Enhanced intelligent (i) driven missing reserve opportunity identification

Abstract

A method, computer system, and computer program product for identifying missing reserves in a reservoir are provided. Well site data of a well site in the reservoir is ingested. A first machine learning model generates post-casing opportunities and reservoir quality indicators from the plurality of logs. The second machine learning model determines missing reserves based on reservoir quality indicators between the wellsite and the well. A third machine learning model determines candidate wells based on missing reserves. A fourth machine learning model predicts economic outcomes of intervention options for candidate wells. And oil field decision making is supported based on the predicted economic result.

Description

[0001] Cross - Reference to Related Applications

[0002] This application is a non - provisional application of U.S. Provisional Application 63 / 409,112, filed on September 22, 2022, and claims the benefit thereof, the entire content of which is incorporated herein by reference. Background Art

[0003] Reserves are the amount of oil expected to be commercially recovered from a known reservoir starting from a given date. The identification of new reserves is an important topic for the long - term sustainable development of oil and gas enterprises. Identifying reserves involves the integration among various disciplines, such as subsurface data, petrophysical information, reservoir models, and production data. Reserve estimates will be revised as additional geological or engineering data become available or as economic conditions change.

[0004] However, due to, for example, large amounts of data, the availability of reservoir models, lack of integration, and lack of insights, traditional techniques for finding new reserves or identifying missing reserves are both labor - intensive and time - intensive. For example, large amounts of data are generated and collected during field production. The consumption and analysis of this data to identify reserve opportunities typically use manual inspection of the data, which is both resource - intensive and mundane. Traditional subsurface, reservoir, and petrophysical analyses are time - intensive and lack new insights due to manual and mundane tasks.

[0005] In addition, building an accurate reservoir model is a resource - intensive process, usually relying on the integration of multiple analysis platforms. Integration typically uses the collaboration of domain experts to work together throughout the process. Due to the lack of data flow and resource - intensive characteristics, the reservoir model is unavailable or not up - to - date, thus hindering further analysis. Summary of the Invention

[0006] Generally, in one aspect, one or more examples relate to a method for identifying missing reserves in a reservoir. The method includes ingesting well logs of wellsite data in the reservoir. The method further includes generating multiple post - casing opportunities from multiple well logs by a first machine - learning model. The method also includes determining missing reserves based on reservoir quality metrics of multiple wellsites by a second machine - learning model. The method additionally includes determining candidate wells based on the missing reserves by a third machine - learning model. The method further includes predicting economic outcomes and ranking multiple intervention options for the candidate wells by a fourth machine - learning model. The method also includes controlling oilfield decisions based on the predicted economic outcomes.

[0007] Other aspects of the invention will be apparent from the following description and the appended claims. Brief Description of the Drawings

[0008] Figure 1Depicts a cloud-based data sharing environment in which one or more embodiments can be implemented.

[0009] Figure 2 Illustrates a computing system according to one or more embodiments of the present invention.

[0010] Figure 3 Illustrates a machine learning framework according to one or more embodiments of the present invention.

[0011] Figure 4.1 、 4.2 And 4.3 illustrate a transformer architecture according to one or more embodiments of the present invention.

[0012] Figure 5 Illustrates a flowchart according to some embodiments of the present invention.

[0013] Figure 6 、 Figure 7 、 Figure 8 、 Figure 9 、 Figure 10 、 Figure 11 And Figure 12 Illustrates an example according to some embodiments.

[0014] Figure 13.1 And 13.2 Illustrates a computing system according to some illustrative embodiments.

[0015] For consistency, like elements in the various figures are denoted by like reference numerals. Detailed Description

[0016] Generally speaking, embodiments relate to decision support systems and methods for using augmented intelligence (AI)-driven technologies to identify missing reserves. Augmented intelligence is an alternative conceptualization of artificial intelligence that uses machine learning and data analysis to process large amounts of data and improve human intelligence. Illustrative embodiments use the concept of AI to accelerate the automatic identification of missing reserves. The decision support system is divided into two parts: the "planning" space and the "operation" space.

[0017] The "planning" phase integrates multiple decision-making phases for automatically and quickly identifying opportunities in missing reserves. Opportunity assessment, data discovery and validation, ML-assisted opportunity identification, and ML-based RQI (reservoir quality index) and well contribution space.

[0018] The "operation" space integrates multiple decision-making phases through an AI platform for identifying target locations, ranking and evaluating options to support expert decision-making. Data-driven target point analysis, production assessment and predictive analysis, risk / cost, and artificial intelligence for missing reserve opportunities. The solution also presents intervention opportunities that can assist the "operation" space while managing uncertainty and risk.

[0019] Figure 1 depicts a cloud-based data sharing environment in which one or more embodiments may be implemented. In one or more embodiments, one or more of the modules and elements shown may be omitted, repeated, and / or replaced Figure 1 by one or more of the modules and elements shown. Thus, embodiments should not be considered limited to Figure 1 the particular arrangement of the modules shown.

[0020] As Figure 1 shown, the data sharing environment includes remote systems (111), (113), (115), (117), data acquisition tools (121), (123), (125), (127), and a data platform (130) connected to the data acquisition tools (121), (123), (125), (127) via a communication link (132) managed by a communication repeater (134).

[0021] In one or more embodiments, the data acquisition tools (121), (123), (125), and (127) are configured to collect data. In particular, the various data acquisition tools are adapted to measure and detect physical properties of physical objects and structures. Other data may also be collected, such as historical data, analyst user input, economic information, and / or other measurement data and other parameters of interest.

[0022] In one or more embodiments, the remote systems (111), (113), (115), (117) are operatively coupled to the data acquisition tools (121), (123), (125), (127), and in particular, may be configured to send commands to the data acquisition tools and receive data from them. Thus, the remote systems (111), (113), (115), (117) may be provided with computer facilities for receiving, storing, processing, and / or analyzing data from the data acquisition tools. In one or more embodiments, the remote systems may also be provided with the mechanism of the data acquisition tools (121), (123), (125), (127) or have the function of the mechanism for actuating the data acquisition tools (121), (123), (125), (127). The data acquisition tools may be located at a physical location different from the physical location of the remote system. As an example, the location may be, for example, a processing facility location, a data center location (e.g., server farm, etc.), an oil rig location, a well site location, a wind farm, a solar farm, etc. In one or more embodiments, the remote system may then send command signals in response to the received, stored, processed, and / or analyzed data, for example, to control and / or optimize the various operations of the data acquisition tools.

[0023] In one or more embodiments, remote systems (111), (113), (115), (117) are communicatively coupled to a data platform (130) via a communication link (132). In one or more embodiments, communication between the remote systems and the data platform can be managed by a communication repeater (134). For example, a satellite, a tower antenna, or any other type of communication relay can be used to collect data from multiple remote systems and transmit the data to a remote data platform for further analysis. In one or more embodiments, the data platform is an E&P system that is configured to analyze, model, control, optimize, or perform management tasks for E&P field operations based on data provided from the remote systems. In one or more embodiments, the data platform (130) is provided with functions for manipulating and analyzing data. In one or more embodiments, the results generated by the data platform can be displayed for a user to view the results in a two-dimensional (2D) display, a three-dimensional (3D) display, or other suitable displays. Although the remote systems are shown as being separate from the Figure 1 data platform in

[0024] In one or more embodiments, the data platform (130) is implemented by deploying an application in a cloud-based infrastructure. As an example, the application can include a web application that is implemented and deployed on the cloud and accessible from a browser. Users (e.g., external clients of third parties and internal clients of the data platform) can log in to the application and execute the functions provided by the application to analyze and interpret data, including data from the remote systems (111), (113), (115), (117). The data platform (130) can correspond to a computing system, such as Figure 12 the computing systems shown in FIGS. 12.1 and 12.2 and described below.

[0025] Figure 2 A computing system (200) is shown, which can be the same as the computing system of the Figure 1 data platform (130) in Figure 12 . The hardware components of the computing system (200) are described in further detail below and in

[0026] According to an illustrative embodiment, a computing system (200) overcomes one or more challenges in identifying missing reserves by providing an integrated approach for multi-domain analysis and assessment. The system combines digital technologies such as data analysis and artificial intelligence (AI) / machine learning (ML) techniques with an expert advisory system, as compared to other known analysis techniques in traditional siloed domains. The system provides new insights for scalable opportunities across multiple platforms to optimize mature field operations with greater decision confidence in shorter decision cycle times.

[0027] In one or more embodiments of the present invention, a data repository (202) is any type of storage unit and / or device for storing data (e.g., a file system, a database, a data structure, or any other storage mechanism). Additionally, the data repository (202) can include multiple different, potentially heterogeneous storage units and / or devices.

[0028] The data repository (202) stores data related to at least one reservoir (208) served by one or more well sites (210). A well site (210) can be associated with a rig or production equipment, a wellbore, and other well site equipment configured to perform wellbore operations such as logging, drilling, fracturing, production, or other applicable operations. For example, a well site can be associated with a rig, a wellbore, and drilling equipment to perform drilling operations. A well site can be associated with surface storage tanks and / or transport pipelines to perform production operations.

[0029] The well site (210) can be operatively coupled to Figure 1 data acquisition tools (121), (123), (125), (127). In particular, Figure 1 the data acquisition tools (121), (123), (125), (127) can generate well site data (212) for E&P operations, which is stored in the data repository (202) as one or more logs (214A, 214.2, 214N).

[0030] Well site data (212) can include drilling data such as drilling rate, depth measurements, fluid properties and composition, bit and drill string parameters (e.g., torque, weight on bit), and mud logging data including gas content and lithology. Well site data (212) can include geological data such as formation evaluation data (e.g., resistivity and porosity measurements), core samples and cuttings analysis, and seismic and logging data for subsurface imaging. Well site data (212) can include pressure and temperature data such as wellbore pressure and temperature measurements during drilling and well control operations, pore pressure and fracture gradient data, and annulus pressure measurements.

[0031] Wellsite data (212) can include current production data such as flow rates (e.g., produced oil, gas, and water), wellhead pressure, and temperature. Wellsite data (212) can include production history and decline curve data such as historical production rates and volumes, including analysis of well performance over time, reservoir pressure, and productivity index, pressure drop / accumulation data, and flowmeter measurements.

[0032] Figure 2 The system shown can also include a server (204). The server (204) is one or more computing systems operating in a potentially distributed computing environment. The data repository (202) can be local to the server (204) (sharing the same physical location) or can be remote from the server (204). The server (204) can be Figure 12 .1 and Figure 12 the computing system of the computing systems shown in.2 and can include a network environment.

[0033] The server (204) includes one or more machine learning models (206). A machine learning model is a computer program that has been trained to identify certain types of patterns. There are many different types of machine learning models, but broadly, machine learning models are classified as supervised and unsupervised machine learning models, which relate to how the machine learning model is trained. A supervised machine learning model is trained based on known data that is compared to the output of the machine learning model during training. An (or more) unsupervised machine learning model is trained without known data during training. One or more embodiments can use a supervised or unsupervised machine learning model.

[0034] In an example, the machine learning model (206) is a neural network model employing a transformer architecture, such as Figures 4.1 - 4.3 shown. In this example, the machine learning model (206) is trained on the wellsite data (212). Training the (one or more) machine learning models changes the (one or more) machine learning models by changing the parameters defined for the (one or more) machine learning models. Thus, once changed, the machine learning model can be referred to as a "trained" machine learning model. The (one or more) trained machine learning models are different from the (one or more) untrained machine learning models because the training process transforms the (one or more) untrained machine learning models. Training can be an ongoing process. Thus, a trained machine learning model can be retrained and / or continuously trained. Additionally, the (one or more) untrained machine learning models can be (one or more) pre-trained machine learning models that have had a certain amount of training performed.

[0035] In an example, the machine learning model (206) can utilize various machine learning algorithms, such as linear, non-linear, tree-based and boosting algorithms, median, and blind test results that have been evaluated with available data sets to optimize the overall error. For example, in some embodiments, the machine learning model (206) can include a random forest classifier, a convolutional neural network, and / or a light gradient boosting machine classifier.

[0036] In an example, the reservoir quality indicator (RQI) (216) is a parameter or set of parameters generated by the machine learning model (206) from wellsite data (212). The reservoir quality indicator (RQI) (216) can be derived from well logs and includes, for example, average porosity, average permeability, net pay thickness, flow capacity, water saturation, hydrocarbon storage capacity, and permeability / porosity, as well as other user-defined RQIs. The indicator is used to evaluate the quality of subsurface reservoir rock, particularly in terms of its suitability for hydrocarbon (oil and gas) accumulation and production.

[0037] In an example, the missing reserves (218) are predictions generated by one of the machine learning models (206) based on the reserve quality indicator (216) from wellsite data (212). More generally, the term "missing reserves" refers to hydrocarbon reserves that are expected to be present in a given reservoir but have not been discovered or quantified through exploration and drilling activities.

[0038] Figure 2 The overall process flow is shown. For machine learning modeling purposes, the missing reserves can be divided into 2 categories. In the first category, the missing reserves can be identified as behind-casing opportunities (BCOs), where the pay zone is not registered as hydrocarbon-bearing when the wellsite (210) is drilled and is registered as hydrocarbon-bearing when the wellsite (210) is drilled but not produced due to the completion strategy. In the second category, the missing reserves can be identified between wells using seismic, 3D data sets, and core measurements with various reservoir properties such as lithology, porosity, and water saturation.

[0039] The machine learning model consumes the provided inputs, such as physical parameters from well logs (214), well information and well characteristic data of the well site (210) of the candidate well (220). The machine learning quickly performs quality control and conditioning of the well log data, predicts new well log data by filling in missing information or missing regions, and provides newly constructed well logs. These generated well logs are used to find the BCOs of the missing intervals that potentially can be produced within the well site (210). Additionally, the generated well logs can be combined with well characteristic data (such as core measurement data, special core analysis) to produce the RQI (216) for each well and well interval. The (one or more) machine learning models (206) use the RQI information and analyze the average production of the well and well intervals to produce a second set of missing reserves. The machine learning model (206) can map the potential regions between well sites (210) to provide potential opportunities for the missing reserves (218).

[0040] In some embodiments, the candidate well (220) can be one of the well sites (210) identified by the machine learning model (206) based on the missing reserves (218). For example, the candidate well (220) can be targeted by one of the machine learning levels (206) for intervention based on the comparison of RQI with production performance to identify work on candidate wells with good reservoir quality and poor production performance. The analysis can be performed at both the well level and the completion level. The production performance metrics can include data such as cumulative oil / gas production and peak productivity rate.

[0041] In some embodiments, the candidate well (220) can be an undeveloped and / or unproduced region between well sites (210). For example, for regions between well sites (210) with missing well logs or missing physical measurements, using machine learning-based reservoir property generation, the machine learning model (206) can help identify missing reserve opportunities between well sites (210). The economic outcome (222) is the prediction by the machine learning model of multiple intervention options for the candidate well. For example, using the candidate well (220) and data on interventions performed at other well sites (210), the machine learning model (206) can perform probabilistic data-driven reservoir simulations to estimate the amount of oil or gas produced by each well, i.e., the value of each well after the intervention. The costs of creating and operating new and existing wells are also considered. These costs and values can then be combined to calculate the net present value (NPV) of each well and thus the discounted profitability index (DPI) of the entire oil field based on the determined workover options.

[0042] Turning to Figure 3 shows a machine learning framework according to an illustrative embodiment. Figure 3 The framework shown can be implemented using Figure 2 the machine learning model (206).

[0043] As shown, Figure 3 the framework provides a decision support system and method for using enhanced intelligence (AI)-driven technology to identify missing reserves. Enhanced intelligence is an alternative conceptualization of artificial intelligence that uses machine learning and data analysis to process large amounts of data and improve human intelligence. Illustrative embodiments use AI to accelerate the automatic identification of missing reserves.

[0044] The decision support system is divided into two parts - a planning (310) space and an operation (320) space. The planning (310) integrates multiple decision-making stages for automatically and quickly identifying opportunities in missing reserves. The operation (320) integrates multiple decision-making stages for identifying target locations, ranking, and evaluating options through an AI platform to support expert decision-making.

[0045] The planning (310) includes opportunity assessment (312), data discovery and validation (314), machine learning-assisted opportunity identification (316), and machine learning-based RQI (reservoir quality indexing) and well contribution space (318).

[0046] As shown, the opportunity assessment (312) receives available wellsite data from a data repository (such as Figure 2 the data repository (202)). The data includes wellsite data collected from and / or generated by multiple data sources for E&P operations. For example, the data can include production data (monthly), pressure / volume / temperature (PVT) data, top structure maps, volume maps (including hydrocarbon pore volume (HCPV), pore volume distribution (PVD), net volume (NV), gross rock volume (GRV)), fault and aquifer locations, permeability trends, reservoir pressure, well trajectories (X, Y, Z), well tops, well events (including zone open / close), log data, conventional core analysis (RCA) data, and special core analysis data (SCAL), such as relative permeability, capillary pressure, and wettability.

[0047] Focusing on the available data collected from various data sources, the data is input into multiple discovery and validation models (314), which employ one or more machine learning algorithms focused on data collection and curation from various data sources, including operations such as data preparation, outlier detection, data transformation, and reconstruction of missing data.

[0048] According to an illustrative embodiment, the discovery and validation model (314) can view and process available data using one or more systems of data analysis, AI / machine learning (ML) techniques, and domain expert guidance. The discovery and validation model (314) can utilize various reservoir simulation and software frameworks, such as Petrel, Techlog, and various machine learning frameworks, such as Dataiku, Python & Jupiter notebook.

[0049] According to an illustrative embodiment, the discovery and validation model (314) can predict reservoir properties directly from measurements such as effective porosity (PHIE), total porosity (PHIT), permeability (PERM), volume of shale in the rock volume (VSHALE), and / or surface well test (SWT).

[0050] For example, the discovery and validation model (314) can derive missing feature values from known information about the data distribution of the features. For categorical features, the missing values can be replaced by the most common value of the feature (e.g., the mode value of the feature). Additionally, one-hot encoding can be applied to categorical features to facilitate consumption by machine learning models. For numerical features, the missing values can be replaced with the value corresponding to the 50th percentile of the feature (e.g., the median of the feature). To address possible typographical and / or input errors, outliers can be removed from numerical features. For example, removing outliers removes less than 1% of the training data in the sample training dataset.

[0051] The discovery and validation model (314) can perform cross-validation and oversampling using, for example, synthetic minority over-sampling technique (SMOTE). A cross-validation procedure can be applied to calculate the accuracy of the predictions using the training and validation datasets. For example, the discovery and validation model (108) can apply a modified SMOTE oversampling technique to the training dataset to help reduce data imbalance for one or more tasks.

[0052] The output from the discovery and validation model (314) is sent to one or more opportunity identification models (316) to perform various machine learning-assisted workflows, such as machine learning-assisted log quality control (QC), and machine learning-driven behind-casing opportunities (BCO), machine learning-based reservoir quality index (RQI) and well contribution, and data-driven target point analysis.

[0053] As used herein, a workflow can be a process that includes multiple work steps. The work steps can operate on data, e.g., to create new data, update existing data, etc. As an example, a system can operate on one or more inputs and, for example, create one or more results based on one or more algorithms. A workflow can include one or more work steps that access modules such as plugins (e.g., external executable code, etc.). A workflow can be implemented in various reservoir simulation and machine learning software frameworks to operate on data (102), including frameworks such as Petrel, Techlog, Dataiku, Python & Jupiter notebook.

[0054] The opportunity identification model (316) uses a machine learning-assisted log QC model, which is applied to automatically assist in adjusting the workflow for data curation, making the data available across legacy systems and reducing uncertainty. The machine learning model consumes the provided inputs, such as physical parameters from log measurements, interpreted values, oil saturation of well intervals. Machine learning quickly performs quality control and adjustment of log data to evaluate the quality of physical properties, such as porosity, permeability, VSHALE, and saturation. The machine learning algorithm predicts new log data by filling in missing information or missing zones and provides newly constructed logs. The machine learning model uses several such logs and utilizes the possible connectivity of zones between wells to perform the interpretation of zones between wells.

[0055] The opportunity identification model (316) can include a machine learning-driven BCO, which uses the output from a previous machine learning log model. For an interval of interest, if the quality trust factor is high, the model uses this information to use machine learning to label or identify hydrocarbon intervals behind the casing. In the case of a low quality trust factor, the model uses its own machine learning prediction module to predict the characteristics of the interval. In both cases, the machine learning model will identify the missing of the behind-casing interval opportunity (BCO), which was not previously considered or was identified as a null value but has potential.

[0056] The machine learning-based RQI and well contribution model (318) evaluates the reservoir quality index (RQI) of a well. The index can be derived from logs and includes, for example, average porosity, average permeability, net pay thickness, flow capacity, water saturation, hydrocarbon storage capacity, and permeability / porosity, as well as a user-defined RQI.

[0057] The above - trained machine - learning model uses BCO, RQI, and well deliverability to highlight specific zone - level opportunities for high - quality remaining reserves. However, instead of a single deterministic case used traditionally, the machine - learning model creates hundreds of realizations by parameterizing characteristic values above and below actual or predicted values, thus providing a probability map of the scenario distribution. This is further used to create a stochastic map of remaining oil in - place reserves and can be visualized in a 3D geological model. It replaces the effort of running a dynamic simulation model over a small fraction of the time through a machine - learning agent.

[0058] The operation (320) includes data - driven best - point analysis (322), production evaluation and prediction analysis (324), risk / cost and AI for missing reserve opportunities (326). The solution also proposes interventions to identify missing reserve opportunities (328), which can help the "operation" space while managing uncertainty and risk.

[0059] The data - driven target - point analysis (322) is one or more machine - learning algorithms for performing candidate screening and candidate ranking. For existing wells, it uses a machine - learning - driven BCO model to highlight the potential production that can be achieved from a specific well and also specifies which intervals or pay zones have potential. In this case, missing reserves can be produced from existing perforations or new pay zones can be perforated to increase production.

[0060] The data - driven target - point analysis (322) also utilizes machine - learning - based RQI to identify missing reserves between wells using RQI, well - to - well pay - zone correlations, and available reservoir information. It identifies specific pay zones between wells and generates the locations of one or more drilling sites or "target points" within the oilfield. Machine learning generates various integrations, meaning realizations that utilize all permutations and combinations of reservoir characteristics, and provides such maps of hundreds of possible missing locations with success probabilities. By generating machine - learning simulations on a cloud with flexible computing power, its performance is significantly improved, leading to a very fast analysis of such scenarios.

[0061] As used herein, a "target point" or "best point" is a possible drilling site with a favorable combination of various geological, geochemical, and petrophysical characteristics that can indicate an expected hydrocarbon zone over the subsurface domain of an oilfield.

[0062] The production evaluation and prediction analysis (324) is one or more machine - learning algorithms for comparing RQI with production performance metrics and identifying work on candidate wells with good reservoir quality but poor production performance. Production performance metrics can include data such as cumulative oil / gas production and peak productivity. The analysis can be performed at both the well level and the completion level.

[0063] For example, poor performance identification can be based on production heterogeneity analysis, which uses the average behavior of wells as a reference to compare the behavior of individual wells in a group. This comparison can be based on a modified heterogeneity index - a dimensionless calculated variable, which is defined as:

[0064] Where:

[0065] MHI = Modified Heterogeneity Index;

[0066] VALUE Group.Avg = The arithmetic mean of all selected wells at the current time step;

[0067] VALUE MaxWell = The maximum value at the current time step among all selected wells; and

[0068] VALUE MinWell = The minimum value at the current time step among all selected wells.

[0069] The risk / cost of missing reserve opportunities and AI(326) are one or more machine learning algorithms for performing probabilistic reservoir simulation to estimate the amount of oil or gas produced by each well, i.e., the value of each well. The costs of creating and operating new and existing wells are also considered. These costs and values can then be combined to calculate the net present value (NPV) of each well and, thus, the discounted profitability index (DPI) for the entire oil field based on the determined workover options.

[0070] One or more embodiments also propose intervention opportunities (328), which can assist the "operation" space while managing uncertainty and risk. For example, a platform can be connected to an interface (e.g., user interface, application programming interface) that enables access to one or more predicted missing reserve opportunities (328) for a selected area. These missing reserve opportunities (328) are then used to see the impact on production enhancement and predictive analysis, where a risk / cost analysis is performed to determine the return on investment. A domain expert guidance system can be employed to identify and finalize missing reserve opportunities based on the processed data.

[0071] Figures 4.1 - 4.3 A converter architecture is shown. The converter architecture (400) can be used to implement Figure 2 the machine learning model (206). Compared with a recurrent neural network (RNN), a converter is less susceptible to the vanishing gradient problem, which is a characteristic of networks using gradient-based optimization techniques (i.e., reduced efficacy due to the decay of temporal information, where earlier layers learn more slowly than later layers).

[0072] The Transformer architecture (400) relies on a self-attention (intra-attention) mechanism, thus eliminating the recurrent operations computed in recurrent neural networks, which can be used to compute the latent space representations on both the encoder (410) and decoder (412) sides. Positional encoding (414) is added to the input and output embeddings (416, 418) without replication. Positional information similar to the time steps in a recurrent network provides the order of the input and output sequences to the Transformer network. A combination of absolute positional encoding and relative positional information can be used. The input from previously generated symbols is used autoregressively by the model for the next prediction, which is organized as a stack of encoder-decoder networks. Additionally, uniform layers constitute both the encoder (410) and decoder (412), and each layer is built from two sub-layers: a multi-head self-attention layer (420) and a per-position feed-forward network (FFN) layer (422). The multi-head sub-layer (420) enables the use of multiple attention functions at an equivalent cost of leveraging attention, while the FFN sub-layer (422) uses a fully connected network to process the attention sub-layer. The FFN applies multiple linear transformations at each position and the rectified linear unit (ReLU), which extends the self-attention mechanism to effectively consider the representation of relative positioning (i.e., the distance between sequence elements).

[0073] Figure 4.2 and Figure 4.3 illustrates a multi-head attention head architecture according to one or more embodiments. Multiple attention heads ( Figure 4.2 ) can be combined to form a multi-head attention ( Figure 4.3 ) that can be implemented in the Transformer architecture (400) of FIG. 4.

[0074] As Figure 4.2 shown, the attention function can be described as mapping a query and a set of key-value pairs to an output, where the query, key, value, and output are all vectors. The attention head is initialized with a set of weights that will be learned during training to assign different weights to various parts of the input sequence.

[0075] The attention head computes three vectors for each position in the input sequence: a query (Q), a key (K), and a value (V). In the "encoder-decoder attention" layer, the query comes from the previous decoder layer, and the memory keys and values come from the output of the encoder. The query (Q) vector represents the position under consideration, while the key (K) and value (V) vectors represent all other positions in the sequence. These vectors are obtained by multiplying the input sequence by learnable weight matrices, which are unique for each of the query, keyword, and value vectors. The query, keyword, and value are packed together into matrices that provide input to various linear functions. This allows each position in the decoder to attend to all positions in the input sequence.

[0076] Calculate the similarity scores between the query vector and all the key vectors in the sequence. By calculating the dot product between the query vector and each key vector, the attention function can be multiplicative. Alternatively, an additive function can be used, where the compatibility function can utilize a feed-forward network with a single hidden layer. The resulting scores can be normalized using a scaling factor.

[0077] For each position in the sequence, calculate the attention scores by applying the SoftMax function to the obtained similarity scores. The attention scores represent the weights assigned to each position in the sequence, indicating how much attention should be given to that position.

[0078] Then use the attention scores to calculate the weighted sum of the value vectors. This summation produces a single output vector, which represents the weighted combination of all the value vectors in the sequence. In other words, the output is calculated as the weighted sum of the values, where the weights assigned to each value are calculated through the compatibility function of the query with the corresponding key.

[0079] Multi-head attention( Figure 4.3 ) allows the model to jointly address information from different representation subspaces at various positions. Depending on the matrix dimensions, the query, key, and value are projected h times using different learned linear projections. The attention function is performed in parallel on each of these projected versions of the query, key, and value, resulting in dv-dimensional output values. The dimensional outputs are concatenated and projected again to obtain the final value. Since the size of each head is reduced, the total computational cost of multi-head attention is similar to the total computational cost of single-head attention with full dimensions.

[0080] Although Figure 1 -4 shows the configuration of the components, other configurations can be used without departing from the scope of the present invention. For example, various components can be combined to create a single component. As another example, functions performed by a single component can be performed by two or more components.

[0081] Figure 5 A flowchart of a method for identifying missing reserves in a reservoir according to some embodiments is shown. Figure 5 The method of Figure 2 can be implemented in one or more components of the computer system shown in

[0082] At block 510, ingest multiple well logs of wellsite data in the reservoir. It also uses other data, such as well information, RCA, SCAL, PVT, and historical production data. In some embodiments, the wellsite data includes a set of production curves for each wellsite.

[0083] At block 520, a first machine learning model consumes the provided quality control inputs, such as log measurements, reservoir information, well information, and other physical measurements, such as core and special core analysis data, to generate multiple quality control or reconstructed logs from the machine learning model. The machine learning model then also generates reservoir quality metrics from the multiple inputs.

[0084] In some embodiments, generating multiple reservoir quality metrics includes generating predicted production curves for multiple possible interventions to be applied to the well site. For example, identifying an intervention applied at the well site within the set of production curves; generating a prediction curve for the well site based on the set of production curves and the identified intervention.

[0085] At block 530, a second machine learning model determines missing reserves based on the reservoir quality metrics of multiple well sites. For an interval of interest, if the quality confidence factor is high, the model uses this information to mark or identify hydrocarbon intervals behind the casing using machine learning. In the case of a low quality confidence factor, the model uses its own machine learning prediction module to predict the characteristics of the interval. In both cases, the machine learning model will identify the missing behind-casing interval opportunities (BCOs) that were previously not considered or were identified as null but have potential.

[0086] At block 540, a third machine learning model determines candidate wells based on the missing reserves. Candidate wells can be selected from existing well sites in the reservoir (i.e., wells that would benefit from an intervention) or new wellbores and / or drillholes to access and identify the optimal points.

[0087] Candidate wells can be identified based on a comparison with existing wells. For example, a vector representation of the well site can be generated from well site data. Then the vector similarity between well sites is determined using the well site data collected prior to the intervention.

[0088] The above trained machine learning models use BCOs, RQIs, and well deliverability to highlight specific zone-level opportunities for high-quality remaining reserves. However, instead of a single deterministic case used traditionally, the machine learning models create hundreds of realizations by parameterizing the characteristic values above and below the actual or predicted values, thus providing a probability map of the scenario distribution. This is further used to create a stochastic remaining oil in place map and can be visualized in a 3D geological model. It replaces the effort of running a dynamic simulation model over a fraction of the time with machine learning agents.

[0089] At block 550, a fourth machine learning model uses the economics provided for the oilfield and predicts the economic outcomes of multiple intervention options for the candidate wells, such as NPV, rate of return.

[0090] For example, the production curve of a candidate well can be extrapolated in time using the production curves of similar well sites.

[0091] At block 560, oilfield operations can be controlled based on predicted economic outcomes.

[0092] Although the various blocks in the flowchart are presented and described in sequence, at least some of the blocks can be executed in a different order, can be combined or omitted, and at least some of the blocks can be executed in parallel. Additionally, the blocks can be executed either actively or passively.

[0093] Figures 6 - 12 is an example. The Figures 6 - 12 example provided is for illustrative purposes only and is not intended to limit the scope of the present disclosure.

[0094] Figure 6 is an example of machine learning-assisted log quality control and interpretation of the characteristics of various zones connecting wells. It also shows how machine learning models can be used for machine learning-based reservoir characteristics in areas with missing logs or missing physical measurements. This can help identify missing reserve opportunities between wells.

[0095] Figure 7 and Figure 8 shows an example of the evaluation of the reservoir quality index (RQI) of a well and the distribution of the overall quality of measurement-based characteristic metrics.

[0096] Figure 9 is an example of a graph of the reservoir quality index versus the average actual production rate over 3 months. The production rate variable can be changed according to the research requirements for an average of 6 months or one year or longer. The machine learning model can create a collection of various such implementations to map for analysis of high-potential wells that are not fully produced. This could be an opportunity to drill deeper and identify candidates for missing reserves.

[0097] Figure 10 is an example of a conventional non-uniformity analysis graph for validating the oil production rate versus the water production rate of potential candidate wells. The aim is to select wells with high oil production and a trend of lower water production.

[0098] Figure 11 is an example showing potential additional thicker layers identified by a machine learning algorithm supported by an SWT curve. The predicted BCO can have low, medium, and high confidence levels that can be verified and approved by domain experts.

[0099] Figure 12 is an example of a heatmap of missing reserves identified within a reservoir. The map compares the results generated by a traditional Bayesian model with the results generated by one or more machine learning models (such as Figure 2 the machine learning model (206)). As shown, compared to traditional statistical methods, the estimates of the AI system can identify opportunity zones within the reservoir faster and more accurately.

[0100] Embodiments can be implemented on a computing system specifically designed to achieve improved technical results. When implemented in a computing system, the features and elements of the present disclosure provide significant technical advancements over computing systems that do not implement the features and elements of the present disclosure. Any combination of mobile devices, desktops, servers, routers, switches, embedded devices, or other types of hardware can be improved by including the features and elements described in the present disclosure. For example, as Figure 13.1 shown, the computing system (1300) can include one or more computer processors (1302), non-persistent storage devices (1304), persistent storage devices (1306), a communication interface (1312) (e.g., a Bluetooth interface, an infrared interface, a network interface, an optical interface, etc.), and many other elements and functions that implement the features and elements of the present disclosure. The (one or more) computer processors (1302) can be integrated circuits for processing instructions. The (one or more) computer processors can be one or more cores or microcores of a processor. The (one or more) computer processors (1302) include one or more processors. The one or more processors can include a central processing unit (CPU), a graphics processing unit (GPU), a tensor processing unit (TPU), combinations thereof, etc.

[0101] The input device (1310) can include a touch screen, a keyboard, a mouse, a microphone, a touchpad, an electronic pen, or any other type of input device. The input device (1310) can receive input from a user in response to data and messages presented by the output device (1308). The input can include text input, audio input, video input, etc., which can be processed and transmitted by the computing system (1300) according to the present disclosure. The communication interface (1312) can include an integrated circuit for connecting the computing system (1300) to a network (not shown) (e.g., a local area network (LAN), a wide area network (WAN) (such as the Internet), a mobile network, or any other type of network) and / or connecting to another device (such as another computing device).

[0102] In addition, the output device (1308) can include a display device, a printer, an external storage device, or any other output device. One or more of the output devices can be the same as or different from the input devices. The input and output devices can be locally or remotely connected to the computer processor (1302). There are many different types of computing systems, and the foregoing input and output devices can take other forms. The output device (1308) can display the data and messages sent and received by the computing system (1300). The data and messages can include text, audio, video, etc., and include the data and messages described above in other figures of the present disclosure.

[0103] Software instructions in the form of computer-readable program code for executing the embodiments may be stored, in whole or in part, temporarily or permanently, on a non-transitory computer-readable medium, such as a CD, DVD, storage device, floppy disk, magnetic tape, flash memory, physical memory, or any other computer-readable storage medium. Specifically, the software instructions may correspond to computer-readable program code that, when executed by one or more processors, is configured to execute one or more embodiments of the present invention, which may include sending, receiving, presenting, and displaying the data and messages described in other figures of the present disclosure.

[0104] Figure 13.1 The computing system 1300 in [reference] can be connected to a network or be part of a network. For example, as Figure 13.2 shown, the network (1320) can include multiple nodes (e.g., node X (1322), node Y (1324)). Each node can correspond to a computing system, such as Figure 13.1 the computing system shown, or a combined set of nodes can correspond to Figure 13.1 the computing system shown. By way of example, the embodiments can be implemented on nodes of a distributed system connected to other nodes. By another example, the embodiments can be implemented on a distributed computing system having multiple nodes, where each part can be located on a different node within the distributed computing system. Additionally, one or more elements of the aforementioned computing system 1300 can be located at a remote location and connected to other elements via a network.

[0105] Nodes in the network (1320) (e.g., node X (1322), node Y (1324)) can be configured to provide services to a client device (1326), including receiving requests and sending responses to the client device (1326). For example, the node can be part of a cloud computing system. The client device (1326) can be a computing system, such as Figure 13.1 the computing system shown in [reference]. Additionally, the client device (1326) can include and / or execute all or part of one or more implementations of the present invention.

[0106] Figure 13.1A computing system can include functionality to present raw and / or processed data, such as results of comparisons and other processing. For example, presenting data can be accomplished through various presentation methods. Specifically, data can be presented by being displayed in a user interface, sent to different computing systems, and stored. The user interface can include a GUI that displays information on a display device. The GUI can include various GUI widgets that organize what data is shown and how the data is presented to the user. Additionally, the GUI can present data directly to the user, e.g., data presented as actual data values by text, or visual representations of data rendered by a computing device, such as through a visualization data model.

[0107] As used herein, the term "connected to" contemplates a variety of meanings. A connection can be direct or indirect (e.g., through another component or network). A connection can be wired or wireless. A connection can be a temporary, permanent, or semi-permanent communication channel between two entities.

[0108] The various descriptions of the figures can be combined and can include features described in other figures of the present application or be included within features described in other figures of the present application. The various elements, systems, components, and boxes shown in the figures can be omitted, repeated, combined, and / or altered as shown. Accordingly, the scope of the present disclosure should not be regarded as limited to the specific arrangements shown in the figures.

[0109] In the present application, ordinal numbers (e.g., first, second, third, etc.) can be used as adjectives for elements (i.e., any noun in the present application). The use of ordinal numbers is not to imply or create any particular ordering of the elements, nor to limit any element to only a single element, unless expressly disclosed, such as by using terms like "before," "after," "single," and other such terms. Instead, the use of ordinal numbers is to distinguish elements. By way of example, a first element is different from a second element, and the first element can include more than one element and be after (or before) the second element in the ordering of the elements.

[0110] Additionally, unless expressly stated otherwise, the term "or" is an "inclusive or" and thus includes the term "and." Further, unless expressly stated otherwise, items joined by the term "or" can include any combination of the item and any number of each item.

[0111] In the foregoing description, numerous specific details are set forth to provide a more thorough understanding of the present disclosure. However, it will be apparent to one of ordinary skill in the art that the techniques may be practiced without these specific details. In other instances, well-known features have not been described in detail to avoid unnecessarily complicating the description. Additionally, other embodiments may be devised that are not explicitly described above and that do not depart from the scope of the claims disclosed herein. Accordingly, the scope should be limited only by the appended claims.

Claims

1. A method for identifying missing reserves in a reservoir, the method comprising: Ingest multiple well logs of wellsite data in a reservoir; Generate multiple post-casing opportunities from the multiple well logs by a first machine learning model; Determine missing reserves through a second machine learning model based on the reservoir quality metrics of the multiple well sites; Determine candidate wells through a third machine learning model based on the missing reserves; Predict economic outcomes and rank multiple intervention options for the candidate wells through a fourth machine learning model; and Control oilfield decisions based on the predicted economic outcomes.

2. The method according to claim 1, wherein the wellsite data includes a set of log curves for each wellsite.

3. The method according to claim 2, comprising: Identify interventions applied at the wellsite in the set of well log curves, production curves, core data, and special core analysis data (SCAL data); Generate a predicted reservoir quality index for each well and producing formation within the wellsite based on the set of reservoir and production data and the identified interventions.

4. The method according to claim 3, comprising extrapolating each of the reservoir quality indices to identify missing reserves between the well and a specific oil-producing formation in a timely manner before the intervention using production curves and reservoir quality indices of similar wellsites.

5. The method according to claim 2, comprising: Generate a vector representation of the wellsite; and Determine the similarity between the wellsites before the intervention.

6. The method according to claim 1, wherein the candidate well is selected from existing wellsites in the reservoir.

7. The method according to claim 1, wherein the candidate well is a new wellsite in the reservoir.

8. A computer program product, comprising: A non-transitory computer-readable storage medium storing program code that, when executed by a computer processor of a computing system, causes the computing system to perform the following method: Ingest multiple well logs of wellsite data in a reservoir; Generate multiple post-casing opportunities from the multiple well logs by a first machine learning model; Determine missing reserves through a second machine learning model based on the reservoir quality metrics of the multiple well sites; Determine candidate wells through a third machine learning model based on the missing reserves; Predict economic outcomes and rank multiple intervention options for the candidate wells through a fourth machine learning model; and Control oilfield decisions based on the predicted economic outcomes.

9. The computer program product according to claim 8, wherein the wellsite data includes a set of production curves for each of the wellsites.

10. The computer program product according to claim 9, further comprising: Identify interventions applied at the wellsite in the set of production curves; and Generate a predicted curve for the wellsite based on the set of production curves and the identified interventions.

11. The computer program product according to claim 10, further comprising: Use the production curves of similar wellsites to extrapolate each production curve in a timely manner before the intervention.

12. The computer program product according to claim 9, further comprising: Generate a vector representation of the wellsite; and Determine the similarity between the wellsites before the intervention.

13. The computer program product according to claim 8, wherein the candidate well is selected from existing well sites in the reservoir.

14. The computer program product according to claim 8, wherein the candidate well is a new well site in the reservoir.

15. A system, comprising: Computer processor; Memory; and Instructions stored in the memory and executable by the computer processor to cause the computer processor to perform operations including: Ingest multiple well logs of wellsite data in a reservoir; Generate multiple post-casing opportunities from the multiple well logs by a first machine learning model; Determine missing reserves through a second machine learning model based on the reservoir quality metrics of the multiple well sites; Determine candidate wells through a third machine learning model based on the missing reserves; Predict economic outcomes and rank multiple intervention options for the candidate wells; and Control oilfield decisions based on the predicted economic outcomes.

16. The system according to claim 15, wherein the well site data includes a set of production curves for each well site.

17. The system according to claim 16, further comprising: Identify interventions applied at the wellsite in the set of production curves; and Generate a predicted curve for the wellsite based on the set of production curves and the identified interventions.

18. The system according to claim 17, further comprising extrapolating each of the production curves in a timely manner before the intervention using production curves of similar well sites.

19. The system according to claim 16, further comprising: Generate a vector representation of the wellsite; and Determine the similarity between the wellsites before the intervention.

20. The system according to claim 16, wherein the candidate well is selected from existing well sites in the reservoir.

21. The system according to claim 16, wherein the candidate well is a new well site in the reservoir.

Images