIVE MACHINE LEARNING BASED PETROPHYSICS INTERPRETATION”, filed May 26, 2023, which is incorporated herein by reference in its entirety. BACKGROUND 2. TECHNICAL FIELD

[002] This disclosure generally refers to solutions for analyzing and interpreting geophysical and petrophysical data and, in particular, for using generative machine learning models to characterize and predict formation and / or reservoir properties. 3. INTRODUCTION

[003] Petrophysical interpretation is the process of analyzing and interpreting geophysical and petrophysical measurement data to characterize and predict reservoir or formation properties. Data used to perform petrophysical interpretations may include, but are not limited to, well logs (e.g., measurements of physical properties of rocks and fluids in the subsurface), core samples, production data, seismic data, and other data sources. BRIEF DESCRIPTION OF THE FIGURES

[004] Certain features of the technology in question are defined in the appended claims. However, the appended drawings, which Petition 870250088766, dated 09 / 30 / 2025, page 12 / 66 / 36, were included to provide greater understanding, illustrate the aspects disclosed and, together with the description, serve to explain the principles of the technology in question. In the drawings: Figure 1A is a schematic diagram of an example of a cable profiling environment, according to some aspects of the disclosed technology.

[005] Figure 1B is a schematic diagram of an example of a profiling environment from Figure 1A, according to some aspects of the disclosed technology.

[006] Figure 2 illustrates a conceptual system diagram for generating petrophysical interpretations using a generative machine learning (ML) model, according to some aspects of the disclosed technology.

[007] Figure 3 illustrates an example of a system that includes a Variational Autoencoder (VAE) ML model, according to some aspects of the disclosed technology.

[008] Figure 4 illustrates a visualization for clustering latent space data, according to some aspects of the disclosed technology.

[009] Figure 5 illustrates an example of a workflow that can be used to perform latent space clustering using a VAE, according to some aspects of the disclosed technology.

[0010] Figure 6 illustrates an example workflow for using a generative machine learning model to perform latent space clustering, according to some aspects of the disclosed technology.

[0011] Figure 7 illustrates an example workflow for performing latent space clustering using a VAE and a generative machine learning model, according to some aspects of Petition 870250088766, dated 09 / 30 / 2025, page 13 / 66 / 36, technology disclosed.

[0012] Figure 8 illustrates an example of a generative adversarial network (GAN) system, according to some aspects of the disclosed technology.

[0013] Figure 9 illustrates a system diagram that uses GANs to perform lithographic classification, according to some aspects of the disclosed technology.

[0014] Figure 10 illustrates an example workflow for implementing a generative model with an uncertain number of inputs, according to some aspects of the disclosed technology.

[0015] Figure 11 illustrates an example of a process for using a generative machine learning model to generate a petrophysical interpretation, according to some aspects of the disclosed technology.

[0016] Figure 12 illustrates an example of a machine learning architecture, according to some aspects of the disclosed technology.

[0017] Figure 13 illustrates an example of a processor-based system with which some aspects of the technology in question can be implemented. DETAILED DESCRIPTION

[0018] The detailed description presented below is intended to be a description of various configurations of the technology in question and is not intended to represent the only configurations in which the technology in question may be practiced. The attached drawings are incorporated into this document and form part of the detailed description. The detailed description includes specific details in order to provide a more complete understanding of the technology in question. However, it will be clear and evident that the technology in question is not limited to the specific details presented here and may be practiced without these details. In some cases, structures and components are shown in block diagram form to avoid obscuring them. Petition 870250088766, dated 09 / 30 / 2025, page 14 / 66 / 36 certain concepts.

[0019] The data needed to perform petrophysical interpretations may include raw well log data, including measurements such as electrical resistivities, acoustic velocities, and transverse relaxation time (T2) distributions of nuclear magnetic resonance (NMR). Other data sources, such as seismic data and core samples, may also be used as input. The result of a petrophysical interpretation may be a set of interpreted values ​​of reservoir properties such as porosity, permeability, and fluid saturation, as well as other geological features such as lithology and structural elements.

[0020] Challenges may arise in performing petrophysical interpretations due to the complexity and heterogeneity of subsurface geological formations. Interpreting well log data may require significant experience and knowledge in petrophysics and geology, as well as an understanding of the specific characteristics of the reservoir being analyzed. Therefore, the interpretation process can be time-consuming and prone to errors due to the subjectivity of interpretation and the inherent uncertainty of subsurface data.

[0021] Furthermore, traditional petrophysical interpretation can involve a manual process of selecting and analyzing individual well log curves, which can be tedious and prone to errors. It can also require human experts to make subjective decisions about which interpretation methods and parameters to use, which can lead to variability in results.

[0022] In summary, petrophysical interpretation can be an essential step in characterizing and predicting the properties of subsurface geological formations. However, the complexity and heterogeneity of the data can make the process challenging and time-consuming, and traditional interpretation methods may be subject to errors and variability. Petition 870250088766, dated 09 / 30 / 2025, page 15 / 66 / 36

[0023] The disclosed technology aspects meet the aforementioned needs by providing machine learning solutions to facilitate the petrophysical interpretation process in various application scenarios, including, but not limited to, lithological interpretation, reservoir characterization, geomechanical interpretation, and joint petrophysical and geophysical interpretation.

[0024] In some examples, petrophysical interpretation can be facilitated using generative machine learning, which involves training a model to generate petrophysical interpretations of geophysical and petrophysical data. This approach can be used to automate the interpretation of petrophysical data, which can improve the efficiency and accuracy of the interpretation process in the oil and gas sector.

[0025] Generative machine learning models, including but not limited to generative adversarial networks (GANs), variational autoencoders (VAEs), flow-based generative models (FBGMs), Naive Bayes classifiers, Gaussian Mixture Models (GMMs), Hidden Markov Models (HMMs), and Restricted Boltzmann Machines (RBMs), can be trained on large petrophysical datasets (or measurement data or formation measurement data), such as well logs, seismic data, and core samples. In some cases, these models can learn the underlying patterns and relationships in the measurement data and generate new interpretations based on this knowledge.

[0026] The generated petrophysical interpretations can be used to predict reservoir properties such as rock types, fluid types, porosity, permeability, formation pressure, and saturation, which can be essential for reservoir characterization and production forecasting.In some cases, this approach can also be used to identify anomalies or patterns in the data that may not be immediately apparent to human experts. Petition 870250088766, dated 09 / 30 / 2025, page 16 / 66 / 36

[0027] In some examples, an advantage of generative machine learning-based petrophysical interpretation is that it can help reduce the time and cost of manually interpreting petrophysical data, which can be very time-consuming and require significant expertise. Generative machine learning-based petrophysical interpretation may also have the potential to improve the accuracy and consistency of the interpretation, as it can incorporate a large amount of data and learn the underlying patterns in a way that can be difficult for humans.

[0028] The disclosure now turns to Figures 1A-B to provide a brief introductory description of the larger systems that can be employed to practice the concepts, methods, and techniques disclosed here. A more detailed description of the methods and systems for implementing the enhanced similarity processing techniques of the disclosed technology will then be presented.

[0029] Figure 1A is a schematic diagram of an example of a cable 100 profiling environment, according to various aspects of the technology in question. As shown in this example, a drilling platform 102 supports a tower 104 which has a movable block 106 for raising and lowering the drill string 108. A kelly 110 supports the drill string 108 as it is lowered via a rotating cable 112. A drill bit 114 is driven by a downhole motor and / or by the rotation of the drill string 108. As a drill bit 114 of the drill string 108 rotates, it drills a well hole 116 that passes through one or more formations 118. A pump 120 circulates drilling fluid through a feed pipe 122 to the kelly 110 inside the drill string 108 and holes in the drill bit 114, back to the surface through the ring around the drill string 108 and into Petition 870250088766, dated 09 / 30 / 2025, page 17 / 66 / 36 of a retention tank 124. The drilling fluid transports the cuttings from the wellbore to tank 124 and helps maintain the integrity of the wellbore.

[0030] A downhole tool 126 may take the form of a drill collar (e.g., a thick-walled tubular pipe that provides weight and rigidity to aid the drilling process) or any other known and / or suitable arrangement. Furthermore, the downhole tool 126 may include one or more logging tools, such as, for example and without limitation, one or more acoustic logging tools (e.g., sonic, ultrasonic, etc.) and / or one or more other types of logging tools and / or corresponding components. The downhole tool 126 may be integrated into a downhole assembly 125 near the drill bit 114. As the drill bit 114 extends the wellbore through the formations, the downhole assembly 125 may collect logging data and / or sensor data (e.g., NMR data and / or any other logging and / or sensor data).The 126 downhole tool may include transmitters (e.g., monopole, dipole, quadrupole, etc.) to generate and transmit signals / waves in the wellbore environment, such as, but not limited to, acoustic signals / waves, radio frequency (RF) signals / waves, optical signals / waves, and / or any other signals / waves. These signals / waves propagate within and along the wellbore and surrounding formation(s) and create signal responses or waveforms, which are received / recorded by one or more receivers.

[0031] For communication purposes, a downhole telemetry sub 128 may be included in the downhole assembly 125 to transfer measurement data to a surface receiver 132 and receive commands from the surface (e.g., from a device on the surface, such as a computer and / or a transmitter). Mud pulse telemetry is an example of a telemetry technique for transferring measurements from tools to Petition 870250088766, dated 09 / 30 / 2025, page 18 / 66 / 36 surface receivers and receive commands from the surface. However, other telemetry techniques can also be used. Other non-limiting examples of telemetry techniques that can be implemented may include fiber optic telemetry, electrical telemetry, acoustic telemetry through the tube, and electromagnetic (EM) telemetry, among others. In some respects, the 128 telemetry sub can store profiling data for later retrieval on the surface when the profiling set is recovered.

[0032] On the surface, surface receiver 132 can receive the uplink signal from downhole telemetry sub 128. Surface receiver 132 may include, for example and without limitation, a wireless receiver, a computer (e.g., a laptop, a desktop computer, a tablet, a server computer and / or any other type of computer) and / or any other device with data communication capabilities (e.g., wired and / or wireless). In some cases, surface receiver 132 may communicate the signal from downhole telemetry sub 128 to a data acquisition system (not shown). Such a data acquisition system may be part of surface receiver 132 or may be a separate device, such as, for example, a computer, a storage device, etc.Surface receiver 132 may include one or more processors, storage devices, input devices, output devices, memory devices, software, and / or the like. Surface receiver 132 may collect, store, and / or process data received from tool 126, as described herein.

[0033] In some examples, surface receptor 132 may include a single receptor or multiple receptors. In some cases, surface receptor 132 may include a set of evenly spaced receptors or a set of receptors in any other arrangement. Surface receptor 132 may include multiple receptors arranged in a Petition 870250088766, dated 09 / 30 / 2025, page 19 / 66 / 36 matrix and / or uniformly spaced (or spaced in any other configuration / arrangement) to facilitate the capture and processing of response signals at specific intervals. The response signals / waves can be analyzed to determine properties and / or characteristics of the wellbore and the adjacent formation. Depending on the implementation, other logging tools can be implemented. For example, logging tools configured to measure electrical, nuclear, gamma, and / or magnetism levels can be used. Logging tools can also be implemented to measure other properties, events, and / or conditions, such as, but not limited to, pressure, measuring fluid viscosity, measuring temperature, performing fluid identification, measuring tool orientation, and / or obtaining any other measurements.

[0034] At various times during the well drilling process, the drill string 108 may be removed from the wellbore 116, as shown in environment example 101, illustrated in Figure 1B. After removal of the drill string 108, logging operations may be conducted using the downhole tool 126 (e.g., a logging tool, a sensing instrument probe, etc.) suspended by a transport means (e.g., the transport means 144 shown in Figure 1B). In one or more examples, the transport may be or include a cable with conductors to carry power to the tool and telemetry from the tool to the surface. In some examples, the downhole tool 126 may have centering pads and / or springs to keep the tool close to the centerline of the wellbore or to tilt the tool toward the wellbore wall as the tool is moved up or down the well.

[0035] In some examples, the drilling tool 126 may include an acoustic or sonic logging instrument that collects acoustic logging data within the wellbore 116. As mentioned Petition 870250088766, dated 09 / 30 / 2025, page 20 / 66 / 36 above, other logging instruments may be used additionally or alternatively. A logging installation may include a computer system, such as computer system 600 described with reference to Figure 6, to collect, store and / or process the data / measurements collected by the downhole tool 126. For example, the logging installation may include a logging data management system to modify NMR logging data to be compatible with a temperature correction algorithm determined based on laboratory-generated NMR data and apply the temperature correction algorithm to the modified logging data, if necessary.

[0036] In one or more instances, a downhole tool 126 transport may include at least one of the following: wires, conductive and / or non-conductive cable (e.g., steel cable, etc.) and / or tubular transport means such as coiled tubing, tubing string or downhole tractor. In some cases, the downhole tool 126 may have a local power source such as batteries, a downhole generator and / or the like. When employing a non-conductive cable, coiled tubing, tubing string or a downhole tractor, communication may be supported using, for example, wireless protocols (e.g., EM, acoustic, etc.) and / or measurements and logging data may be stored in local memory for later retrieval. In some respects, electrical or optical telemetry is provided using conductive cables and / or fiber optic signal paths.

[0037] In Figure 1B, a tool with tool body 146 can be employed in “wire-on” systems to perform logging or other operations. For example, instead of using drill string 108 of FIG. 1A to lower tool body 146, which may contain sensors or other instrumentation to detect and record features and conditions near the drill well and surrounding formation, the Petition 870250088766, dated 09 / 30 / 2025, p. 21 / 66 / 36 The tool body 146 can be lowered by a cable transport system 144. Thus, as shown in Figure 1B, the tool body 146 can be lowered into the borehole 116 by cable transport 144. Cable transport 144 can be anchored to a drilling rig 142 or to portable means such as a truck. Cable transport 144 can include one or more wires, flat cables, wire ropes and / or the like, as well as tubular transports such as spiral tubing, joint tubing or other tubulars.

[0038] The wire transport illustrated 144 may provide support for the tool (e.g., tool body 146), enable communication between tool processors on the surface, and / or provide a power source. The cable transport 144 may include fiber optic cabling to carry out communications. The transport cable 144 may be sufficiently strong and flexible to secure the tool body 146 through the borehole 116, while enabling communication via the transport cable 144 with one or more local processors 148B and / or one or more remote processors 148A, 148N. Power may be supplied via the wire transport 144 to meet the tool's power requirements. For smooth cable or coiled tubing configurations, power may be supplied at the bottom of the hole with a battery or via a downhole generator, for example.

[0039] Although Figures 1A and 1B show specific wellbore configurations, it is understood that this disclosure is suitable for use in drilling wells with other orientations, including vertical, horizontal, inclined, multilateral and similar drilling wells. Although Figures 1A and 1B depict an onshore operation, it should also be understood that this disclosure is suitable for use in offshore operations. Furthermore, this disclosure is not limited to offshore environments. Petition 870250088766, dated 09 / 30 / 2025, page 22 / 66 / 36 represented in Figures 1A and 1B and can also be used in other well operations, such as, for example and without limitation, production piping operations, articulated piping operations, coiled piping operations, combinations of these operations and / or similar operations.

[0040] It is also understood that the drilling wells discussed and the production from such drilling wells, which may be an end result of this disclosure, may, in addition to wells for hydrocarbon production, also include wells for geothermal energy production and wells for injection, sequestration or storage of materials (e.g., steam, waste, CO2 or hydrogen), for which the well construction and completion processes are similar to those for hydrocarbons, and the implementations contained herein also apply.

[0041] Figure 2 illustrates a conceptual system diagram 200 for generating petrophysical interpretations 206 using a generative machine learning model 204, according to some aspects of the disclosed technology. A petrophysical interpretation 206 may include the analysis of physical and chemical properties of rocks and the fluids they contain, usually from well log data (e.g., as illustrated by the measurement data 202 in Figure 2).

[0042] The input or measurement data (or measurements) 202 for the generative machine learning model 204 may include well log data, which may include a detailed record of the geological formations penetrated by a wellbore. Examples of well log data may include, but are not limited to, resistivity records, gamma ray records, porosity records, sonic / acoustic records, NMR records, and / or formation pressure records, etc. Measurement data 202 may also include other types of data such as core samples, seismic data, production rates, fluid flow, pressure data, and geological information. Those skilled in the art will recognize other types of Petition 870250088766, dated 09 / 30 / 2025, page 23 / 66 / 36 measurement data 202 that can be inserted into the generative machine learning model 204, without leaving the scope of the disclosed technology.

[0043] By way of example, petrophysical interpretation data 206 may include data corresponding to properties and / or characteristics associated with underground rock formations, such as rock types, porosity, permeability, saturation, lithology, formation pressure and fluid types. Experts in the field will appreciate additional properties and / or characteristics corresponding to petrophysical interpretation data 206.

[0044] As discussed above, the types of generative machine learning models 204 may include, but are not limited to, GANs, VAEs, FBGMs, Naive Bayes classifiers, GMMs, HMMs, and RBMs. Those versed in the technique will appreciate other types of generative machine learning models 204.

[0045] The use of a generative machine learning model 204 can enable a fast, accurate, and consistent interpretation of large-scale petrophysical data that may not be feasible using conventional approaches. For example, the generative machine learning model 204 can use measurement data 202 for data augmentation and generate new data samples by learning underlying patterns and relationships in the data. In another example, the generative machine learning model 204 can be trained on available data and then used to generate additional synthetic data (e.g., well log data) that can mimic real-world scenarios. Furthermore, the generative machine learning model 204 can fill data gaps based on patterns learned from training data. The generative machine learning model 204 can also simulate different subsurface properties that may change under various conditions.

[0046] The petrophysical interpretation 206 may depend on the type of Petition 870250088766, dated 09 / 30 / 2025, page 24 / 66 / 36 measurement data 202 provided to the generative machine learning model. For example, four application scenarios could include a lithological interpretation, reservoir characterization, geomechanical interpretation, and a joint interpretation based on petrophysics and geophysics.

[0047] In a lithological interpretation, the inputs (e.g., measurement data 202) could include physical properties of the rocks in the subsurface, such as gamma ray logging and spontaneous potential data. The output (e.g., petrophysical interpretation 206) could include descriptions of different rock types present in the formation, which could be classified according to lithology, such as mineral composition, grain size, texture, etc.

[0048] In a reservoir characterization, measurement data 202 may include well logs, core samples, seismic data, production rates, fluid flow, and pressure data. The corresponding petrophysical interpretation 206 may include: (1) Geological features (e.g., structural features such as faults, folds and fractures, as well as stratigraphic features such as lithology, porosity, permeability and thickness).

[0049] (2) Petrophysical properties (e.g., rock properties such as porosity, permeability and fluid saturation, which may be critical for estimating the reservoir's capacity to store and transmit fluids).

[0050] (3) Fluid properties (e.g., fluid properties such as viscosity, density, and composition, which may be critical for predicting fluid flow behavior in the reservoir).

[0051] (4) Three-dimensional (3D) models and maps (e.g., a visual representation of the subsurface reservoir, including the distribution of geological features, petrophysical properties and fluid properties).

[0052] (5) Production forecasts (e.g., estimates of rates Petition 870250088766, dated 09 / 30 / 2025, page 25 / 66 / 36 (regarding production and expected reservoir volumes in different production scenarios).

[0053] (7) Risk assessments (e.g., assessments of the uncertainties and risks associated with the reservoir, such as the probability of encountering unexpected geological features or variations in fluid properties).

[0054] (8) Recommendations (e.g., suggested courses of action based on reservoir characterization results, such as where to drill new wells, how to improve production from existing wells, and how to manage the reservoir over time).

[0055] In a geomechanical interpretation, measurement data 202 may include well logs, core samples, seismic data, and geological information. The corresponding petrophysical interpretation 206 may include: (1) Maps and cross-sections of the geomechanical properties of rocks.

[0056] (2) Quantitative estimates of the mechanical properties of rocks, such as strength, elasticity and ductility.

[0057] (3) Identification of potential areas of subsurface instability, such as areas with a high risk of rock failure, fault reactivation or formation damage.

[0058] (4) Recommendations to mitigate the risks of subsurface instability, such as modifications to drilling and completion practices, or changes to the overall development plan.

[0059] In a petrophysical and geophysical interpretation, measurement data 202 may include well logs, core samples, seismic data, and geological information. The corresponding petrophysical interpretation 206 may include: (1) 3D models and maps of subsurface geology and Petition 870250088766, dated 09 / 30 / 2025, page 26 / 66 / 36 reservoir properties, which can be used to guide drilling and development decisions.

[0060] (2) Quantitative estimates of the petrophysical properties of rock formations, such as porosity, permeability and fluid saturation, which can be used to estimate the reservoir's capacity to store and transmit fluids.

[0061] (3) Identification of underground structures, such as faults and fractures, that may impact reservoir behavior.

[0062] (4) Recommendations for optimizing production, such as recommendations on where to drill new wells or how to optimize production from existing wells.

[0063] The four application scenarios discussed above can use various combinations of generative machine learning models 204 (e.g., GANs, VAEs, FBGMs, Naive Bayes classifiers, GMMs, HMMs, RBMs) applicable to each respective scenario.

[0064] A VAE may need to be trained to learn a compact, structured representation of input data (e.g., measurement data 202) in its latent space. During training, a VAE learns to encode the input data in this latent space and then decode it back into its original form.

[0065] Figure 3 illustrates an example of system 300 that includes a Variational Autoencoder (VAE), according to some aspects of the disclosed technology. In some aspects, system 300 can be used to train a VAE. For example, the VAE can receive training measurement data as input X 302 (e.g., well logs, core samples, seismic data) and pass it through the encoder 304, which can translate the input X 302 into a distribution over latent space, z (e.g., Q(z|X) as illustrated in Figure 3). In some cases, after training a VAE, the latent space data can be Petition 870250088766, dated 09 / 30 / 2025, page 27 / 66 / 36 labeled and / or distributed in groups (for example, in clusters), which will be discussed in Figure 4 below.

[0066] In a VAE, instead of producing a fixed vector for each input X 302, the encoder 304 can generate parameters that define a probability distribution of possible latent vectors. For example, representing data in latent space as distributions can account for variability and uncertainty in latent space. Then, the sampled latent vector 306 (e.g., z is sampled from Q(z|X) denoted as z~ Q(z|X)) can be randomly drawn or sampled from the distribution of possible latent vectors. In other words, feeding a VAE with the same input X 302 multiple times can result in a different sampled latent vector, z 306, each time.

[0067] The sampled latent vector 306 can then be transmitted to the decoding network 308, which can attempt to reconstruct the input X 302 based on the sampled latent vector 306 (e.g., z~Q(z|X)) and produce the Output X' 310. P(X|z) can represent the probability distribution of reconstructing the original input data X 302 given a specific sampled latent vector, z 306. In other words, given a sampled latent vector 306, there can be multiple reconstructions of the input X 302, and P(X|z) can provide the probability of each potential reconstruction.

[0068] After the 304 encoder, latent space, and 308 decoder are well constructed (e.g., after training), clustering and labeling can be performed on the vectors in the latent space, as illustrated in Figure 4.

[0069] Figure 4 illustrates a 400 visualization for latent space data clustering, according to some aspects of the disclosed technology. After the encoder, latent space, and decoder, as discussed above in Figure 3, are well constructed or well trained, clustering algorithms can be applied to the latent space, z for Petition 870250088766, dated 09 / 30 / 2025, page 28 / 66 / 36, group similar points into distinct groupings. As illustrated in Figure 4, similar rock types or properties can be grouped as designated by types 1 to 9 (e.g., limestone, sandstone, shale, etc.).

[0070] Examples of clustering algorithms that can be applied to latent space data may include, but are not limited to, K-means clustering, Density-Based Spatial Clustering with Noise (DBSCAN), hierarchical clustering, Gaussian Mixture Model (GMM) clustering, spectral clustering, mean shift clustering, sorting points to identify clustering structure (OPTICS), and Balanced Iterative Reduction and Clustering Using Hierarchies (BIRCH). Those skilled in the art will appreciate additional examples of clustering algorithms that can be applied to the latent space of a generative machine learning model.

[0071] Figure 5 illustrates an example of a 500 workflow that can be used to perform latent space clustering using a VAE, according to some aspects of the disclosed technology. As an example, consider a lithological interpretation, where the X 502 input represents raw measurement data acquired from subsurface rocks, such as gamma ray logging and spontaneous potential readings. For example, these measurements can provide information about the physical properties of subsurface formations.

[0072] The Variational Autoencoder (VAE) can process X502 input data and compress it into a lower-dimensional VAE 504 latent space. For example, the VAE 504 latent space can include X502 input data features in a more compact format. In the context of lithological interpretation, the VAE can be trained to capture the most relevant features of different rock types from Petition 870250088766, dated 09 / 30 / 2025, page 29 / 66 / 36 input measurements X 502.

[0073] After the data are transformed into the VAE 504 latent space, the 506 clustering model can apply clustering algorithms to group similar data points. These clusters can represent distinct rock types or lithologies. By analyzing the patterns and similarities in the VAE 504 latent space, the 506 clustering model can identify which measurements likely belong to the same rock type.

[0074] After clustering, each cluster can be associated with a specific rock type or lithological description, as shown in output Y 508. For example, a cluster can be identified as limestone based on its gamma ray and spontaneous potential readings. This step can translate the clusters in the latent space of VAE 504 into meaningful lithological descriptions and provide a clear interpretation of subsurface formations.

[0075] In addition, with the exception of the above grouping, VAE can also be used for regression, to map one input to another output, such as using resistivity, acoustic velocity P and velocity S to obtain porosity.

[0076] The process described above can also be used for other scenarios, including, but not limited to, reservoir characterization, geomechanical interpretation, and a joint interpretation based on petrophysics and geophysics, as discussed above in Figure 2.

[0077] Figure 6 illustrates an example of workflow 600 for using a generative machine learning model to perform latent space clustering, according to some aspects of the disclosed technology. As an example, for a lithological interpretation, the X 602 input could represent raw measurement data collected from subsurface rocks, as discussed above in relation to Figure 5. Petition 870250088766, dated 09 / 30 / 2025, page 30 / 66 / 36

[0078] The flow-based generative model, as illustrated in Figure 6, can be a type of deep learning architecture that can model complex distributions by transforming a simple base distribution (e.g., Gaussian distribution) into the target distribution using a series of invertible mappings. For example, a flow-based generative model can be used to convert input data into a latent space representation 604 of the flow-based generative model. The latent space of the flow-based generative model 604 can act as a compressed or encoded version of the input data X 602. The clustering model 606 can categorize similar data points in latent space 604 into clusters, and the output Y 608 of the workflow 600 can be the result of the clustering model.

[0079] In some respects, since a flow-based generative model is strictly reversible, this means that the input and output are the same as in theory, and it cannot be used in regression.

[0080] Figure 7 illustrates a workflow 700 for performing latent space clustering using a VAE and a generative machine learning model, according to some aspects of the disclosed technology.

[0081] There may be some gaps in the latent space constructed by the VAE. In some cases, gaps may cause some uncertainty in the predicted results. The flow-based generative model can be used together with a VAE to generate the latent space and obtain a fully covered latent space. The clustering model can be applied to generate the output. The example above may not only include flow-based generative models, but may also include other generative models, such as a restricted Boltzmann machine (RBM), in which the hidden layer can act as the latent space. Similar approaches can also be used for clustering or classification. Petition 870250088766, dated 09 / 30 / 2025, page 31 / 66 / 36

[0082] Figure 8 illustrates an example of a generative adversarial network (GAN) system, according to some aspects of the disclosed technology. The GAN includes perhaps two models, a generator and a discriminator, which can be trained together. For example, the generator can generate a batch of samples, and these, along with real examples from the domain, can be fed to the discriminator and classified as real or fake. The discriminator can then be updated to improve the discrimination of real and fake samples in the next round, and the generator can be updated based on how well, or not, the generated samples fooled the discriminator.

[0083] In some respects, there are two scenarios for applying a GAN. In a first scenario, when there is not enough input data for training, a GAN can be used to generate data samples that capture the characteristics of the input data to compensate for the lack of data samples. In a second scenario, the discriminator of a GAN can be used for classification.

[0084] Figure 9 illustrates a 900 system diagram that uses GANs to perform lithographic classification, according to some aspects of the disclosed technology. The 900 system diagram, as illustrated in Figure 9, shows an example of a GAN application for rock type classification. Different GAN models can be trained for different rock types. Furthermore, each GAN can be used independently for the identification of a specific rock type.

[0085] Figure 10 illustrates an example of a workflow 1000 to implement a generative model with an uncertain number of inputs, according to some aspects of the disclosed technology. In some cases, the number of inputs for a generative model 1004 may be fixed. In petrophysical interpretation, some logging data may not be available at certain depths. As a result, there may be a Petition 870250088766, dated 09 / 30 / 2025, page 32 / 66 / 36 different number of inputs for different depths. An adaptive layer 1002 can be used before the input data enters the generative model 1004. As illustrated in Figure 10, the full number of inputs is N and the number of inputs is n (n<= N). The adaptive layer 1002 can map n inputs to N inputs. For missing inputs, default values ​​can be filled in with the adaptive layer 1002. The adaptive layer 1002 can ensure that the number of inputs to the generative model 1004 is a fixed number and that a trained generative model 1004 does not need to be retrained due to some missing data.

[0086] Figure 11 illustrates an example of a process 1100 for using a generative machine learning model to generate a petrophysical interpretation, according to some aspects of the disclosed technology. In step 1102, process 1100 includes providing a set of formation measurement data to a generative machine learning model. For example, formation measurement data (e.g., measurement data 202 as illustrated in Figure 2) may include data obtained from various tools and methods used to investigate and analyze subsurface rock formations that can help characterize the rock properties, fluid content, and overall structure of subsurface formations. Formation data may include well logs, core samples, seismic data, production rates, fluid flow, pressure data, and geological information.Those skilled in the art may appreciate other types of formation data that can be used for petrophysical interpretation.

[0087] In block 1104, process 1100 includes the generation, using the generative machine learning model, of a latent space dataset corresponding to the formation measurement dataset. For example, the generative machine learning model may include GANs, VAEs, FBGMs, Naive Bayes classifiers, GMMs, Petition 870250088766, dated 09 / 30 / 2025, page 33 / 66 / 36 HMMs and RBMs. Latent space data can be generated by the generative machine learning model (e.g., an encoder) as a representation of the formation measurement dataset.

[0088] In block 1106, process 1100 includes clustering the latent space dataset to generate a clustering set. For example, a clustering model can generate a clustering set by identifying and grouping closely related points in the latent space data. For example, for a lithological interpretation application with an effectively trained VAE model, the clustering model can group similar rock types.

[0089] In block 1108, process 1100 includes the determination of a petrophysical interpretation based on the cluster set. For example, in a lithological interpretation application, the petrophysical interpretation might include descriptions of different rock types (e.g., shale, limestone, sandstone) based on or corresponding to the cluster set. In other words, in this example, each cluster in the cluster set might be associated with a specific rock type.

[0090] Figure 12 is an example of a machine learning (also neural network) architecture 1200 that can be used to implement all or part of the systems and techniques described here (e.g., the neural network 1200 can be used to implement a symbolic regression model, as discussed above). The neural network 1200 includes several hidden layers 1222a, 1222b through 1222n.The hidden layers 1222a, 1222b, up to 1222n include a number “n” of hidden layers, where “n” is an integer greater than or equal to one. The number of hidden layers can be adjusted to include as many layers as needed for the given application. The 1200 neural network also includes an output layer 1221 that provides an output resulting from the processing performed by the hidden layers 1222a, 1222b up to 1222n. Petition 870250088766, dated 09 / 30 / 2025, page 34 / 66 / 36

[0091] The 1200 neural network is a multilayer neural network of interconnected nodes. Each node can represent a piece of information. The information associated with the nodes is shared between the different layers, and each layer retains the information as it is processed. In some cases, the 1200 neural network may include a feed-forward network, in which case there are no feedback connections where the network outputs are fed back. In some cases, the 1200 neural network may include a recurrent neural network, which may have loops that allow information to be transported through the nodes during input reading.

[0092] Information can be exchanged between nodes through node-to-node interconnections between the various layers. Nodes in the input layer 1220 can activate a set of nodes in the first hidden layer 1222a. For example, as shown, each of the input nodes in input layer 1220 is connected to each of the nodes in the first hidden layer 1222a. The nodes in the first hidden layer 1222a can transform the information from each input node by applying activation functions to the input node information. The information derived from the transformation can then be passed and can activate nodes in the next hidden layer 1222b, which can execute their own designated functions. Example functions include convolution, upsampling, data transformation, and / or any other suitable functions. The output from hidden layer 1222b can then activate nodes in the next hidden layer, and so on.The output of the last hidden layer 1222n can activate one or more nodes of the output layer 1221, in which an output is provided. In some cases, although the nodes in the neural network 1200 are shown as having multiple output lines, a node may have a single output and all lines shown as output from a node represent the same output value.

[0093] In some cases, each node or interconnection between nodes may have a weight which is a set of parameters derived from the network training. Petition 870250088766, dated 09 / 30 / 2025, page 35 / 66 / 36 neural 1200. After the neural network 1200 is trained, it can be called a trained neural network, which can be used to classify one or more activities. For example, an interconnection between nodes can represent learned information about the interconnected nodes. The interconnection can have an adjustable numerical weight that can be adjusted (e.g., based on a training dataset), allowing the neural network 1200 to be adaptable to inputs and able to learn as more and more data is processed.

[0094] The neural network 1200 is pre-trained to process the data features in the input layer 1220 using the different hidden layers 1222a, 1222b up to 1222n to provide the output through the output layer 1221.

[0095] In some cases, the 1200 neural network can adjust node weights using a training process called backpropagation. A backpropagation process may include a forward pass, a loss function, a backward pass, and a weight update. The forward pass, loss function, backward pass, and parameter / weight update are performed for one training iteration. The process can be repeated for a certain number of iterations for each training dataset until the 1100 neural network is trained well enough that the layer weights are accurately adjusted.

[0096] To perform the training, a loss function can be used to analyze the error in the output. Any suitable loss function definition can be used, such as a Cross Entropy loss. Another example of a loss function includes the mean squared error (MSE), defined as E_total = Σ(1 / 2 (target-output)A2). The loss can be defined as equal to the value of E_total.

[0097] The loss (or error) will be high for the initial training data, as the actual values ​​will be very different from the predicted output. The Petition 870250088766, dated 09 / 30 / 2025, page 36 / 66 / 36 The objective of the training is to minimize the amount of loss so that the predicted output is the same as the training output. The 1200 neural network can perform a backward pass, determining which inputs (weights) contributed most to the network's loss, and can adjust the weights so that the loss decreases and is eventually minimized.

[0098] The 1200 neural network can include any suitable deep network. An example includes a Convolutional Neural Network (CNN), which includes an input layer and an output layer, with multiple hidden layers between the input and output layers. The hidden layers of a CNN include a sequence of convolutional, non-linear, downsampling (for subsampling), and fully connected layers. The 1200 neural network can include any other deep network besides a CNN, such as an autoencoder, Deep Belief Networks (DBNs), Recurrent Neural Networks (RNNs), among others.

[0099] As understood by those skilled in the art, machine learning-based classification techniques can vary depending on the desired implementation. For example, machine learning classification schemes may utilize one or more of the following, alone or in combination: hidden Markov models; RNNs; CNNs; deep learning; Bayesian symbolic methods; Generative Adversarial Networks (GANs); support vector machines; picture registration methods; and enforceable rule-based systems. When regression algorithms are used, they may include, but are not limited to: a Stochastic Gradient Descent Regressor, a Passive-Aggressive Regressor, etc.

[00100] Machine learning classification models can also be based on clustering algorithms (e.g., a K-average Mini-batch clustering algorithm), a recommendation algorithm (e.g., a Minwise Hashing algorithm or a Petition 870250088766, dated 09 / 30 / 2025, page 37 / 66 / 36 Euclidean Locality-Sensitive Hashing (LSH) algorithm) and / or an anomaly detection algorithm, as a local outlier factor. Additionally, machine learning models may employ a dimensionality reduction approach, such as one or more of: a Mini-batch Dictionary Learning algorithm, an Incremental Principal Component Analysis (PCA) algorithm, a Latent Dirichlet Allocation algorithm, and / or a Mini-batch K-meaning algorithm, etc.

[00101] Figure 13 illustrates an example of a processor-based system with which some aspects of the technology in question can be implemented. For example, the 1300 processor-based system can be any computing device that comprises it, or any component thereof, in which the system components communicate with each other using the 1305 connection. The 1305 connection can be a physical connection via a bus or a direct connection to the 1310 processor, as in a chipset architecture. The 1305 connection can also be a virtual connection, a network connection, or a logical connection.

[00102] In some embodiments, the 1300 computing system is a distributed system in which the functions described in this disclosure may be distributed within a data center, across multiple data centers, on a peer-to-peer network, etc. In some embodiments, one or more of the described system components represent many such components, each performing part or all of the function for which the component is described. In some embodiments, the components may be physical or virtual devices.

[00103] The example system 1300 includes at least one processing unit (Central Processing Unit (CPU) or processor) 1310 and connection 1305 that couples various system components, including system memory 1315, such as Read-Only Memory (ROM) 1320 and Petition 870250088766, dated 09 / 30 / 2025, page 38 / 66 / 36 Random Access Memory (RAM) 1325 to processor 1310. The computing system 1300 may include a high-speed memory cache 1312 connected directly, near, or integrated as part of the processor 1310.

[00104] The 1310 processor may include any general-purpose processor and a hardware or software service, such as the 1332, 1334, and 1336 services stored in the 1330 storage device, configured to control the 1310 processor, as well as a special-purpose processor where software instructions are incorporated into the actual processor design. The 1310 processor may essentially be a completely self-contained computing system, containing multiple cores or processors, a bus, memory controller, cache, etc. A multi-core processor may be symmetric or asymmetric.

[00105] To enable user interaction, the computing system 1300 includes an input device 1345, which may represent any number of input mechanisms, such as a microphone for speech, a touch screen for gesture or graphic input, keyboard, mouse, motion input, speech, etc. The computing system 1300 may also include the output device 1335, which may be one or more of a number of output mechanisms known to those skilled in the art. In some cases, multimodal systems may allow a user to provide multiple types of input / output to communicate with the computing system 1300. The computing system 1300 may include the communication interface 1340, which may generally govern and manage user input and system output.The communication interface can perform or facilitate the reception and / or transmission of wired or wireless communications via wired and / or wireless transceivers, including those using an audio connector / plug, a microphone connector / plug, a Universal Serial Bus (USB) port / plug, an Apple® Lightning® port / plug, or a port / plug. Petition 870250088766, dated 09 / 30 / 2025, p. 39 / 66 / 36 Ethernet, a fiber optic port / plug, a proprietary wired port / plug, a BLUETOOTH® wireless signal transfer, a low-energy (BLE) BLUETOOTH® wireless signal transfer, an IBEACON® wireless signal transfer, a Radio Frequency Identification (RFID) wireless signal transfer, a Near Field Communication (NFC) wireless signal transfer, a Dedicated Short Range Communication (DSRC) wireless signal transfer, a Wi-Fi® 802 wireless signal transfer.11, a Wireless Local Area Network (WLAN) signal transfer, a Visible Light Communication (VLC) signal transfer, a Worldwide Interoperability for Microwave Access (WiMAX) signal transfer, an infrared (IR) communication wireless signal transfer, a Public Switched Telephone Network (PSTN) signal transfer, an Integrated Services Digital Network (ISDN) signal transfer, a 3G / 4G / 5G / LTE cellular data network wireless signal transfer, an ad-hoc network signal transfer, a radio wave signal transfer, a microwave signal transfer, an infrared signal transfer, a visible light signal transfer, an ultraviolet light signal transfer, a wireless signal transfer across the electromagnetic spectrum, or some combination thereof.

[00106] The 1340 communication interface may also include one or more Global Navigation Satellite System (GNSS) receivers or transceivers that are used to determine a location of the 1300 computing system based on the reception of one or more signals from one or more satellites associated with one or more GNSS systems. GNSS systems include, but are not limited to, the U.S. Global Positioning System (GPS), the Russian Global Navigation Satellite System (GLONASS), the Chinese BeiDou Navigation Satellite System (BDS), and the European Galileo GNSS. There is no restriction regarding operation in any particular hardware arrangement and therefore the basic characteristics Petition 870250088766, dated 09 / 30 / 2025, page 40 / 66 / 36 described in this document can be easily replaced by improved hardware or firmware arrangements as they are developed.

[00107] Storage device 1330 may be a non-volatile and / or non-transient and / or computer-readable memory device and may be a hard disk or other types of computer-readable media that can store data that is accessible by a computer, such as magnetic cassettes, flash memory cards, solid-state memory devices, digital versatile disks, cartridges, a floppy disk, a floppy disk, a hard disk, magnetic tape, a magnetic stripe / band, any other magnetic storage medium, flash memory, memristor memory, any other solid-state memory, a read-only memory optical disc (CD-ROM), a rewritable CD optical disc, a digital video disc (DVD) optical disc, a Blu-ray Disc (BD) optical disc, a holographic optical disc, other optical media, a Secure Digital (SD) card, a micro SD (microSD) card, a Memory Stick® card, a smart card chip, an EMV chip,a Subscriber Identity Module (SIM) card, a mini / micro / nano / pico SIM card, another Integrated Circuit (IC) chip / card, Random Access Memory (RAM), Active RAM (SRAM), Dynamic RAM (DRAM), Read-Only Memory (ROM), Programmable ROM (PROM), Erasable PROM (EPROM), Electrically Erasable PROM (EEPROM), flash EPROM (FLASHEPROM), cache memory (L1 / L2 / L3 / L4 / L5 / L#), Resistive RAM (RRAM / ReRAM), Phase Change Memory (PCM), Spin Transfer Torque RAM (STTRAM), another memory chip or cartridge and / or a combination thereof.

[00108] The storage device 1330 may include software services, servers, services, etc., which, when the code defining such software is executed by the processor 1310, causes the system 1300 to perform a function. In some embodiments, a hardware service that performs a specific function may include the software component. Petition 870250088766, dated 09 / 30 / 2025, page 41 / 66 / 36 stored on a computer-readable medium in connection with the necessary hardware components, such as processor 1310, connection 1305, output device 1335, etc., to perform the function.

[00109] Modalities within the scope of this disclosure may also include tangible and / or non-transient computer-readable storage media or devices for carrying or having computer-executable instructions or data structures stored thereon. Such tangible computer-readable storage devices may be any available device that can be accessed by a general-purpose or special-purpose computer, including the functional design of any special-purpose processor as described above.By way of example, and not limitation, such tangible computer-readable devices may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other device that can be used to transport or store the desired program code in the form of computer-executable instructions, data structures, or processor chip design. When information or instructions are provided via a network or other communications connection (whether wired, wireless, or a combination thereof) to a computer, the computer properly views the connection as a computer-readable medium. Thus, any such connection is properly termed a computer-readable medium. Combinations of the foregoing should also be included within the scope of computer-readable devices.

[00110] Computer executable instructions include, for example, instructions and data that cause a general-purpose computer, a special-purpose computer, or a special-purpose processing device to perform a particular function or group of functions. Computer executable instructions also include modules. Petition 870250088766, dated 09 / 30 / 2025, page 42 / 66 / 36 of programs that are executed by computers in autonomous or networked environments. Generally, program modules include routines, programs, components, data structures, objects, and functions inherent in the design of special-purpose processors, etc., that perform tasks or implement abstract data types. Computer executable instructions, associated data structures, and program modules represent examples of the means of program code to execute steps of the methods disclosed in this document. The particular sequence of such executable instructions or associated data structures represents examples of corresponding actions to implement the functions described in such steps.

[00111] Other forms of dissemination can be practiced in networked computing environments with many types of computer system configurations, including personal computers, portable devices, multiprocessor systems, programmable or microprocessor-based consumer electronics, networked Personal Computers (PCs), minicomputers, mainframes, and the like. Forms can also be practiced in distributed computing environments where tasks are performed by local and remote processing devices that are connected (either by wired links, wireless links, or a combination thereof) through a communications network. In a distributed computing environment, program modules can be located on both local and remote memory storage devices. SELECTED EXAMPLES

[00112] Illustrative examples of disclosure include: Aspect 1. A device comprising: at least one memory; and at least one processor coupled to at least one memory, or at least one processor configured to: provide a set of training measurement data to a machine learning model. Petition 870250088766, dated 09 / 30 / 2025, p. 43 / 66 / 36 generative; generate, using the generative machine learning model, a latent space dataset corresponding to the formation measurement dataset; cluster the latent space dataset to generate a cluster set; and determine a petrophysical interpretation based on the cluster set.

[00113] Aspect 2. The apparatus of Aspect 1, in which the generative machine learning model is a Variational Autoencoder (VAE).

[00114] Aspect 3. The apparatus of either of Aspects 1-2, where the generative machine learning model is a flow-based generative model (FBGM).

[00115] Aspect 4. The apparatus of any of Aspects 1-3, wherein the generative machine learning model is a generative adversarial network (GAN).

[00116] Aspect 5. The apparatus of any of Aspects 1 to 4, wherein the formation measurement dataset comprises at least one of the following: well logs, core samples, seismic data, production rates, fluid flow, pressure data, geological information or a combination thereof.

[00117] Aspect 6. The apparatus of any of Aspects 1 to 5, wherein the petrophysical interpretation comprises at least one of the following rock types: porosity, permeability, saturation, lithology, formation pressure, fluid types, or a combination thereof.

[00118] Aspect 7. The apparatus of Aspect 2, in which the VAE comprises a network of encoders, a sampled latent vector and a network of decoders.

[00119] Aspect 8. A computer-implemented method comprising: providing a set of formation measurement data to a generative machine learning model; generating, by means of the model, Petition 870250088766, dated 09 / 30 / 2025, p. 44 / 66 / 36 generative machine learning, a latent space dataset corresponding to the formation measurement dataset; cluster the latent space dataset to generate a cluster set; and determine a petrophysical interpretation based on the cluster set.

[00120] Aspect 9. The computer-implemented method of Aspect 8, where the generative machine learning model is a variational autoencoder (VAE).

[00121] Aspect 10. The computer-implemented method of any of Aspects 8 to 9, wherein the generative machine learning model is a flow-based generative model (FBGM).

[00122] Aspect 11. The computer-implemented method of any of Aspects 8 to 10, wherein the generative machine learning model is a generative adversarial network (GAN).

[00123] Aspect 12. The computer-implemented method of any of Aspects 8 to 11, wherein the formation measurement dataset comprises at least one of the following: well logs, core samples, seismic data, production rates, fluid flow, pressure data, geological information, or a combination of these data.

[00124] Aspect 13. The computer-implemented method of any of Aspects 8 to 12, wherein the petrophysical interpretation comprises at least one of the following: rock type, porosity, permeability, saturation, lithology, formation pressure, fluid type, or a combination thereof.

[00125] Aspect 14. The computer-implemented method of Aspect 9, in which the VAE comprises an encoding network, a sampled latent vector, and a decoding network.

[00126] Aspect 15. A non-transient, computer-readable storage medium comprising at least one instruction to make Petition 870250088766, dated 09 / 30 / 2025, page 45 / 66 / 36, whereby a computer or processor: provides a set of formation measurement data to a generative machine learning model; generates, through the generative machine learning model, a set of latent space data corresponding to the set of formation measurement data; clusters the latent space data to generate a set of clusters; and determines a petrophysical interpretation based on the set of clusters.

[00127] Aspect 16. The non-transient, machine-readable storage medium of Aspect 15, where the generative machine learning model is a Variational Autoencoder (VAE).

[00128] Aspect 17. The non-transient, machine-readable storage medium of any of Aspects 15 to 16, wherein the generative machine learning model is a stream-based generative model (FBGM).

[00129] Aspect 18. The non-transient, computer-readable storage medium of any of Aspects 15 to 17, wherein the generative machine learning model is a generative adversarial network (GAN).

[00130] Aspect 19. The non-transient, computer-readable storage medium of any of Aspects 15 to 18, wherein the formation measurement dataset comprises at least one of the following: well logs, core samples, seismic data, production rates, fluid flow, pressure data, geological information, or a combination thereof.

[00131] Aspect 20. The non-transient, computer-readable storage medium of any of Aspects 15 to 19, wherein the petrophysical interpretation comprises at least one of the following: rock type, porosity, permeability, saturation, lithology, formation pressure, fluid type, or a combination thereof. Petition 870250088766, dated 09 / 30 / 2025, page 46 / 66 / 36

[00132] The various embodiments described above are provided for illustrative purposes only and should not be construed to limit the scope of the disclosure. For example, the principles in this document apply equally to optimization as well as general improvements. Various modifications and changes may be made to the principles described in this document without following the embodiments and example applications illustrated and described in this document, and without departing from the spirit and scope of the disclosure. Claim language reciting “at least one of” a set indicates that one member of the set or multiple members of the set satisfy the claim.

[00133] The claim language or other language in the disclosure stating “at least one of” a set and / or “one or more” of a set indicates that one member of the set or several members of the set (in any combination) satisfy the claim. For example, the claim language stating “at least one of A and B” or “at least one of A or B” means A, B, or A and B. In another example, the claim language stating “at least one of A, B, and C” or “at least one of A, B, or C” means A, B, C, or A and B, or A and C, or B and C, or A and B and C. The language “at least one of” a set and / or “one or more” of a set does not limit the set to the items listed in the set. For example, the claim language stating “at least one of A and B” or “at least one of A or B” may mean A, B, or A and B and may additionally include items not listed in the set of A and B. Petition 870250088766, dated 09 / 30 / 2025, pp. 47 / 66

Claims

1 / 4 CLAIMS 1. Apparatus, characterized in that it comprises: at least one memory; and at least one processor coupled to at least one memory, or at least one processor configured to: provide a set of formation measurement data to a generative machine learning model; generate, by means of the generative machine learning model, a set of latent space data corresponding to the set of formation measurement data; cluster the set of latent space data to generate a set of clusters; and determine a petrophysical interpretation based on the set of clusters.

2. Apparatus according to claim 1, characterized in that the generative machine learning model is a variational autoencoder (VAE).

3. Apparatus according to claim 1, characterized in that the generative machine learning model is a flow-based generative model (FBGM).

4. Device according to claim 1, characterized in that the generative machine learning model is a generative adversarial network (GAN).

5. Apparatus according to claim 1, characterized in that the formation measurement data set comprises at least one of the following: well logs, core samples, seismic data, production rates, fluid flow, pressure data, geological information, or a combination of these data.

6. Apparatus according to claim 1, characterized by the fact that the petrophysical interpretation comprises at least one of the following: rock type, porosity, permeability, saturation, lithology, formation pressure, fluid type, or a combination of these elements.

7. Apparatus according to claim 2, characterized in that the VAE comprises a network of encoders, a sampled latent vector and a network of decoders.

8. A computer-implemented method, characterized in that it comprises: providing a set of formation measurement data to a generative machine learning model; generating, by means of the generative machine learning model, a latent space dataset corresponding to the formation measurement dataset; clustering the latent space dataset to generate a set of clusters; and determining a petrophysical interpretation based on the set of clusters.

9. A computer-implemented method according to claim 8, characterized in that the generative machine learning model is a variational autoencoder (VAE).

10. A computer-implemented method according to claim 8, characterized in that the generative machine learning model is a flow-based generative model (FBGM).

11. A computer-implemented method according to claim 8, characterized in that the generative machine learning model is a generative adversarial network (GAN).

12. Computer-implemented method according to claim 8, characterized in that the formation measurement dataset comprises at least one of the following: well logs, core samples, seismic data, production rates, fluid flow, pressure data, geological information, or a combination of these data.

13. A computer-implemented method according to claim 8, characterized in that the petrophysical interpretation comprises at least one of the following: rock type, porosity, permeability, saturation, lithology, formation pressure, fluid type, or a combination thereof.

14. Computer-implemented method according to claim 9, characterized in that the VAE comprises a network of encoders, a sampled latent vector, and a network of decoders.

15. A non-transient, computer-readable storage medium, characterized in that it includes at least one instruction to cause a computer or processor to: provide a set of formation measurement data to a generative machine learning model; generate, through the generative machine learning model, a set of latent space data corresponding to the set of formation measurement data; cluster the latent space data to generate a set of clusters; and determine a petrophysical interpretation based on the set of clusters.

16. A non-transient, computer-readable storage medium according to claim 15, characterized in that the generative machine learning model is a variational autoencoder (VAE).

17. Non-transient, computer-readable storage medium according to claim 15, characterized in that the generative machine learning model is a flow-based generative model (FBGM).

18. A non-transient, computer-readable storage medium according to claim 15, characterized in that the generative machine learning model is a generative adversarial network (GAN).

19. A computer-readable non-transient storage medium according to claim 15, characterized in that the formation measurement dataset comprises at least one of the following: well logs, core samples, seismic data, production rates, fluid flow, pressure data, geological information, or a combination thereof.

20. A computer-readable non-transient storage medium according to claim 15, characterized in that the petrophysical interpretation comprises at least one of the following types of rock, porosity, permeability, saturation, lithology, formation pressure, fluid types, or a combination thereof. Petition 870250088766, dated 09 / 30 / 2025, pp. 51 / 66

Images