WO2019023255A1

WO2019023255A1 - Developing oilfield models using cognitive computing

Info

Publication number: WO2019023255A1
Application number: PCT/US2018/043524
Authority: WO
Inventors: Hubertus Victor Thomeer; Diankui Fu
Original assignee: Bj Services, Llc
Priority date: 2017-07-24
Filing date: 2018-07-24
Publication date: 2019-01-31

Abstract

In examples, a system comprises a processor and storage coupled to the processor. The storage stores executable code which, when executed by the processor, causes the processor to obtain historical data pertaining to wells within a predetermined area; process the historical data to identify a relationship; and generate a model based on the relationship. The system comprises a cognitive computing system.

Description

Developing Oilfield Models Using Cognitive Computing

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] The present application claims priority to U.S. provisional application serial number 62/536,081 , filed July 24, 2017, which is herein incorporated in its entirety by reference.

BACKGROUND

[0002] Wells are drilled through various portions of the Earth, called "formations," to appropriate depths at which hydrocarbons may be extracted. The process of creating a new well or group of wells involves numerous steps, including well planning, well design, drilling operations, formation evaluation and testing, and well completion. During the completion cycle of the well, hydraulic fracturing is used to stimulate hydrocarbon flow by producing fractures deep within the reservoir. These fractures are created by high- pressure injection of fracturing fluids including water, proppants, and additives into the formation. This process, commonly known as 'Tracking", has evolved over the years from being done in vertical wells to now being done horizontal wells in precise locations along the horizontal section to ensure maximum effect. This has resulted in the need to collect an extensive amount of data when performing a fracking job.

BRIEF DESCRIPTION OF THE DRAWINGS

[0003] Various examples are described below referring to the following figures:

[0004] Figure 1 is a block diagram of a cognitive computer system, in accordance with various embodiments.

[0005] Figure 2 is a flow diagram of a method for developing and refining an oilfield operations model using the cognitive computer system of Figure 1 , in accordance with various embodiments.

DETAILED DESCRIPTION

[0006] Disclosed herein are various embodiments of systems and techniques for using cognitive computing (also known as machine learning and artificial intelligence) to plan an oilfield operation, such as a tracking job, for a target well based on a model produced using data gathered from one or more neighboring wells. In the illustrative context of hydraulic fracturing, a cognitive computer uses such data gathered from one or more neighboring wells to generate and refine a model that is then used to plan the fracking job of the target well, assuming that the target well has characteristics (e.g., mineralogy) similar to those of the neighboring wells. The scope of this disclosure, however, is not limited to the application of cognitive computing to only fracking jobs. Cognitive computers may be used to create, train, apply, and refine models for any of a variety of tasks in the oil and gas context, and this disclosure is intended to encompass any and all such embodiments. Thus, although the description below is provided in the illustrative context of fracking jobs, the same description may be adapted to virtually any task in the oil and gas context. Models generated using cognitive computers as described herein may present numerous technical advantages, such as a reduced requirement for target well data when planning the target well (e.g., when developing a fracking job for the target well). Additional materials supporting the various features and embodiments disclosed herein are provided below in Appendix A. Further, although this disclosure describes techniques in the context of target wells and wells neighboring the target wells, the techniques also may apply to target wells and wells that do not neighbor the target wells but that have similar formation profiles as the target wells.

[0007] Figure 1 depicts an illustrative cognitive computing system 100 in accordance with various embodiments. The illustrative cognitive computing system 100 comprises a cognitive computer 102, which may be any suitable computing system employing cognitive computing hardware such as neurosynaptic processor cores and capable of executing probabilistic, machine-learning algorithms encoded in executable code. In some embodiments, the cognitive computer 102 may not contain neurosynaptic hardware but may nevertheless be configured to execute probabilistic, machine-learning algorithms encoded in executable code. The cognitive computer 102 couples to an input interface 104, which may include, for instance, a keyboard, a mouse, a touchscreen, a microphone, or a network connection to another electronic device. The cognitive computer 102 also couples to an output interface 106, such as a display, audio speaker, or network connection to another electronic device. [0008] The cognitive computer 102 may further couple to a network interface 108 via which the cognitive computer 102 communicates with one or more other electronic devices, such as other cognitive or von Neumann computing devices. The system 100 also includes a storage 110, which may comprise any suitable type of storage, including random access memory, read-only memory, or another type of memory, and it may be co-located with the cognitive computer 102, located remotely from the cognitive computer 102, and/or be a distributed storage that is located in multiple locations at once. The cognitive computer 102 may additionally couple to and communicate with storage 112, which is a local storage that is co-located with the cognitive computer 102 and, in some embodiments, is co-housed within the same physical device as the cognitive computer 102. The storage 112 may store cognitive algorithm executable code 114 that is designed based on probabilistic principles, not on deterministic algorithms, and that is able to engage in machine learning using one or more training sessions. The cognitive computer 102, upon executing the cognitive algorithm executable code 114, may perform some or all of the steps of the method depicted in Figure 2. In some embodiments, the cognitive algorithm executable code 114 is suitably programmed to facilitate performance of the steps in Figure 2. In some embodiments, the executable code 114 is programmed to learn to perform some or all of the steps described with respect to Figure 2.

[0009] Figure 2 depicts a flow Table of an illustrative method 200 for a cognitive computer to plan a fracking job in a target well based on data pertaining to wells near (e.g., within 1 mile, within 5 miles, or within another predetermined distance from) the target well. Although the method 200 is illustratively described in the context of fracking jobs, the principles of the method may be extended to apply to any suitable type of oilfield operation.

[0010] The method 200 includes a step 202 in which data pertaining to one or more nearby wells is collected. Such data may comprise, for example, historical information pertaining to any and all phases of the well development cycle, including well planning (e.g., seismic data), well design, drilling operations, formation evaluation and testing, and well completion. Production data also may be included. For instance, such data may include seismic information; drilling information; completion information; production information; rock properties; total organic content; natural fractures; horizontal stress; density; porosity; well logs (e.g., gamma, neutron, acoustic, etc.); tracking job designs; production history; well trajectory; reservoir measurements; petrophysical data; surface equipment measurements; and field wellhead pressures over time. Such data may also include information pertaining to the target well, such as well design information (e.g., optimal well trajectory); optimized cluster spacing; total fracture surface area; and optimal fluid and proppant design. Additional types of information may include synthetic data logs; fracturing designs (e.g., perforation spacing, perforation depth, proppant used, etc.) of nearby wells; data pertaining to chemicals provided downhole; fracking gradients; illustrative sources of data may include solids measurements; fluids measurements; equipment measurements; surface measurements; wellhead measurements; reservoir measurements (e.g., clay content and calculated information, such as data based on Young's modulus); and nearby reservoir measurements. The scope of this disclosure is not limited to the foregoing illustrative data examples; it encompasses any other types of data that may relate in any way to the intended oilfield operation (e.g., planning frac jobs).

[0011] The cognitive computer 102 may collect such data from any suitable source. For example, the cognitive computer 102 may receive such data from oilfield personnel, from another electronic device via the network interface 108, from the storage 110, or from the storage 112. Other sources of data are contemplated. Such data may, for instance, be collected by the cognitive computer 102 and stored in either or both of the storages 110, 112.

[0012] In step 204, the cognitive computer 102 categorizes the data collected in step 202 into groups. For example, in some embodiments, the cognitive computer 102 categorizes the data into either a "known" group or an "unknown" group. Data in the "known" group are data that are known with either complete certainty or at least certainty beyond a particular threshold, where the threshold is determined by oilfield personnel on an application-specific basis. Data in the "unknown" group are data that are not known with either complete certainty or at least certainty beyond the threshold. For instance, data in the "known" group may include solids measurements (e.g., in and out of the well(s)); fluids measurements (e.g., in and out of the well(s)); well depth and position; fluid properties pumped in wells; formation measurements; and production history. Additionally, for instance, data in the "unknown" group may include flow regime; rock properties; natural fracture geometry; formation permeability; formation fluid properties; matrix permeability; fracture propagation mechanisms; nano darcy rock fluid flow mechanisms; high and low permeability zones; unconfined compressive strength; and mineralogy; anisotropy; data determined using Poisson's ratio; physics-based equations; water saturation; fracture toughness; and closure stress. The scope of disclosure is not limited to the foregoing illustrative examples. The cognitive computer 102 may categorize such data on any suitable basis; for instance, the cognitive computer 102 may be trained to identify "known" and "unknown" data and to categorize the data accordingly.

[0013] The data categorization of step 204 may be performed on data that is structured and contextual. Structured data may include, for instance, data that is stored or presented in a database or data structure with fields, such as logging data. Contextual data is non- structured information, such as textual descriptions and drawings. Due to the size of data sets, particularly structured data sets, the performance of step 204 may require sequence tagging, structured hierarchies, sorting similar terms and measurements, and reduction of randomness in the data and variables. The cognitive computer 102 may achieve this with principle component analysis, multilinear subspace learning, and generalized discriminant analysis. The cognitive computer 102 may also detect and account for data anomalies and outliers through various algorithms looking for unexpected patterns in the seismic, drilling, completion and production data. Performance of the step 204 facilitates the interpretation of the data in step 206.

[0014] The method 200 further comprises the cognitive computer 102 processing the organized data to observe relationships between various data points and to discard irrelevant data from the data set (step 206). Relationships may be observed, for instance, by dynamically varying one or more parameters and observing changes in one or more other parameters. Numerous such relationships between data may be identified in this manner. Data associated with parameters that are unaffected by changes in other parameters may be discarded. Similarly, data associated with parameters which, when adjusted, cause no changes in other parameters also may be discarded, as both such categories of data may be deemed irrelevant. In this manner, the cognitive computer 102 may produce a refined, organized data set with multiple identified relationships between the data in the set. In assessing the relationships between various data parameters in this step, the cognitive computer 102 may also identify the degree of relevance (e.g., weight) of each parameter on other parameters. Stated another way, the cognitive computer 102 may identify the relative magnitudes of influence that data parameter variations may have on other data parameters. The cognitive computer 102 may use various analysis tools to perform step 206, such as ordinary least squares regressions, random forest, gradient boosting machines, support vector regressions, and kriging models.

[0015] In step 208, the cognitive computer 102 uses the refined, organized data set and the multiple, weighted relationships between the data in the data set to generate a model that expresses the various relationships. The model may, for instance, be expressed in a matrix format, although the scope of this disclosure is not limited as such. The cognitive computer 102 may develop the model using supervised learning, unsupervised learning, and reinforcement learning. In some embodiments, training the model involves selecting one model out of a set of possible models or a Bayesian framework. Modeling techniques may include or use function approximation or regression analysis, including time series prediction, fitness approximation, and modelling. The cognitive computer 102 may use pattern and sequence recognition, novelty detection, and sequential decision-making. The cognitive computer 102 may further use filtering techniques, clustering techniques, and blind source separation and compression. The cognitive computer 102 may use one or more of these tools to develop a model that requires less new well data to plan a fracking job for the new well than would otherwise be the case. Similarly, the cognitive computer 102 may use one or more of these tools to develop a model that requires no new well data to plan a fracking job for the new well.

[0016] In step 210, the cognitive computer 102 validates and refines the model against new and/or existing (i.e., historical) data. Stated another way, the cognitive computer 102 tests the model it produced in step 208 when drilling a new well in a relevant location (e.g., within a predefined radius of the wells for which data were provided in step 202) or against historical data, such as well data collected in step 202 or other well data not yet used in the method 200. In such tests, the cognitive computer 102 may use certain data (e.g., surface log data) as inputs to the model and it may use the model to predict one or more outputs. The cognitive computer 102 may then compare the outputs predicted by the model to the actual outputs of the well(s) to adjust the model. The cognitive computer 102 may repeat this testing process any number of times to repeatedly refine and improve the model.

[0017] In step 212, the cognitive computer 102 uses the model to design a fracturing job in a new well. Relevant inputs are provided to the model and the outputs of the model are assumed as valid and are used to plan and execute a fracking job in a new well or, in some embodiments, in an existing well. The cognitive computer 102 may use the outcomes of such projects to continually refine its model.

[0018] In some examples, a well-developed oil field is to increase production. The target formation has been drilled and completed in many surrounding wells and other stacked reservoir formations have also been drilled and completed. The oil and gas company wants to change or reduce the spacing of the new wells drilled. Historical information is gathered for that field that includes seismic, drilling, completion and production information that includes everything from reservoir log measurements, surface measurements during various operations, written reports with contextual information that writes about issues and successes during the operation, other field development information and any other information, previous modelling, production history, etc., that can be used to interpolate the new wells that will be drilled in the reduced spacing. One goal is the extraction of patterns and information from multiple sources of various field exploration and development data for processing.

[0019] The next step is to process this data and information in such a way that can segment the information into two categories of known and unknown certainty with the structural and contextual data. This process allows the awareness of the information available and understanding of the strength and weaknesses of the data and information. The analytics involve ranking importance and reliability of the information weighting the data for relevance. Due to size of the data sets, this may require sequence tagging, structured hierarchies, sorting similar terms and measurements, reduction of randomness in the data and variables, etc. This can be done with principle component analysis, multilinear subspace learning and generalized discriminant analysis. Anomalies and outliers are also detected and accounted for through various algorithms looking for unexpected patterns in the seismic, drilling, completion and production data. [0020] The next step would be to process the information with a reasoning from both analytical methods and subject matter experts to pare down the information to relevant form for the future wells that will be completed in the reduced spacing of the field. This will be done with statistical analysis using tools such as ordinary least squares regressions, random forest, gradient boosting machine, support vector regression and kriging model. This will lead to predictive input-output models that can be evaluated. It may use artificial neural networks that interconnect the information and data to link inputs and outputs to understand relationships and relevance of information such as reservoir properties, modelling parameters, treatment compatibility with formation, etc. This may be used for supervised learnings, unsupervised learnings and reinforcement learnings. Training the neural network model involves selecting one model out of the set of models or in a Bayesian framework. This may lead to function approximation or regression analysis, including time series prediction, fitness approximation and modelling. It does classification, including pattern and sequence recognition, novelty detection and sequential decision making. The data processing includes filtering, clustering, blind source separation and compression. One or all of these methods will develop a model that will require less new well data to stimulate this new well in a reduced spacing.

[0021] The next step is to validate the models created with historical wells in the field and out of the field of work. The model may treat the well in that field as if it only has the drilling information and no other information to create a job program to be completed and stimulated. The model may use supervised machine learning techniques to adjust the model to tune for best results. The model may be used in any future new well in any mature field being drilled and completed without the need for data except the normal execution data that is received during the drilling process.

[0022]The present disclosure relates generally to maximizing production recovery through optimizations and, more specifically, it relates to an optimized fracturing grid to yield increased production recovery for improved production after a fracturing operation through maximizing relevant inputs, historical data, production history and an automation process. The disclosure generally relates to a maximized production recovery through optimization which includes a method of processing and managing the compilation of contextual, real time and historical relevant reservoir field information to automate completion and stimulation of a new well in an existing mature oil and gas field to ensure optimal performance through the life of the well.

[0023]The disclosure is not limited in its application to the details of construction or to the arrangements of the components set forth in the following description or illustrated in the drawings. The disclosure is capable of other embodiments and of being practiced and carried out in various ways. Also, it is to be understood that the phraseology and terminology employed herein are for the purpose of the description and should not be regarded as limiting.

[0024] An object is to provide an optimized fracturing grid to yield increased production recovery for improved production after a fracturing operation through maximizing relevant inputs, historical data, production history and an automation process.

[0025] Another object is to provide an Optimized Fracturing Grid To Yield Increased Production Recovery that utilizes the past relevant information during the discovery, evaluation and development phase of an oil and gas well to automate the fracturing stimulation design.

[0026] Another object is to provide an Optimized Fracturing Grid To Yield Increased Production Recovery that creates a methodology for maximizing and streamlining inputs by using nearby well trajectories, geologic properties and production historical information to create an automated design based without data from the current well that is being stimulated in the grid.

[0027] Another object is to provide an Optimized Fracturing Grid To Yield Increased Production Recovery that reduces the amount of downhole measurements and formation properties for the job design of the current well being completed and stimulated.

[0028] Another object is to provide an Optimized Fracturing Grid to Yield Increased Production Recovery that uses analytics to match and correlate items of known certainty with unknown certainty. [0029] Other objects and advantages of the present disclosure will become obvious to the reader and it is intended that these objects and advantages are within the scope of the present disclosure.

[0030] Various Tables are presented below. These Tables include:

[0031] TABLE 1: Table 1 is a block Table illustrating the overall of the present disclosure. Phases in an oil and gas field development with data collected.

[0032] TABLE 2: Table 2 is a flowchart illustrating the overall operation of the present disclosure. Data with known and unknown certainty.

[0033] TABLE 3: Table 3 is a flowchart illustrating the overall operation of the present disclosure. Methodology used in getting fracture potential of new wells.

[0034] TABLE 4: Table 4 is a flowchart illustrating the overall operation of the present disclosure. Wellsite Measurements Flow received at the wellsite.

[0035] TABLE 5: Table 5 is a flowchart illustrating a sub-operation of the present disclosure. Wellsite Measurements, Data and modeling Workflow.

[0036] TABLE 6: Table 6 is a flowchart illustrating the overall operation of the present disclosure. Workflow of Job Design that would be automated.

[0037] TABLE 7: Table 7 is a block Table illustrating the overall of the present disclosure. Automated Frac Design Workflow.

[0038] TABLE 8: Table 8 is a flow Table of a method for developing and refining an oilfield operations model using a cognitive computer system, in accordance with various embodiments.

[0039] Overview Turning now descriptively to the Tables, in which similar reference characters denote similar elements throughout the several views, the Tables illustrate a method of processing and managing the compilation of contextual, real time and historical relevant reservoir field information to automate completion and stimulation of a new well in an existing mature oil and gas field to ensure optimal performance through the life of the well.

[0040] Shifting From A New Oil And Gas Field Development Phase To A Mature Oil And Gas Field Development Phase Table 1 shows the shift from collection of data to utilizing historic data to design and complete a well in a mature field. Utilizing analytics and machine learning allows the shift from awareness and learning to go to an automated process. Table 5 shows where measurement is gathered and used in models in the development phase. The disclosure includes a methodology to use a variety of field development information to determine relevant information to reduce future reservoir characterization and data of new well development.

[0041]Utilizing Drilling, Fracturing Treatment Designs And Production History To Create An Automated New Well Frac Design Table 6 shows the methodology of using current drilling information of the new well, historical fracturing treatment and production of nearby wells will create an optimal solution to for new frac design. Using multiple sources of existing nearby and new well drilling information along with previous well stimulation treatments along with historical production information, a machine learning and/or analytic process will generate an optimal frac stimulation design for the new well.

[0042] Automated Frac Design Process Table 7 shows a method of processing and managing the compilation of contextual, real time and historical relevant reservoir field information to automate completion and stimulation of a new well in an existing mature oil and gas field to ensure optimal performance through the life of the well. The disclosure includes a method of processing and managing the compilation of contextual, real time and historical relevant reservoir field information to automate completion and stimulation of a new well in an existing mature oil and gas field to ensure optimal performance through the life of the well.

[0043] Connections of Main Elements and Sub-Elements of Disclosure Part of the data reconciliation will require using analytics to match and correlate items of known certainty with unknown certainty.

[0044] Description of Some Embodiments An optimized fracturing grid is disclosed to yield increased production recovery for improved production after a fracturing operation through maximizing relevant inputs, historical data, production history and an automation process. The optimized fracturing grid to yield increased production recovery generally includes a method of processing and managing the compilation of contextual, real time and historical relevant reservoir field information to automate completion and stimulation of a new well in an existing mature oil and gas field to ensure optimal performance through the life of the well.

[0045] The above discussion is meant to be illustrative of the principles and various embodiments of the present disclosure. Numerous variations and modifications will become apparent to those skilled in the art once the above disclosure is fully appreciated. It is intended that the following claims be interpreted to embrace all such variations and modifications.

Claims

CLAIMS What is claimed is:

1. A system comprising:

a processor; and

storage coupled to the processor, the storage storing executable code which, when executed by the processor, causes the processor to: obtain historical data pertaining to wells within a predetermined area; process the historical data to identify a relationship; and generate a model based on the relationship,

wherein the system comprises a cognitive computing system.

2. A method comprising:

collecting, by a cognitive computer, historical data pertaining to a plurality of wells;

organizing, by the cognitive computer, the collected historical data;

processing, by the cognitive computer, the collected historical data to observe relationships between data points in the collected historical data;

generating, by the cognitive computer, a model expressing said relationships; refining, by the cognitive computer, the model using the collected historical data, new well data, or both; and

using, by the cognitive computer, the model to plan an aspect of an oilfield operation.