RU2417443C2 - Optimisation of solutions for multitude of objects with various basic uncertainties - Google Patents

Abstract

FIELD: oil and gas production.

SUBSTANCE: here is disclosed method for optimisation of solutions for multitude of objects like oil and/or gas deposits. The method includes stages when there is received information setting alternate decisions and variable uncertainties for multitude of objects. This information specifies a corresponding set of one or several algorithms for each object. In the method there is formed a decision vector; the decision vector is evaluated for determination of at least value of a global target function for the decision vector; there is found an optimiser, thus there is created a multitude of decision vectors and corresponding values of the global target function. Data are retained in a memory unit. A row of boreholes is drilled in a multitude of deposits; also, number of boreholes is determined by choice of one of vectors of decision of a reference set.

EFFECT: increased yield of oil and gas.

29 cl, 10 dwg

Description

BACKGROUND OF THE INVENTION

1. The technical field to which the invention relates.

In general, this invention relates to the field of decision making, uncertainty and optimization analysis, and more specifically to a system and method for optimizing decisions related to a system of objects in the presence of uncertainty.

2. Description of the prior art

There is a large group of works that are usually carried out in the process of exploration, development and operation of oil and gas fields (facilities), for example, works such as

land lease;

seismic data recording;

exploratory well drilling;

production wells drilling;

well completion;

installation of equipment for the process and storage of production from wells.

Although large capital expenditures are required to carry out these works, these expenditures are considered as investments in the hope of generating sales revenues (for example, from the sale of oil and gas) that significantly exceed the total cost. Therefore, in this application, these works can be attributed to investment activities.

For each investment activity, various decisions are required, for example, decisions regarding

how much land to rent and what land to rent;

how to record seismic data and how much seismic data to record;

how many exploratory wells to drill and where to drill exploratory wells;

how to develop deposits;

how many productive and injection wells to drill, where to drill wells and what well design (i.e., the path of the wellbore through space) should be for each of the productive wells;

how many perforation intervals to do in each well and how to distribute the location of the perforation intervals along the well structure for each well;

what order of drilling and perforation of wells;

how much to devote to the construction of technological equipment and how to connect wells with equipment and with integrated oil and gas treatment systems;

how to process fluids extracted from the geological formation;

what facilities and in what order to develop and carry out production at the facilities in case of a large number of potential productive facilities.

In addition, these decisions must be made taking into account the large number of fundamental uncertainties, such as the uncertainties of factors such as

futures prices for leasing oil and gas areas;

oil and / or gas reserves in the field;

the shape and physical properties of the reservoirs in each field;

the amount of time that will be spent on drilling each exploratory well and each production well;

the possibility of obtaining equipment in the future, such as drilling equipment and a mobile installation for completing wells;

costs associated with the operation and maintenance of productive and injection wells and equipment;

oil and gas futures prices;

weather resistance.

In any case, when making a decision, these fundamental uncertainties entail uncertainties in gain, namely in production volume, sales volume and profit. Therefore, there is a need for systems and methods that can help in making decisions (each decision has a range of possible options) related to the development of an oil and gas field, taking into account fundamental uncertainties.

SUMMARY OF THE INVENTION

According to one set of implementations, a method for optimizing decisions regarding a large number of objects may comprise steps in which

(a) receive information specifying decision variables and uncertainty variables for a large number of objects and setting for each object an appropriate set of one or more algorithms (for example, algorithms that simulate production of fluids from the reservoir, economic indicators, and other output data for objects);

(b) form a solution vector, wherein the solution vector includes a value for each of the decision variables;

(c) performing an estimation process with respect to the solution vector to determine at least the value of the global objective function for the solution vector;

(d) an optimizer is performed, wherein said optimizer implementation includes the repeated implementation of (b) and (c), resulting in a large number of solution vectors and corresponding values of the global objective function, and the optimizer uses a large number of solution vectors and corresponding the values of the global objective function for the formation and multiple adjustments of the reference set of solution vectors in order to optimize the global objective function;

(e) store data, including a large number of solution vectors and the corresponding values of the global objective function, in a storage device (for example, in the system memory of the host computer).

In addition, the method may include displaying a graphical representation of at least one subset of the solution vectors from the reference set.

Optimization can be implemented using any of a number of optimizers, in particular using a stochastic optimizer. For example, the optimizer may use a forbidden search procedure and / or a scatter search procedure.

The evaluation process operation may include

(c1) the formation of an uncertainty vector, wherein the uncertainty vector includes a value for each of the uncertainty variables;

(c2) the formation for each object of a data set for the corresponding set of algorithms using the corresponding subset of one or more values in the solution vector and the corresponding subset of one or more values in the uncertainty vector;

(c3) initiating the execution of a set of algorithms for each object based on the corresponding data set to obtain output data corresponding to the object;

(c4) calculating for each object one or more values of one or more corresponding values, wherein each of the values of the values for the object is calculated based on at least a portion of the output of the set of algorithms for the object;

(c5) performing operations including (c1) to (c4) several times, resulting in a set of values of each quantity for each object;

(c6) calculating one or more statistical indicators for each object based on the corresponding one of the aggregates; and

(c7) combining the first subset of statistical indicators to determine the value of the global objective function corresponding to the solution vector.

According to some implementations, a subset of one or more value values for an object can be calculated using input data (e.g., data from a data set) as well as output from a set of algorithms for the object.

Information (related to (a) above) may specify one or more functional relationships between uncertainty variables. The operation (c1) of generating the uncertainty vector can be carried out in a way that takes into account one or more functional dependencies between the uncertainty variables.

In addition, the information may specify one or more correlation relationships between the uncertainty variables. In this case, the indicated correlation relationships are taken into account in the operation (c1) of the formation of the uncertainty vector. Uncertainty variables may include two or more subsets related to two or more objects, respectively. Correlation relationships may include correlation relationships between uncertainty variables for various objects.

In addition, the information may specify one or more constraints on the decision variables. In this case, the operation (b) of the formation of the solution vector can be performed in compliance with one or more restrictions on the decision variables.

Information can also define one or more functional dependencies between decision variables. In this case, the operation (b) of the formation of the solution vector can be performed taking into account one or more functional dependencies between the decision variables.

In the operation (c) of the decision process with respect to the solution vector, the corresponding value of the auxiliary function can be obtained in addition to the specified value of the global objective function. Therefore, operation (c) may additionally include (c8) combining a second subset of statistical indicators to determine the value of the auxiliary function corresponding to the solution vector.

The optimizer can use the values of the global objective function and the values of the auxiliary function corresponding to a large number of solution vectors for the specified multiple adjustment of the reference set in order to optimize the objective function subject to one or more constraints.

The user can specify the structure of one or more constraints. For example, the user can set a restriction on the functional combination of the auxiliary function and the global objective function (the functional combination itself can be set by the user). As another example, a user may define a restriction on an auxiliary function.

Operation (c8) may include combining a second subset of statistical indicators in accordance with a linear combination, the coefficients of which are specified by the specified information.

Each of the decision variables has a set of attainable values related to it, representing the possible results of the corresponding solution. In addition, each of the uncertainty variables has a set of attainable values related to it. The sets of reachable values can be finite sets of numerical values, data structures, programs, etc. In addition, the sets of attainable values can be intervals of a real number line or, more generally, user-defined areas in n-dimensional space, where n is a positive integer.

According to one embodiment, the method may also include displaying a graph of the dependence of the value of the global objective function on the value of the auxiliary function for each of the specified large number of decision vectors.

The save operation (e) may include storing solution vectors from a large number of solution vectors together with their corresponding values of the global objective function and the corresponding values of the auxiliary function in the database in the storage device.

According to some implementations, the method may further include

carrying out actions on the database to identify a subset of a large number of decision vectors that have optimal values of the global objective function for a certain value of the auxiliary function;

repeating the specified action for the set of values of the auxiliary function; and

saving subsets identified at the specified repetition.

According to some implementations, the method may also include

taking action on the database to determine the geometric location of the optimal values of the global objective function, taking into account the value of the auxiliary function; and

display of a geometrical place of optimum values.

A graphical representation of a subset of a large number of solution vectors that correspond to a point at a geometric location of optimal values can be displayed in response to a user selecting a point.

The user can set a large number of objects, decision variables and uncertainty variables for each object and a set of one or more algorithms for each object. Decision variables are variables that undergo optimization.

Uncertainty variables are variables that are examined in a random (or pseudo-random) manner to create variations in the input data introduced into the algorithms for each object.

A set of one or more algorithms for at least one of the objects may include an algorithm for estimating oil and gas production over time and an algorithm for evaluating the economic performance of an object. An algorithm for estimating oil and gas production over time can be

a material balance simulator that operates with the data structure of the reservoir model;

calculation of material balance performed in spreadsheet models;

a fluid motion simulator in a reservoir that operates with a three-dimensional data structure (for example, a fluid motion simulator in a reservoir with a finite-difference representation, a flow line based representation, or a finite element representation);

an algorithm defined by a set of one or more user-defined algebraic expressions.

In addition, a set of one or more algorithms for at least one of the objects may include one or more of

algorithms for evaluating the transportation of oil and gas to the surface, for example, a process simulator in a network of onshore pipelines,

algorithms for evaluating the transportation of oil, gas and water to the surface, for example, an analytical model of a network of land pipelines or a set of one or more user-defined algebraic expressions, and

algorithms for estimating the volume of oil and gas and the three-dimensional properties of the reservoir.

An algorithm for estimating the volume of oil and gas and the three-dimensional properties of the reservoir can be implemented using

geostatistical algorithm for the formation of the properties of the geological environment;

an algorithm for the formation of the structural structure of the geological environment;

an algorithm defined by a set of one or more user-defined algebraic expressions; or

a simulator of material balance.

The algorithm for calculating the economic indicators of the object can be implemented as

spreadsheet-based algorithms, which may include full fiscal models and production sharing models;

algorithms based on analytical economic models and a set of one or more user-defined algebraic expressions.

According to another set of implementations, a method for optimizing decisions regarding a large number of objects by executing program code on a first computer (eg, a server computer) may include operations in which

(a) receive information defining decision variables and uncertainty variables for a large number of objects and defining for each object an appropriate set of one or more algorithms;

(b) form a solution vector, wherein the solution vector includes a value for each of the decision variables;

(c) performing an estimation process with respect to the solution vector to determine at least the value of the global objective function for the solution vector;

(d) an optimizer is performed, wherein said optimizer implementation includes the repeated implementation of (b) and (c), as a result of which a large number of solution vectors and corresponding values of the global objective function are generated, while the optimizer uses a large number of solution vectors and corresponding the values of the global objective function for the formation and multiple adjustments of the reference set of solution vectors in order to optimize the global objective function;

(e) store data, including a large number of solution vectors and their corresponding global objective function values, in a storage device.

Solution variables may include variable numerical solutions and scenario decision variables. A numerical solution variable is a solution variable whose attainable values are numerical values. For example, a unique decision variable can be used to represent possible results for a number of wells to be drilled at a particular site. A scripted decision variable is a decision variable whose reachable values are data structures that describe the corresponding scenarios. For example, a scenario solution variable can be used to represent a solution between different scenarios, with each scenario representing a different set of alternatives for a number of wells at a particular site, well drilling schedules, and well drilling geometries.

The value of the global objective function can be calculated on the basis of output data generated using one or more algorithms that operate with decision variables. According to some implementations, the value of the global objective function may be the total carrying amount of oil or gas, or total production, or net present value. According to another implementation, the value of the global objective function may be statistical, for example, the average value of stocks or net present value. According to some implementations, other statistical indicators of the value of the global objective function are needed to match, for example, the standard deviation condition or any quantile value.

Operation (c) of the evaluation process may include

(c1) the formation of N _I uncertainty vectors, with each uncertainty vector containing the assigned values for each of the uncertainty variables, where N _I is a positive integer;

(c2) the formation of N _I sets of iterative data for each object, wherein each of the sets of iterative data is formed using part of the solution vector and part of the corresponding one of the uncertainty vectors;

(c3) distributing N _I sets of iterative data for each object to one or more secondary computers to initiate the execution of sets of algorithms for the object based on each of the corresponding N _I sets of iterative data, and each time a set of algorithms for the object is executed based on the set of iterative data corresponding set of output data;

(c4) receiving N _I sets of output data for each of the objects from one or more secondary computers and storing N _I sets of output data in a storage device;

(c5) calculating for each object and each of N _I sets of output data corresponding to the object, one or more values of one or more corresponding values based on data in the set of output data, resulting in a collection of N _I values of each value for each object;

(c7) calculating one or more statistical indicators for each object based on the corresponding one of the aggregates; and

(c8) combining the first subset of statistical indicators to determine the value of the global objective function corresponding to the solution vector.

Some of one or more values calculated for each object can be calculated on the basis of input data (for example, data of the corresponding set of iterative data), as well as output from a set of algorithms for the object.

The first computer and one or more secondary computers can be connected using a computer network, such as a local area network, intranet, wide area network, Internet, etc.

At the stage of the operation of distributing sets of iterative data, distributed execution of sets of algorithms based on sets of iterative data on a large number of secondary computers is initiated. Secondary computers can be combined into a network structure.

According to one embodiment, a subset of the iterative data sets can be performed locally using the first computer, which partly depends on the available performance of the data processing system on the first computer.

Information can be obtained from the user during the process on the client computer and transmitted to the first computer during this process.

According to another set of implementations, a method for optimizing decisions regarding a large number of objects may include operations in which

(a) receive information specifying a large number of objects, the objective function of the object for each of the objects, and a set of algorithms for each object;

(b) form a solution vector including values for a set of decision variables;

(c) form N _I uncertainty vectors, wherein each of the N _I uncertainty vectors is the result of a separate specification of a set of uncertainty vectors, where N _I is a positive integer;

(d) calculate the first set of N _I values for each objective function of the object, performing N _I iterations of the execution of the corresponding set of algorithms and operating with the output data generated at each of the N _I iterations of the specified execution, and with each execution of the corresponding set of algorithms they operate with the input data determined using the values of the decision variables and using the corresponding one of the uncertainty vectors;

(e) calculating the first statistical indicator of the object for each objective function of the object based on the corresponding set of values;

(f) calculating the value of the global objective function based on the first statistical indicators of the object;

(g) an optimizer is performed, wherein said optimizer implementation includes performing (b) to (e) multiple times, resulting in a large number of solution vectors and corresponding values of the global objective function, and where the optimizer uses a large number of solution vectors and corresponding values of the global objective function for the formation and multiple adjustments of the reference set of solution vectors in order to optimize the global objective function.

The method may also include operations

calculating a second statistical indicator of the object for each object according to the corresponding one of the populations; and

calculating the value of the auxiliary function based on the second statistical indicators of the object.

When performing the optimizer, an attempt is made to optimize (that is, maximize or minimize) the global objective function, subordinate to the restriction on one or more functions from the auxiliary function and the global objective function.

By using the global objective function, which is a functional combination (for example, a linear combination with positive weights), the methods of the various embodiments described in this application can be implemented to optimize decisions regarding a large number of objects taken together. In the methods of the embodiments, operations can be performed on uncertainty variables including dependencies on various objects. In addition, by methods according to an implementation, a user may be given the opportunity to set constraints on decision variables, including restrictions on decision variables on various objects.

Methods (or parts thereof) according to any or all of the implementations described in this application can be implemented based on program instructions. Program instructions can be stored on various types of computer-readable media, such as magnetic disk, magnetic tape, various types of semiconductor memory, cylindrical magnetic domain memory, read-only compact disc, etc.

Program instructions may be stored in a storage device of a computer system (or in storage devices of a group of computer systems). A computer system processor may be configured to read and execute program instructions from a storage device and, thus, to implement methods according to an implementation (s) or parts of methods.

In addition, program instructions (or subsets of program instructions) can be transmitted through any of a variety of transmission media, for example, over a computer network.

Application for registration number 10/653829 for US patent, filed 09/03/2003 under the name "Method and system for scenario and case decision management", in which the inventors are Cullick, Narayanan and Wilson, is fully incorporated into this application by reference.

Brief Description of the Drawings

A better understanding of the present invention can be achieved by considering the detailed description presented in this application, in combination with the accompanying drawings, in which:

figure 1 is an illustration of the modules of the client application and the streaming server according to one set of implementations;

figure 2 is an illustration of a flow according to one series of implementations configured to perform an analysis of optimization decisions related to a large number of objects in the presence of various fundamental uncertainties (which are modeled by uncertainty variables);

figure 3 is an illustration of the optimizer 210 of figure 2 according to one series of implementations;

figure 4 is an illustration of the evaluation process 220 of figure 2 according to one series of implementations;

figure 5 is an illustration of a method for optimizing decisions related to a large number of objects, according to one series of implementations;

figure 6 is an illustration of the operation 520 (that is, the operation of the evaluation process) of figure 5 according to one series of implementations;

figure 7 is an illustration of a method for optimizing solutions according to another series of implementations;

figure 8 is an illustration of the operation 730 (that is, the operation of the evaluation process) of figure 7 according to one series of implementations with emphasis on the action of distributing sets of iterative data to another computer (s) for execution on another computer (s);

figure 9 is an illustration of a method for optimizing solutions according to another series of implementations;

figure 10 is a structural diagram of a computer system according to one set of implementations, which can serve as a server computer in some implementations and a client computer in some implementations.

Although various modifications and alternative forms of the invention are allowed, specific embodiments thereof are shown by way of example in the drawings and are described in detail in this application. However, it should be understood that the drawings and detailed description are not intended to limit the invention to the particular forms disclosed, but rather, the invention covers all modifications, equivalents, and variations that fall within the spirit and scope of the present invention as defined by the appended claims.

DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS

In one set of implementations, a decision management system may be configured to provide a user with means (e.g., graphical user interfaces) for

(a) constructing an arbitrary flow;

(b) constructing a set of models including descriptions of a certain number of uncertainties and a certain number of solutions;

(c) defining a data set for execution control, including information for qualifying and controlling flow execution;

(d) initiating flow execution based on a set of models and a set of data to control execution; and

(e) analyzing the results of the execution of the stream.

A stream is a set of interconnected algorithms designed to solve a problem. A decision management system may provide a graphical user interface that allows a user to select, import, or create algorithms and link algorithms to each other to form a stream.

A stream can be arranged to include one or more computational cycles. Loops can be nested. Flow loops can be used for any of a variety of purposes. For example, a stream may include a loop for optimizing over a space corresponding to a number of variables. Variables that are optimized are referred to in this application as decision variables. As another example, a flow may include a cycle to investigate the effect of variation in space corresponding to a number of variables that represent uncertainties. Variables that represent uncertainties are referred to herein as uncertainty variables.

The loop may include an iterator, an algorithm (or, in a more general case, a library of algorithms) and, if desired, a dispatcher. When executing the stream, the iterator is responsible for generating sets of iterative data for the corresponding executions of the algorithms. Each execution of the algorithm is based on the corresponding one of the sets of iterative data. Each iterative data set contains data used as input for the algorithm. In this application, each execution of the algorithm is referred to as an “iteration”. However, the term “iteration” does not mean the requirement of sequential processing of one iteration after another. In fact, the iterator and dispatcher can be configured to perform parallel iterations. For example, in anticipation of an appropriate number of iterations, the iterator may be configured to form a number of sets of iterative data. The dispatcher can detect other computers (or processors) that have an unoccupied range of computing capabilities, and distribute iterative data sets to other computers so that iterations can be performed in parallel or, depending on the number of free computers, at least partially in parallel.

According to one embodiment, the dispatcher can compress one or more sets of iterative data to form an executable file and transfer the executable file to one of the remote computers (having an unoccupied range of computing capabilities). On the remote computer, the executable file can be unpacked to restore the sets of iterative data and the algorithm is executed, once for each of the sets of iterative data. The algorithm may include code that transfers the results of each execution back to the first computer (that is, to the computer on which the dispatch program is running) for storage.

As noted above, an executable file may include compressed sets of iterative data. If the remote computer does not yet have a copy of the executable code for the algorithm, the dispatcher can compress the copy of the executable code and include the compressed code in the executable file.

As indicated above, a stream is a set of interconnected algorithms designed to solve a problem. Algorithms work on the basis of data structures. Algorithms of various kinds work on the basis of data structures of various kinds. For example, a reservoir fluid simulator designed to predict oil, gas and water production volumes operates with a different set of data structures than the economic algorithm designed to predict economic returns. A decision management system may provide an interface through which a user can specify locations of file sources (e.g., locations of a file system on a network) and / or databases containing data structures. In addition, the decision management system may initiate tools that allow the user to create and modify data structures.

Each data structure includes a plurality of one or more data values. However, some data values from the data structure may be parameters or values that are uncertain. Therefore, a decision management system can provide the user with a mechanism for selecting the data value in the data structure and for moving the data value into the uncertainty variable by introducing the set of achievable values for the uncertainty variable and the probability distribution for the uncertainty variable. (By default, the original data value may be included as one of the values in the set of achievable values of the uncertainty variable). Any number of data values from any number of data structures can thus be advanced into uncertainty variables.

In addition, the value of the data in the data structure may represent one possible outcome of the decision. The user may want to consider other possible outcomes of the solution. Therefore, a decision management system can provide the user with a mechanism for selecting the data value from the data structure and for promoting the data value in the decision variable by introducing a plurality of achievable values representing possible decision results. (For example, the number of wells to be drilled in a given field can be equal to any integer from A to B. Including. As another example, the productivity of a particular pump can be any value in the range from X to Y, expressed in cubic feet per second ) By default, the original data value can be included as one of the elements of the set of reachable values. Any number of data values from any number of data structures can thus be promoted into decision variables.

The set of attainable values for a solution variable or a variable of uncertainty can be a finite set (e.g., a finite set of numerical values specified by the user) or an infinite set (e.g., an interval of a real line or region in n-dimensional Euclidean space, where n is a positive integer).

The user can set the probability distribution for the variable uncertainty in various ways. If the uncertainty variable has a set of achievable values that is continuous, the decision management system can provide the user with the opportunity to set the probability distribution density (PRV) by choosing the type of probability distribution density (for example, normal, uniform, triangular, etc.) and set the parameters that determine the specific probability distribution density within the selected type. If the uncertainty variable has a set of achievable values, which is finite, the decision management system can provide the user with the ability to enter probability values (or numerical values that can later be normalized with respect to probability values) for the achievable values of the uncertainty variable. The decision management system may provide the user with the opportunity to bring the probability distribution density into qualitative and quantitative agreement with the existing set of values for the uncertainty variable (possibly by analogy).

A case is possible when there are two or more data structures, each of which can be introduced as input into a specific algorithm (or a specific set of algorithms). The user may not know exactly which of two or more data structures to use. For example, for the same reservoir, there may be two geocellular models, and it may not be clear which model is the best representation of the physical reservoir. Therefore, the user may want to create a thread in which the algorithm is executed several times, each time using a randomly selected one of two or more data structures. Therefore, a decision management system may provide the user with a mechanism for generating a variable uncertainty whose achievable values are consistent with two or more data structures, respectively. The user can assign probability values to two or more attainable values. Each probability value represents the probability that the corresponding data structure will be selected from any particular implementation of the uncertainty variable. The user can make a choice to form any number of uncertainty variables of this kind. A set of data structures that are thus converted to attainable values of the uncertainty variable is called a "subordinate to the uncertainty variable."

In addition, there may be a case where the data structures are two or more, while they represent alternative solutions. For example, there may be two or more data structures, each representing a possible implementation for well locations and equipment piping connections. Therefore, a decision management system may provide the user with a mechanism for generating a decision variable whose achievable values are consistent with two or more data structures, respectively. A set of data structures that are thus converted to reachable values of the decision variable is called a “subordinate of the decision variable”.

Therefore, the decision management system provides the user with the opportunity

(1) specify source locations for data structures, create data structures, and modify data structures;

(2) to advance data values from data structures into uncertainty variables and decision variables;

(3) subordinate data structures to uncertainty variables;

(4) subordinate data structures to decision variables.

Some data structures can be categorized as models because they describe (characterize) systems of interest. For example, data structures may include geo-cellular reservoir models, well design models, etc. Objects arising from operations (2), (3) and (4) are also referred to as models in this application.

A decision management system can be configured to support the use of very accurate physical reservoir models, well models, production flow models, and economic models.

Some of the uncertainty variables may be correlated. Therefore, the decision management system provides the user with the opportunity to set correlation relationships between pairs of uncertainty variables. For example, a user can select a pair of uncertainty variables and enter a correlation coefficient to specify a correlation between a pair of uncertainty variables. In some embodiments, a user may select a group of two or more uncertainty variables and enter a correlation matrix for the group. When the flow is executed, the uncertainty variables are specified in a way that takes into account the correlation relationships specified by the user.

Some uncertainty variables can have functional dependencies among themselves, for example, functional dependencies of the form Y = f (X), Y = f (X ₁ , X ₂ ), Y = f (X ₁ , X ₂ , X ₃ ), etc. where X, X ₁ , X ₂ , X _3, and Y are uncertainty variables. Therefore, a decision management system may provide a user interface that allows the user to define such functional dependencies. Any of a wide variety of functions (including linear and non-linear functions) can be supported. According to some implementations, the user interface allows the user to construct an arbitrary algebraic function of the uncertainty variables. When a thread executes, the uncertainty variables are instantiated in a manner in which these functional dependencies are taken into account. For example, if Y = f (X), the concretization for Y can be obtained first of all by concretizing X and then evaluating the function f based on the concretization of X. Similarly, decision variables can have functional relationships among themselves, for example, functional dependencies of the form Y = f ( X), Y = f (X ₁ , X ₂ ), Y = f (X ₁ , X ₂ , X ₃ ), etc., where X, X ₁ , X ₂ , X ₃ and Y are solution variables. Therefore, the same user interface (or another user interface) can be used to specify such functional dependencies between decision variables.

Some of the decision variables or functional combinations of decision variables may be subject to constraints. Therefore, a decision management system may provide a user interface that allows the user to set restrictive conditions, such as

Y ″ inequality ″ X,

Y ″ inequality ″ f (X),

Y ″ inequality ″ f (X ₁ , X ₂ ),

Y ″ inequality ″ f (X ₁ , X ₂ , X ₃ ),

etc.,

to decision variables such as X, X ₁ , X ₂ , X _3, and Y, where ″ inequality ″ can be any of the inequality operators, <, ≤,> or ≥. Any of a wide variety of f functions (including linear and non-linear functions) can be supported. According to some implementations, the user interface allows the user to construct an arbitrary algebraic function of the decision variables. When the flow is executed, optimization is carried out, in which the limiting conditions set by the user on the decision variables are taken into account.

In addition, a decision management system may provide a user interface through which the user can define a data set for execution control. The data set for execution control contains information for regulating and qualifying the flow. For example, a data set for execution control may include information such as

a way of concretizing the variables of uncertainty in the stream; and

number of iterations for the corresponding cycles in the stream.

In the Monte Carlo concretization method, the uncertainty variables (or some subset of the uncertainty variables) can be instantiated concretely by using random number generators. In the Latin hypercube-based concretization method, uncertainty variables (or some subset of the uncertainty variables) can be instantiated using the well-known Latin hypercube method. In the concretization method based on the design of the experiment, the uncertainty variables (or some subset of the uncertainty variables) can be instantiated using any of various well-known experimental design methods.

Over time, the user can design a certain number of threads, a certain number of sets of models and a certain number of data sets to control the execution. The decision management system may provide the user with the ability to select a stream, a set of models and a data set to control the execution and initiate the execution of the stream, taking into account the data set for execution control and a set of models. In addition, the decision management system may provide the user with greater control over how the flow can be performed. For example, according to some implementations, the stream may be performed to perform specific iterations, which for various reasons may adversely affect the previous execution of the stream. According to other implementations, additional iterations can be performed on the threads that were performed previously.

The decision management system may provide the user with various means for creating flows of various kinds. For example, according to some implementations, the decision management program may include a decision flow builder that is configured to automate the flow construction process in order to optimize decisions related to developing a set of objects (e.g., a set of oil and gas fields) in a manner that takes into account various uncertainties, related to a set of objects.

Client server architecture

A decision management system can be implemented by executing a decision management program in one group of one or more computers.

The decision management program can be divided into a client application and a server application with a communication protocol between the client application and the server application. For example, a communication protocol can be implemented using a remote method call (RMI) or Web services.

The client application can run on the client computer, and the server application can run on the server computer. However, the condition that the client application and the server application must be run on separate computers is not necessary. If the computer has sufficient computing power, the client application and the server application can run on that computer.

The server application can be configured to use dispatchers. A dispatcher is a program (or system of programs) configured to distribute independent groups of calculations (for example, iterations of a large task) among other processors (or computers) for execution on these other processors. Other processors may be distributed over a network, such as a local area network, wide area network, or the Internet. In addition, other processors can be part of a shared memory hardware device, such as a cluster, and can be used as a single unit (for example, in a situation of parallel computing).

In some embodiments, the core modules of the client application and the server application (i.e., the streaming server) may be as shown in FIG. 1.

Client and server applications

The client application 100 may include a stream builder 107, a model manager 102, a stream executor 109, and an output analyzer 112.

The builder 107 threads provides a user interface through which the user sets the task and creates a thread to solve it.

The model manager 102 provides an interface through which the user can create, build, and modify models that include decision variables and uncertainty variables.

Thread artist 109 is used to initiate thread execution. In response to a user request for thread execution, the thread executor sends the stream along with a compatible set of models and a compatible set of execution control data to the stream server so that the stream can be executed. In addition, the thread executor can receive status information from the streaming server related to the currently executing streams and successive streams, and display status information to the user. The streaming server may send the results of the stream to the stream executor and / or store the results in a database.

The streaming server may include a remote application programming interface 160 (API) for communicating with client applications, such as client application 100.

The output analyzer 112 provides the user with the ability to analyze data based on the results of the stream.

Stream server 155 may execute threads (or control the execution of threads) that are transmitted as jobs from the client application. The streaming server 155 may provide information about the status of the jobs being performed to the streaming agent. In addition, the streaming server provides the ability to remove tasks.

Client Application Modules

As noted above, the client application can be divided, as shown in figure 1, into a number of modules. Each module can have an application programming interface (API) related to it, and various modules communicate through these interfaces. Thus, the client application may include a model manager API 103, a decision flow builder API 106, a flow builder API 108, a flow executor API 110, and an output analyzer API 113.

According to some implementations, these APIs are only capable of interfacing between modules, so the modules can be developed independently.

In addition, each of the modules, which requires communication with external applications, can have a remote application programming interface. Therefore, according to some implementations, the client application may include a remote model manager API 104 and a remote thread executor API 111.

Model manager

Model manager 102 may provide an interface that allows the user to import, create, edit, and maintain models, for example, models that include uncertainty variables and decision variables described above. The model manager can also accompany models in projects or other user-defined structures.

The types of data structures to be included in the models intended for a particular stream depend on the algorithms included in the stream. Therefore, the model manager can communicate with the stream builder to ensure that the user supplies data structures that are consistent with the algorithms of a particular stream. For example, a list of algorithms in a stream may be received from the flow builder by the model manager. In this case, the model manager can focus on requesting from the user the types of data structures suitable for the above algorithms. Conversely, models selected by the user can determine the type of stream that can be executed. Therefore, if the user creates and enters a data structure, then in this case the user may be recommended to select only those threads that are capable of executing this model. In some embodiments, the model manager may be configured to run well-known software tools to create and edit certain types of models. For example, each model may be of a specific type, and each type may have a specific tool registered for enabling the creation of a model, editing, viewing, etc. Such a solution can be similar to JavaBean property editors, which provide a specialized graphical user interface that is registered using JavaBean technology. In addition, some of the views within the framework of the activation framework using JavaBeans technology can be used in such a solution.

Since there is no need for user-supplied data structures to include decision variables or uncertainty variables, the model manager can provide generic mechanisms for moving data values from data structures to decision variables and uncertainty variables together with data structures and to subordinate data structures decision variables and uncertainty variables.

The model manager can function as a situation manager, allowing you to clone an existing model and change subcomponents.

Thread builder

Algorithms are building blocks of threads. The algorithm can be considered as a process that has input data and forms output data. A stream is a set of algorithms that are interconnected to solve a specific problem. It should be noted that, if desired, the stream itself can be converted into an algorithm.

The thread builder 107 provides a user interface with which the user can create arbitrary threads. For example, the stream builder may provide the user with the option of selecting algorithms from the set of available algorithms and linking the algorithms to each other (that is, linking the output of the algorithms to the input of other algorithms) to form the stream. The set of available algorithms may include a number of algorithms useful for planning exploration, development and / or exploitation of a number of fields, for example, algorithms such as geological or geostatistical simulators, material balance simulators, reservoir fluid flow simulators, tubing simulators, simulators processes in equipment, pipeline network simulators and economic algorithms. In some embodiments, simulators can be performed using one or more tools, either commercial or open source (for example, a Landmark simulator to increase the cost-effectiveness of a collector or other simulator based on a finite difference or finite element method, or a streamline simulator, a simulator Petex GAP onshore pipeline, Microsoft's Excel, Petex MBAL material balance simulator, GSLIB geostatistical generator).

In addition, the thread builder may provide the user with the opportunity to build algorithms from the working memory or import programs that can be used as algorithms. According to some implementations, the flow builder may provide the user with the option of defining an arbitrary algebraic function as an algorithm or importing a neural network for use as an algorithm.

According to one set of implementations, when forming directional relationships between algorithms, the flow builder may impose restrictions on the type of data. For example, the stream builder may allow the attachment of output integer data from one algorithm to the floating-point input of the second algorithm, but the attachment of floating-point output to the input integer may be prohibited.

The thread builder can use a search engine to find algorithms. The search engine can support searches by category of algorithm (such as “mining dynamics algorithm”), type of input, type of output, etc.

In some embodiments, the stream builder may be configured using one or more commercial tools (e.g., a Landmark decision management system in a decision space).

Once created, algorithms and threads can be registered by the model manager as

suitable for use by the user;

suitable for use by team members working together on a project;

suitable for use in the company;

ready for external distribution to customers.

Stream executor

The thread executor 109 may provide a user interface that allows the user to initiate thread execution. When the user issues a command to initiate the execution of a stream, taking into account a set of models and a data set for controlling execution, the stream executor composes a task including a stream, a set of models and a data set for controlling execution, and submits the task to the streaming server. The streaming server can generate a unique identifier for the job and forward the job identifier to the stream executor. In the thread executor, the job identifier can be written.

In addition to receiving streams from the stream builder and models from the model manager, the stream executor can also download a description of the stream and its related models from a set of input files (for example, XML files) specified by the user. The thread executor can display a list of tasks that includes tasks that are currently being performed (including their current status) and tasks that are waiting. The thread executor interface allows the user to take off tasks. In addition, the thread executor’s interface can allow the user to access any existing task results. Results can be displayed through the user interface of the output analyzer.

The streaming artist can be configured to direct jobs to any of a large number of streaming servers. Streaming servers can be created from a group of computers connected through a network. The streaming agent may provide the client with the opportunity to decide which streaming server the job should be directed to. To perform tasks of various kinds, various streaming servers can be made to order.

Output analyzer

The main task of the output analyzer 112 may be to provide the user with a means to comprehend the output obtained from the task (i.e., the stream, taking into account a given set of models and a data set to control the execution). The output can be complex because the stream can have a complex structure, for example, a structure with an outer loop and multiple inner loops.

The output analyzer may be configured to allow the user to examine the output from the executed stream and observe the results from the stream still running on the stream server.

The output analyzer can have access to the entire set of data analysis tools, and is also able to format the output so that it can be analyzed in a third-party tool (such as Microsoft Excel).

Result Calculator Interface

The client application may also include a result calculator interface (not shown in the drawings). The result calculator interface allows the user to specify special calculations that must be built into the stream. Special calculations can be useful when the output of the first algorithm (or the first set of algorithms) is not exactly the input required for the second algorithm (or the second set of algorithms), but can be used to calculate the input data needed for the second algorithm. The result calculator interface allows the user

select one or more output arguments from the first set of one or more flow algorithms and

define a function (for example, an arbitrary algebraic function) of one or more output arguments;

set the input argument of the second algorithm (different from one or more algorithms from the first set), that is, get the result of evaluating the function for the selected output arguments.

In the flow builder or decision flow builder, this information can be used to form a result calculator algorithm that implements a given function so that the result calculator algorithm is embedded in the stream between the first set of algorithms and the second algorithm. The user can specify any desired number of special calculations so that they are performed (during the execution of the stream) between the algorithms or sets of algorithms in the stream.

Stream server

Stream server 155 may be divided, as shown in FIG. 1, into a number of separate modules. The streaming server can be configured to perform tasks. The task includes a stream and a set of input data.

Depending on the number of jobs, the streaming server can execute them in parallel, while the streaming server can set a queue for waiting for executions. A streaming server may have the ability to remove tasks or change the priority of tasks.

The streaming server can be equipped with a simple user interface (for example, in HTML form).

Unlike a client application, a streaming server can usually run on another computer. However, the streaming server can also be run on the same computer as the client application.

The stream server maintains a dialogue with the stream executor, informing the stream executor about the status of tasks, etc.

The connection between the stream executor and the streaming server can be based on any of a wide variety of known protocols, such as remote method call (RMI) or Web services.

Algorithms, Models, Iterators, and Dispatchers

A decision management system is configured to execute threads that include a set of interconnected algorithms. A flow can be performed based on models with related uncertainty variables and decision variables as input. For example, models can include data values that are advanced in decision variables, data values that are advanced in uncertainty variables, data structures that are subordinate to decision variables, data structures that are subordinate to uncertainty variables, or any combination of them.

A stream may include loops configured to iteratively execute sets of algorithms. Dispatchers can be used to distribute iterations across multiple computers for parallel execution.

By default, a dispatcher can be provided that performs all iterations of the related cycle on the local host computer.

In addition, a network dispatcher may be provided that searches for free computers on the network and distributes iterations to these free computers.

In addition, a metacomputer network manager may be provided that distributes iterations over a network of computers.

It is very likely that the number of models, especially models that are constantly in memory, and the number of algorithms that work in accordance with the models will increase over time. Therefore, the decision management system can be configured to block at its discretion the influx of new models and algorithms.

According to some implementations, the decision management system may have a command type interface. User input routed through the command type interface can be collected and stored. If desired, this provides a way to repeat a set of interactions.

Solution Flow Builder

A decision management client application may also include a decision flow builder. The decision flow builder may provide a user interface that accelerates the flow construction process in order to optimize decisions related to exploration, development and operation of facilities such as oil and gas fields.

In some embodiments, the decision flow builder may be specifically configured to speed up the design of the flow, including an external optimization loop and a set of internal “uncertainty studies” cycles, for example, one internal cycle per object.

The decision flow builder may provide a graphical user interface through which the user can

a) set the names of objects to be analyzed;

b) specify a view for each of the objects, for example, from species such as an exploratory field, an discovered field and an exploited field;

c) define one or more algorithms (for example, process simulation algorithms) to be included in the internal loop for each object;

d) specify the sources of data structures consistent with the algorithms specified in step (c);

f) to perform operations on data structures in order to build models for each of the objects, while the models may include uncertainty variables and / or decision variables described above;

f) set global decision variables and global uncertainty variables;

g) define functional relationships between uncertainty variables and functional relationships between decision variables, including functional relationships within and across objects;

h) define correlation relationships between uncertainty variables, including correlation relationships between uncertainty variables within and across objects;

i) define constraints on decision variables or functional combinations of decision variables;

j1) define one or more variables of interest for each object A (k), k = 1, 2, ..., N _A , where each of the quantities represents a functional combination of one or more output arguments of algorithms related to object A (k) ;

j2) set the statistical indicators to be calculated for each object A (k), k = 1, 2, ..., N _A , based on one or more specified values for the object A (k);

k1) define the global objective function S in the form of a functional combination from the first subset of statistical indicators;

k2) define one or more auxiliary functions, with each auxiliary function being a functional combination corresponding to a subset of statistical indicators;

k3) specify one or more restrictions, including one or more auxiliary functions and / or global objective function;

l) select an optimizer (i.e. from a set of supported optimizers) for an external optimization cycle;

m) select a dispatcher from the set of supported dispatchers;

n) specify whether optimization should be maximization or minimization.

In accordance with the information specified by the user, the decision flow builder can form a flow having an external optimization cycle and a set of internal cycles.

As indicated in (b) above, the types of facilities may include an exploratory field, an discovered field and an exploited field. Other types of objects are also contemplated.

As indicated in (c) above, the user can specify one or more algorithms (for example, process simulation algorithms) to be included in the internal loop for each object. The builder of the decision flow can receive user input specifying a specific object that is currently to be addressed, and display a list (or some graphical representation) of a set of algorithms corresponding to the type of the object in question, from which the user can make a choice. Different kinds of algorithms correspond to different objects.

As indicated in (d) above, the user can specify the sources of data structures consistent with the algorithms specified in (c). For example, the decision flow builder interface may be configured so that a user can set file system locations (for example, file system locations on a computer network) for each data file or database containing data structures. In addition, the user can specify information inputs (or sets of information inputs) to databases containing data structures.

Examples of data structures for an object include

(1) subsurface (underground) structural architecture, geometry and volume of the object;

(2) an underground three-dimensional distribution of the properties of an object, including, but not limited to, pore space, oil and gas saturation, water saturation, geological lithology, gaps and cracks;

(3) the physical location of the object relative to the surface of the sea or the earth's surface and the geographical location of the object;

(4) the number of wells and the location of wells at the facility that can be drilled in the future;

(5) the number of wells that have already been drilled at the facility;

(6) the volume of hydrocarbons associated with the asset;

(7) potential reserves associated with the asset;

(8) a picture of the change in time of oil, gas and water production correlated with the object;

(9) a picture of the change in the time of injection of gas and water, correlated with the object;

(10) types of equipment that relates to the facility; and

(11) types of equipment that can be further manufactured for the facility.

Other examples of data structures include reservoir models, economic models, pipeline network models, process equipment models, and well tubing models.

The types of data structures that are specified for an object may depend on the types of algorithms that should be used for the objects. For example, a simulator of material balance, consisting of analytical or algebraic expressions, can be based on a “reservoir” model for an object, while the reservoir model has a single set of properties to represent the entire object. To predict the flow of oil, gas and water over time, a simulator of fluid motion in a reservoir with the representation of partial differential equations of fluid derivatives in porous media and with a finite-difference formulation can be based on a three-dimensional spatial model of the reservoir. The economic algorithm for predicting profit over time can be based on an economic model that includes prices, tax rates, and fiscal complexities. The builder of the decision flow can prompt the user to enter data structures of the kind that is consistent with the algorithms already defined for the object in (c).

As indicated in (e) above, the applicant may perform operations on data structures in order to build models for each of the objects, especially for models that include uncertainty variables and decision variables. For example, a user may

push data values from data structures into uncertainty variables;

push data values from data structures to decision variables;

Subordinate data structures to uncertainty variables and

Subordinate data structures to decision variables.

Examples of solution variables for objects include

number of wells;

surface location of wells;

diameter of pipelines;

volume of compressors;

type of well (having achievable values, such as horizontal, vertical, or multi-hole);

well geometry;

well drilling path (in the reservoir related to the object);

type of platform (having achievable values, such as underwater and on guy lines);

a decision variable that obeys a set of well drilling schedules;

a decision variable that obeys a set of well completion schedules;

a decision variable that obeys a set of equipment commissioning schedules.

Examples of uncertainty variables for objects include

hydrocarbon volume;

hydrocarbon productivity;

cost of wells and / or equipment; and

oil and gas prices.

Global decision variables are decision variables that are ambiguously associated with a single entity. For example, the amount of capital available to a group of objects is an example of a global decision variable. Global uncertainty variables are uncertainty variables that are ambiguously associated with a single entity. For example, the price of oil is an example of a global variable of uncertainty.

As indicated in (g) above, the user can define functional dependencies between uncertainty variables and functional dependencies between decision variables. In some embodiments, the decision flow builder interface may be configured to enable a user to

selecting an uncertainty variable (from a set of V _UN uncertainty variables related to the current task) as a dependent variable and selecting one or more other uncertainty variables (from a set of V _UN ) as independent variables; and

assigning a function (for example, an arbitrary algebraic function) of independent variables to find a functional relationship between independent variables and a dependent variable.

The set V _{UN of the} uncertainty variables related to the current task includes the uncertainty variables defined in (e) and global uncertainty variables.

In addition, the decision flow builder interface can be configured to allow the user to

selecting a solution variable (from a set of V _D solution variables related to the current task) as a dependent variable and selecting one or more other solution variables (from a set of V _D ) as independent variables; and

defining a function (for example, an arbitrary algebraic function) of independent variables to determine the functional relationship between independent variables and the dependent variable.

The set V _{D of the} uncertainty variables related to the current task may include global decision variables and solution variables defined in (e).

The decision flow builder can provide the user with the ability to specify any number of functional dependencies by reusing the decision flow builder interface described above. For example, a user can define functional dependencies of the following types:

the functional relationship between the uncertainty variables (or decision variables) within the object, that is, when the dependent variable and one or more independent variables relate to the same object;

the functional relationship between the uncertainty variables (or decision variables) for the objects, that is, when the dependent variable and one or more independent variables relate to two or more objects; and

a functional relationship between uncertainty variables (or decision variables), where the dependent variable and / or one or more independent variables are global variables.

An “algebraic function” is a function that is defined by an expression that includes only algebraic operations on the arguments of the function, such as addition, subtraction, multiplication, division, extraction of an integer (or rational) root and raising to an integer (or rational) degree.

As indicated in (h) above, the user can specify correlation relationships between the uncertainty variables. In one set of embodiments, the decision flow builder interface may be configured to provide the user with the ability

selecting a pair of uncertainty variables from a set of V _UN uncertainty variables; and

entering a correlation coefficient to specify a correlation between a pair of uncertainty variables.

The user can specify any number of such correlations by reusing the interface of the decision flow builder described above. For example, the user can specify correlation relationships between pairs of uncertainty variables, which include the following forms:

a pair of uncertainty variables within an object (that is, referring to the same object);

a pair of variables of uncertainty over objects (that is, when one variable from a pair refers to one object, and another variable from a pair refers to another object);

a pair of uncertainty variables, where one or both variables are global uncertainty variables.

In some embodiments, a user may select a group of two or more uncertainty variables and enter a correlation matrix for the group.

When the flow is executed, the uncertainty variables are specified in a way that takes into account user-defined correlation relationships.

As indicated in (i) above, the user can set constraints on decision variables or functional combinations of decision variables. In one set of embodiments, the decision flow builder interface may be configured to provide the user with the ability

selecting a first decision variable Y from a set of V _D decision variables;

selecting a set of one or more variables X ₁ , X ₂ , ..., X _{N of a} solution (other than Y) from a set of V _D solution variables, where N is a positive integer;

the task, if desired, of the function f (for example, an arbitrary algebraic function) of one or more variables X ₁ , X ₂ , ..., X _{N of a} solution; and

selecting an inequality operator, denoted as “inequality”, from the set {<, ≤,> or ≥} to define a constraint of the form

Y “inequality” f (X ₁ , X ₂ , ..., X _N ).

In addition, the interface of the decision flow builder can be configured to provide the user with the possibility of setting constraints of the form Y “inequality” C, where C is the user-defined constant.

The set of V _D decision variables may include global decision variables and decision variables defined in (e). Any of a wide variety of f functions (including linear and non-linear functions) can be supported. According to some implementations, the decision flow builder interface allows the user to construct an arbitrary algebraic function of the selected decision variables X ₁ , X ₂ , ..., X _N.

The decision flow builder interface may be configured to allow a user to set any number of constraints on decision variables by reusing the decision flow builder interface described above. For example, the user can set restrictions, which include the following forms:

restriction on decision variables within an object, that is, when all decision variables covered by the restriction relate to the same object;

restriction on decision variables by objects, that is, when decision variables covered by the restriction relate to two or more objects;

restriction on decision variables, including one or more global decision variables.

When a stream is executed, optimization is performed, during which the user-defined restrictions are taken into account.

As indicated in (j1) above, the user can specify one or more values of interest for each object A (k), k = 1, 2, ..., N _A.

In one set of embodiments, a user can specify a value for an object A (k) by

selecting one or more output arguments from the algorithms related to the object A (k), and / or selecting one or more values that were previously set for the object A (k); and

defining a function (for example, an arbitrary algebraic function) of the selected output arguments and / or previously specified values.

By using this procedure, any number of values can be set for each object.

In addition, the decision flow builder interface can display a list of commonly used values consistent with the types of algorithms that relate to the current object A (k) under consideration. The user can specify values of interest by selecting values from the list.

Examples of object-related quantities of interest include such values as net present value, total carrying reserves, total oil produced, etc.

As indicated in (j2) above, the user can set the statistical indicators to be calculated for each object A (k), k = 1, 2, ..., N _A , that is, statistical indicators based on one or more of the values that were set for the object A (k). According to one set of implementations, the decision flow builder interface may provide the user with the ability to set a statistical metric for each of one or more values that have been set for each object by

entering a mathematical expression defining a statistical indicator;

selecting a statistical indicator from the list of commonly used statistical indicators.

Examples of commonly used statistics include

average value;

standard deviation;

quantile of order p, where p is a user-defined value between zero and one.

In addition, the decision flow builder interface may provide the user with the ability to specify more than one statistic based on the same value. For example, a user may want to calculate the mean and standard deviation from the net present value of an object.

As part of the execution of the flow, one or more values specified for each object can be evaluated several times to obtain a set of values for each value. Each statistic specified for a quantity is calculated from the corresponding population.

As indicated in (k1) above, the user can set the global objective function S as a functional combination of the first subset of statistical indicators. To this end, in one set of embodiments, the decision flow builder interface may be configured to provide the user with the ability

finding the first subset by selecting one or more statistical indicators that were set for each object; and

assigning a function (for example, a linear combination or an arbitrary algebraic function) of statistical indicators from the first subset.

In accordance with one typical scenario, the user can select one statistical indicator S _k corresponding to each object A (k), k = 1, 2, ..., N _A , and set the global objective function S as a linear combination of the form

.

.

The user can enter the values of the coefficients c _k so that they are multiplied by the corresponding statistical indicators S _k . For example, the user can give the global objective function S such a shape that it is a linear combination of the average value of the net present value for each object A (k).

As indicated in (k2) above, the user can specify one or more auxiliary functions, with each auxiliary function being a functional combination of the corresponding subset of statistical indicators. According to one set of implementations for defining an auxiliary function, the interface of the decision flow builder may provide the user with the opportunity

selecting one or more statistical indicators that were set for each object; and

specifying a function (for example, an arbitrary algebraic function) of the statistics of the selected statistical indicators.

Using this procedure, the user defines any number of auxiliary functions.

In accordance with one typical scenario, the user can find the auxiliary function H by selecting one statistical indicator T _k corresponding to each object A (k), k = 1, 2, ..., N _A , and setting a linear combination of the form

or kind

.

.

, чтобы они были умножены на соответствующие статистические показатели. Например, пользователь может придать вспомогательной функции Н такую форму, чтобы она представляла собой линейную комбинацию квадрата среднеквадратического отклонения от чистой приведенной стоимости для каждого объекта A(k).User can enter coefficient values

so that they are multiplied by the relevant statistics. For example, the user can give the auxiliary function H such that it is a linear combination of the square of the standard deviation from the net present value for each object A (k).

As indicated in (k3) above, the user can specify one or more restrictions, including one or more auxiliary functions and / or the global objective function S. According to one series of implementations, the user can set one or more restrictions, each of which has the form

J “relationship” C,

where J is a user-defined function whose arguments are selected from the set F _C , which includes one or more auxiliary functions and the global objective function S, while the “relationship” is a dependence selected from the set {<, ≤,>, ≥, =}, where C is a user-defined constant. The user can set such a restriction by

selecting one or more arguments for function J from the set F _C ;

defining a functional combination (for example, an arbitrary algebraic functional combination) from one or more selected arguments;

selecting the relationship "relationship" from the set {<, ≤,>, ≥, =}; and

job constant C.

By using this procedure, any number of restrictions can be specified.

In accordance with one typical scenario, a user can rely on this procedure to specify a constraint of the form J = H / S <C, where H is an auxiliary function and S is a global objective function. For example, H may be a linear combination of squares of standard deviations from the net present value at one deviation per object, and S may be a linear combination of average values of net present values at one average value per object.

As indicated in (l) above, the user can select an optimizer. According to one set of implementations, a decision flow builder may provide a user with a choice of a set of supported optimizers. It is assumed that the global objective function S is purely non-linear and has many local minima (or maxima) throughout the global space defined by the decision variables. In addition, this global space over which optimization should be carried out can have isolated islands of admissible vectors among a huge number of invalid vectors. Therefore, it is desirable to use optimizers that are not limited to finding a single local minimum (or maximum) and do not stop inside a single island. Therefore, the set of supported optimizers may include stochastic optimizers, for example, optimizers that use randomness in the search technique.

According to some implementations, the decision flow builder may provide the user with the opportunity to choose from an optimizer such an optimizer as

(1) scatter search optimizer;

(2) search optimizer with a ban;

(3) an optimizer based on a metaheuristic combination of scatter search, forbidden search, neural networks, and linear programming;

(4) an optimizer based on a genetic algorithm.

A set of V _D decision variables from the stream presents solutions that are manageable. Each decision variable has a user-defined set of reachable values. The Cartesian product of the sets of attainable values of the solution variables in V _D is the global solution space during which optimization should be carried out and at the same time obey any restrictions that the user can set on solution variables or functional combinations of solution variables, and obey one or more of the constraints defined in (k3), including one or more auxiliary functions and / or global objective function.

A vector in the global decision space is called a solution vector. In other words, the solution vector is a vector of values, that is, one value selected from each set of attainable values of the solution variables in V _D.

In some embodiments, the decision flow builder may provide the user with the ability to introduce a set of initial estimation vectors that the optimizer may use in the search process.

A set of V _UN variables of uncertainty from a stream reflects uncertainty. Each uncertainty variable has a user-defined set of attainable values. The Cartesian product of the sets of attainable values of the uncertainty variables in V _UN is a global space of uncertainties that should be iteratively explored. A vector in the global space of uncertainties is called an uncertainty vector. In other words, the uncertainty vector is a vector of values, that is, one value selected from each set of attainable values of the uncertainty variables in V _UN .

In some embodiments, the decision flow builder may generate flow 200 in response to information (a) - (n) specified by the user. As shown in FIG. 2, stream 200 may include an optimizer 210 selected by the user in (l) and an evaluation process 220. Optimizer 210 can generate decision vectors x in the global decision space, subordinate decision variables to user-defined constraints. The evaluation process 220 is carried out with respect to the decision vectors x for generating the corresponding values S (x) of the global objective function S and the corresponding values H (x) of the auxiliary function H described below. The S (x) values of the global objective function and the H (x) values of the auxiliary function are sent to the optimizer. The optimizer uses these values to form adjusted solution vectors when trying to minimize (or maximize) the global objective function using a user-defined constraint on the auxiliary function H (x) (or on a combination of the auxiliary function and the global objective function). The optimizer can work in accordance with any of a number of optimization algorithms. The structure of the cycle, including the optimizer and the evaluation process, is referred to in this application as the "external" cycle.

In one set of embodiments, the optimizer 210 may be configured to perform operations in accordance with the following methodology shown in FIG. 3. The optimizer may be run on a server computer.

In step 310, the optimizer 210 can create a reference set of valid solution vectors (a solution vector is considered valid if it satisfies the user-defined constraints on the solution variables). At the first iteration in accordance with operation 310, a reference set can be formed by forming an initial set of solution vectors that are spatially different (i.e. obtained by using random formation methods), and then transforming the initial set of solution vectors into valid solution vectors. However, in any iteration after operation 310, a reference set can be formed by

selecting the “best” solution vectors in the amount of n ₁ from the previous version of the reference set (that is, the “best” in accordance with the assessment using the global objective function);

the formation of a set of spatially different vectors (for example, obtained by using random formation methods);

transforming various vectors to obtain various valid vectors;

adding a variety of valid vectors in an amount of n ₂ to vectors in an amount of n ₁ to increase the spatial diversity of the reference set.

The optimizer can sort the n ₁ best solution vectors in order of S (x) of their global objective function. The optimizer can also sort n ₂ additional vectors according to the distance measure.

In step 320, the optimizer 210 may form a new vector x by forming the _new combinations (e.g., linear combination) subsets of the reference set of vectors. New vectors do not have to be valid vectors. The subsets may include two or more vectors of the reference set.

In step 330, the optimizer 210 can convert the new solution vectors to feasible solution vectors x _f using a linear program and / or a partially integer program.

In step 340, the optimizer 210 may present valid decision vectors x _f for the evaluation process 220 to calculate the global objective function values S (x _f ) and the auxiliary function values H (x _f ) corresponding to the allowed solution vectors x _f .

In step 350, the optimizer can adjust the reference set by using the global objective function values S (x _f ) and the auxiliary function values H (x _f ) calculated in step 340. According to one embodiment, the optimizer can examine each of the valid vectors x _f so that to establish

(1) the condition that the feasible vector has a better value H (x _f ) of the global objective function than the current worst n ₁ solution vectors in the value function of the global objective function, and in this case, the feasible vector x _f can replace the worst vector ;

(2) the condition that the feasible vector has a greater measure of distance than the current worst n ₂ additional vectors.

In either case, the optimizer can replace the worst vector with a valid vector. If none of the valid vectors x _f satisfies condition (1) or (2), the reference set remains the same, that is, the adjustment is a zero correction.

In step 360, the optimizer can determine if the reference set has changed as a result of step 350. If this happens, the optimizer can perform another iteration according to steps 320-360. If the reference set has not changed, the optimizer can proceed to operation 365.

In step 365, the optimizer may determine whether the termination condition is satisfied. If the termination condition is not satisfied, the optimizer can proceed with operation 310 to regenerate the reference set with another set of n ₂ different valid vectors. If the termination condition is satisfied, the optimizer may continue with operation 370.

According to various implementations, different termination conditions are contemplated. For example, an end condition may be a condition under which

(1) at least one solution vector from the reference set has a better achievable value of the global objective function than the user-defined threshold, while one or more constraints regarding one or more auxiliary functions and / or the global objective function are satisfied; or

(2) a given number of n _max iterations of the outer loop (resulting from operations 310 to 365) is performed; or

(3) a predetermined time T _{max of} calculations elapses; or

(4) any combination of (1), (2) and (3) is satisfied.

The maximum number n _max iterations and the maximum time T _max calculations can be set by the user.

In step 370, the optimizer can store the reference set vectors in the database along with their corresponding global objective function values and auxiliary function values. The output analyzer can read from the database and display a graphical representation of the value of the global objective function and the value (s) of the auxiliary function for each of the vectors of the reference set (or, alternatively, for n ₁ best vectors of the reference set).

According to one set of implementations, the evaluation process 220 may be arranged to perform actions in accordance with the following methodology shown in FIG. 4. The evaluation process may be performed on a server computer.

In operation 410, a decision vector x may be received from the optimizer for the estimation process.

In step 420 of the module, the specification of the estimation process can create N _I uncertainty vectors in a way that takes into account user-defined functional dependencies and correlation between uncertainty variables, while N _I is a user-defined integer positive number. Using N _I uncertainty vectors, the global space of uncertainties is investigated. Uncertainty vectors can be formed in accordance with the user-specified instantiation method described above. Each uncertainty vector U _j , j = 1, 2, ..., N _I , has a specific value for each of the uncertainty variables in the set V _UN (that is, in the set of all uncertainty variables related to the current task). Each specified value is selected from the set of attainable values corresponding to one of the uncertainty variables.

In operation 425 of the estimation process, N _I sets of iterative data can be created based on models in anticipation of N _I iterations of the “inner loop” for each object A (k), k = 1, 2, ..., N _A , where N _I is the number of iterations and N _A is the number of objects. For example, during the evaluation process, iterative data sets may be created in accordance with the following pseudo-code:

for k from 1 to N _A :

for j from 1 to N _I :

a set of IDS (k, j) iterative data is created for the algorithms of object A (k) using the subvector x (k) of the solution vector x that corresponds to the object A (k) and the subvector U _j (k) of the uncertainty vector U _j object A (k);

a set of IDS (k, j) iteration data is stored in the IBUFF (k) iteration buffer for object A (k);

End for

End for.

There is no need to execute the For Loop command (interrupting the loop after executing a fixed number of times) with respect to the exponent k in a strictly sequential manner. For example, this For Loop command can be configured to perform multi-threaded operations, for example, using one thread for each k value.

It should be recalled that the decision-making management system provides the user with the opportunity to build models by

(1) subordinating sets of data structures to decision variables;

(2) subordinating sets of data structures to uncertainty variables;

(3) moving data values from data structures to decision variables; and

(4) moving data values from data structures to uncertainty variables.

The values of the subvector x (k) of the solution and the values of the subvector U _j (k) of uncertainty can be used to determine one or more data structures based on models. In particular, one or more values of the subvector x (k) of the solution and / or one or more values of the subvector U _j (k) of uncertainty can be used to select data structures based on models, including sets of alternative data structures. In addition, one or more values of the subvector x (k) of the solution and / or one or more values of the subvector U _j (k) of uncertainty can be substituted into models that include solution variables and / or uncertainty variables. Data structures defined by the values of the subvector x (k) of the solution and the values of the subvector U _j (k) of the uncertainty can be included in the set of IDS (k, j) iterative data. A set of iterative data may include enough data to execute a set of algorithms that are defined for object A (k). The subvectors x (k) do not have to be disjoint. For example, values corresponding to global decision variables can be found in more than one of the subvectors x (k). Similarly, the subvectors U _j (k) need not be disjoint.

In step 430, the evaluation process manager can distribute (i.e. send) N _I sets of iterative data for each object A (k), k = 1, 2, ..., N _A , to one or more other computers, where N _I is the number iterations to be performed on each object. For example, a dispatcher may operate in accordance with the following pseudo-code:

for k from 1 to N _A :

While (while the condition is fulfilled) that the IBUFF (k) iteration buffer is not empty:

detecting a free computer C _P ;

reading one or more sets of IDS (k, j) iteration data from the IBUFF (k) iteration buffer;

compressing one or more sets of iterative data and packing the compressed data into an executable data file;

transferring the executable file to the computer C _P to create each of one or more sets of iterative data on the computer C _P ;

EndWhile (end on condition)

Endfor.

The dispatcher may include a copy of the executable code for the algorithms related to the object A (k) in the executable data file when the computer C _P does not yet have a copy of this code. In addition, the dispatcher may include addressing the buffer INB (k) of data related to object A (k) in an executable data file. The data buffer INB (k) may be located in the storage device of the server computer. Computer C _P executes a set of algorithms related to object A (k), on the basis of each set of IDS (k, j) iterative data included in the executable data file, collects output data (algorithms) resulting from execution into the corresponding ODS set (k, j) output and transmits a set of ODS (k, j) output to the data buffer INB (k).

There is no need to execute the For Loop command with respect to exponent k in a strictly sequential manner. For example, this For Loop command can be configured to perform multi-threaded operations, for example, using one thread for each k value.

In addition, there is no need to perform operations 425 and 430 in a strictly sequential manner. For example, multiple For Loop task chains can be multi-threaded in step 430 along with multiple For Loop task chains (relative to metric k) in step 425. According to one embodiment, the k-th dispatch thread can be started immediately after the beginning of the formation of the k-th stream of a set of iterative data. For example, when the k-th stream of the iterative data set is generated, the k-th dispatcher stream can start after saving one or more sets of iterative data in the IBUFF (k) iteration buffer.

The server computer may include one or more processors.

In step 440, the accumulating adder of the results of the evaluation process may store the ODS (k, j) sets of output data generated by other computers (onto which the data sets were distributed for execution). The set ODS (k, j) of the output data is the output of the algorithms related to the object A (k), executed on the basis of the set of IDS (k, j) iterative data. According to one set of implementations, the accumulating adder of results can perform operations in accordance with the following pseudo-code:

for k from 1 to N _A :

j ← 0;

While (while the condition is fulfilled) j≤N _I

receiving a set of ODS (k, j) output data from the input data buffer INB (k);

unpacking the content data from the set of ODS (k, j) output data;

the unpacked data from the set of ODS (k, j) output data is stored in the database (that is, in the database equipped with the indices k and j);

j ← j + 1;

EndWhile (end on condition)

EndFor (command).

There is no need to execute the For Loop command with respect to exponent k in a strictly sequential manner. For example, this For Loop command can be configured to perform multi-threaded operations, for example, using one thread for each k value.

In addition, there is no need to perform operations 430 and 440 in a strictly sequential manner. For example, multiple For Loop task chains can be multi-threaded in step 440 along with multiple For Loop task chains in step 430. According to one embodiment, the kth accumulative adder stream can be started immediately after k- th dispatcher thread. For example, the k-th stream of the dispatcher can start the k-th stream of the accumulating adder after receiving the distributed first executable data file for the object A (k).

In step 445, the evaluator of the results of the evaluation process for each object A (k), k = 1, 2, ..., N _A, and for each iteration j = 1, 2, ..., N _I , the value of each user-defined value for the object with using the values of the output argument from the set of ODS (k, j) output. Therefore, the results calculator forms a set of N _I values for each value set by the user.

In step 445B, the result calculator for each object A (k), k = 1, 2, ..., N _A, can calculate the value of each user-specified statistical indicator for the object based on the corresponding one of the populations.

Operations 445A and 445B are collectively referred to as operations 445.

According to one set of implementations, the result computer can be configured to perform operation 445 in accordance with the following pseudo-code:

for k from 1 to N _A :

j ← 0;

While (while the condition j≤N _I is satisfied),

for each user-defined value related to the object A (k):

reading the value (s) from the set of ODS (k, j) output data to calculate the value of a user-defined value;

calculating and storing the value of the value;

Endfor

j ← j + 1;

EndWhile (end on condition)

for each user-defined statistic related to the object A (k):

calculating the value of a user-defined statistical indicator for N _I values of the corresponding user-specified value.

Endfor

Endfor.

The For Loop command relative to index k need not be executed in a thread-wise manner. For example, this For Loop command can be configured to perform multi-threaded operations, for example, using one thread for each k value.

In addition, there is no need to perform operations 440 and 445 in a strictly sequential manner. For example, multiple For Loop task chains can be multi-threaded in step 445 along with multiple For Loop task chains in step 440. According to one embodiment, the kth result calculation stream can be started immediately after k- th stream of summation of results. For example, the k-th stream of summing the results can start the k-th stream of calculating the results after placing the first of the ODS (k, j) sets of output data in the database.

In step 450, the global result calculator can calculate

the value S (x) of the global objective function S based on a user-defined first subset of a set of G _T statistical indicators (i.e., a combination of statistical indicators from each object);

the value of each auxiliary function based on the corresponding user-defined subset of the full set of G _T statistical indicators.

According to one set of implementations, a method for optimizing decisions regarding a large number of objects may include the following operations shown in FIG. 5.

At step 510, a computer (for example, a server computer including one or more processors) may receive information specifying:

decision variables and uncertainty variables for a large number of objects; and

for each object, a corresponding set of one or more algorithms.

At step 515, the computer may form a solution vector. The solution vector includes a value for each of the decision variables. Each of the decision variables has a set of attainable values related to it.

At step 520, the computer can perform an estimation process with respect to the solution vector to determine at least the value of the global objective function for the solution vector. See below for a description of the assessment process.

At operation 530, the computer may perform an optimizer. The implementation of the optimizer includes the repeated implementation of operations 515 and 520, resulting in the formation of a large number of solution vectors and the corresponding values of the global objective function. The optimizer uses a large number of solution vectors and corresponding values of the global objective function to form and repeatedly adjust the reference set of solution vectors in order to optimize the global objective function.

At step 540, the computer can store data in the storage device (for example, in the system memory of the computer) including a large number of decision vectors and the corresponding global objective function values.

At step 545, the computer can display a graphical representation of at least a subset of the solution vectors from the reference set.

The optimizer can be implemented as any of a number of optimizers, in particular as a stochastic optimizer. For example, the optimizer may be

(1) scatter search optimizer;

(2) search optimizer with a ban;

(3) an optimizer based on a metaheuristic combination of scatter search, forbidden search, neural networks, and linear programming; or

(4) an optimizer based on a genetic algorithm.

Operation 520 for performing the evaluation process may include the following operations shown in FIG. 6.

At operation 520A, the computer may generate an uncertainty vector. The uncertainty vector includes a value for each of the uncertainty variables.

At step 520B, the computer can generate a data set for each object for the corresponding set of algorithms using the appropriate subset of one or more values in the solution vector and the corresponding subset of one or more values in the uncertainty vector.

At operation 520C, the computer can initiate a set of algorithms for each object based on the corresponding data set to obtain output arguments corresponding to the object. A computer can initiate execution on other computers, for example, by sending data sets to other computers (and, if necessary, executable code for the corresponding algorithms).

At operation 520D, the computer can calculate for each object one or more values of one or more corresponding values from the output arguments corresponding to the object.

At step 520E, the computer can perform operations including operations 520A through 520D several times, resulting in a set of values of each value for each object.

At step 520F, the computer can calculate one or more statistics for each object based on one of the populations.

In step 520G, the computer can combine the first subset of statistics to determine the value of the global objective function corresponding to the solution vector. The combination of the first subset of statistical indicators can be carried out in accordance with a user-defined functional combination (for example, in accordance with a linear combination).

Information (relating to 510 above) may specify one or more functional relationships between the uncertainty variables. An operation of generating an uncertainty vector 520A may be performed in a manner that takes into account one or more functional relationships between the uncertainty variables. The user can specify functional dependencies, for example, through the client graphical user interface, which is implemented on another computer (for example, on the client computer).

Information can also specify one or more correlations between uncertainty variables. In this case, the operation 520A of the formation of the uncertainty vector can be performed by a method in which these correlation relationships are taken into account. Uncertainty variables can comprise two or more subsets related to two or more objects, respectively. Correlation relationships may include correlation relationships between uncertainty variables for various objects.

Information can also specify one or more constraints on decision variables. In this case, the operation 515 of the formation of the solution vector can be performed taking into account one or more restrictions on the decision variables.

Operation 520 of performing the estimation process with respect to the solution vector can provide an appropriate value of the auxiliary function in addition to the specified value of the global objective function. Therefore, operation 520 may further include an operation (520H) combining a second subset of statistical indicators to determine the value of the auxiliary function corresponding to the solution vector.

The optimizer can use the values of the global objective function and the values of the auxiliary function corresponding to a large number of solution vectors, with the specified multiple adjustments of the reference set, in order to optimize the objective function subject to one or more constraints.

The user can specify the structure of one or more constraints. For example, the user can set a restriction on the functional combination of the auxiliary function and the global objective function (the functional combination itself can be set by the user). As another example, a user can set a restriction only on an auxiliary function. Operation 520H may include combining a second subset of statistical indicators in accordance with a linear combination, the coefficients of which are set using the specified information.

Each of the decision variables has a set of attainable values related to it, representing the possible results of the corresponding solution. In addition, each of the uncertainty variables has a set of attainable values related to it. The sets of reachable values can be finite sets of numerical values, data structures, programs, etc. The sets of reachable values can also be numerical line intervals or, more generally, user-defined areas in n-dimensional space, where n is a positive integer.

According to one embodiment, a graph of a global objective function value versus an auxiliary function value for each of the indicated large number of decision vectors can be displayed using a display device (such as a monitor, projector, head display, etc.).

Data storage operation 540 may include storing a large number of decision vectors together with their corresponding global objective function values and corresponding auxiliary function values in a database in a storage device.

In some embodiments, a computer may

perform operations on the database to identify a subset of a large number of solution vectors that have optimal values of the global objective function for a certain value of the auxiliary function;

repeat the indicated operations for a large number of auxiliary function values; and

save the subsets identified at the specified repetition.

In addition, the computer can also

perform operations on the database to determine the geometric location of the optimal values of the global objective function, taking into account the value of the auxiliary function; and

display the geometric location of the optimal values.

A graphical representation of a subset of a large number of solution vectors that corresponds to a point at the geometrical location of the optimal values can be displayed in response to a user choosing a point.

The user can set a large number of objects, decision variables and uncertainty variables for each object and a set of one or more algorithms for each object. Decision variables are variables that undergo optimization.

Uncertainty variables can be examined randomly (or in a pseudo-random way) to create variations in the input to the algorithms for each object.

A set of one or more algorithms for at least one of the objects may include an algorithm for estimating oil and gas production over time and economic indicators for the object. An algorithm for estimating oil and gas production over time can be

a material balance simulator that operates with the data structure of the reservoir model;

calculation of material balance performed in spreadsheet models;

a fluid motion simulator in a reservoir that operates with a three-dimensional data structure;

an algorithm defined by a set of one or more user-defined algebraic expressions.

In addition, a set of one or more algorithms for at least one of the objects may include one or more of

algorithms for evaluating the transportation of oil and gas to the surface, for example, a process simulator in a network of onshore pipelines,

algorithms for evaluating the transportation of oil, gas and water to the surface, for example, an analytical model of a network of land pipelines or a set of one or more user-defined algebraic expressions, and

algorithms for estimating the volume of oil and gas and the three-dimensional properties of the reservoir.

An algorithm for estimating the volume of oil and gas and the three-dimensional properties of the reservoir can be implemented as

a geostatistical algorithm for the formation of the properties of the geological environment, determined using a set of one or more user-defined algebraic expressions; or

a simulator of material balance.

The algorithm for calculating the economic indicators of the object can be implemented as

spreadsheet-based algorithms, which may include full fiscal models and production sharing models;

algorithms based on analytical economic models and a set of one or more user-defined algebraic expressions.

According to another set of implementations, a method for optimizing decisions relating to a large number of objects, carried out by executing program code on a first computer (eg, a server computer), may include the following operations shown in FIG. 7.

At step 710, the first computer (for example, a server computer including one or more processors) can receive information defining decision variables and uncertainty variables for a large number of objects and defining for each object a corresponding set of one or more algorithms.

At operation 720, the first computer can create a solution vector, the solution vector including a value for each of the decision variables.

At step 730, the first computer can execute the process relative to the solution vector to determine at least the values of the global objective function for the solution vector.

At step 740, the first computer can execute the optimizer, while the specified optimizer execution includes the repeated execution of actions 720 and 730, resulting in a large number of solution vectors and corresponding values of the global objective function, and the optimizer uses a large number of solution vectors and corresponding values global objective function for education and multiple adjustments of the reference set of decision vectors in order to optimize the global objective function.

At step 750, the first computer may store in the storage device data including a large number of solution vectors and their corresponding global objective function values.

Operation 730 for performing the evaluation process may include the following operations shown in FIG. 8.

At step 730A, the first computer can generate N _I uncertainty vectors, with each uncertainty vector containing specific values for each of the uncertainty variables, where N _I is a positive integer.

At step 730B, the first computer can generate N _I sets of iterative data for each object, with each of the sets of iterative data being generated using part of the solution vector and part of the corresponding one of the uncertainty vectors.

At step 730C, the first computer can distribute N _I sets of iterative data for each object to one or more secondary computers to initiate the execution of a set of algorithms for the object based on each of the corresponding N _I sets of iterative data, where at each execution of the set of algorithms for the object based on a set of iterative data, an appropriate set of output data is generated.

At step 730D, the first computer can receive N _I sets of output data for each of the objects from one or more secondary computers and store N _I sets of output data in memory.

At step 730E, the first computer can calculate, for each object and each of the N _I sets of output data corresponding to the object, one or more values for one or more corresponding values from the data in the set of output data, resulting in a collection of N _I values of each value for each object.

At step 730F, the first computer can calculate one or more statistics for each object based on the corresponding one of the populations.

In step 730G, the first computer can combine the first subset of statistics to determine the value of the global objective function corresponding to the solution vector.

The first computer and one or more secondary computers can be connected using a computer network, such as a local area network, intranet, wide area network, Internet, etc.

During the operation of distributing sets of iterative data, distributed execution of sets of algorithms based on sets of iterative data is initiated on a large number of secondary computers. Secondary computers can be combined into a mega-computer network structure.

According to one embodiment, depending on the available processing system performance on the first computer, a subset of the iterative data sets may be executed locally by the first computer.

Information can be obtained from the user during the process on the client computer and transferred to the first computer in accordance with this process.

According to yet another set of implementations, a method for optimizing decisions regarding a large number of objects may include the following operations shown in FIG. 9.

At step 910, a computer (for example, a server computer including one or more processors) can receive information specifying a large number of objects, an objective objective function for each of the objects, and a set of algorithms for each object.

At step 920, the computer can create a solution vector including values for a set of decision variables.

At step 930, the computer can create N _I uncertainty vectors, each of the N _I uncertainty vectors being the result of a separate specification of a set of uncertainty variables, where N _I is a positive integer.

At step 940, the computer can calculate the first set of N _I values for each objective function of the object by performing N _I iterations of the execution of the corresponding set of algorithms and operating with the output data generated at each of the N _I iterations of the specified execution, and each time the corresponding set of algorithms is executed, actions on a set of input data defined using the values of decision variables and the corresponding one of the uncertainty vectors.

At step 950, the computer can calculate the first statistical indicator of the object for each objective function of the object based on the corresponding set of values.

At step 960, the computer can calculate the value of the global objective function based on the first statistic of the object.

At step 970, the computer can perform the optimizer, while the optimizer includes repeatedly performing operations 920 to 960, resulting in a large number of solution vectors and corresponding values of the global objective function, and the optimizer uses a large number of solution vectors and corresponding values of the global objective function for the formation and multiple adjustments of the reference set of solution vectors in order to optimize the global objective function.

In some embodiments, a computer may

calculate the second statistical indicator for each object for the corresponding one of the populations; and

calculate the value of the auxiliary function based on the second statistical indicator of the object.

When performing the optimizer, an attempt may be made to optimize (i.e., maximize or minimize) the global objective function, subject to a restriction imposed on one or more functions from auxiliary functions and the global objective function.

Figure 10 - structural diagram of a computer system

FIG. 10 is a block diagram showing a computer system 1082 according to one set of embodiments that can act as a server computer or a client computer, as described variously in the present application.

Computer system 1082 may include at least one central processing unit 1160 (i.e., a processor) that is connected to a host bus 1162. The central processing unit 1160 may be any of various types, including, but not limited to, an x86 processor, processor PowerPC, a central processor unit from the SPARC family (with scalable processor architecture) of RISC processors (with a simplified instruction set), as well as others. A storage medium, typically including a semiconductor random access memory and referred to in this application as main memory 1166, may be connected to the main bus 1162 via a memory controller 1164. In the main memory 1166 can be stored programs that are executed to implement any or all or any subgroup of various methods according to the implementations described in this application. The main memory can also store operating system software, as well as other software for operating a computer system.

The host bus 1162 may be coupled to an expansion or I / O bus 1170 via a bus controller 1168 or bus bridge logic. The expansion bus 1170 may include slots for various devices, such as a video card 1180, a hard disk 1182, information storage devices 1190 (such as a CD-ROM drive, tape drive, floppy disk drive, etc.) and a network interface 1122. The video card 1180 can be connected to a display device, such as a monitor, projection apparatus, or head-mounted display. A network interface 1122 (e.g., an Ethernet device) can be used to communicate with other computers over the network.

Computer system 1082 may also include input / output devices 1192, such as a mouse, keyboard, and speakers.

A computer system 1082 according to embodiments aimed at being used as a server computer can be enhanced by higher processor performance (for example, there may be several processors), more storage capacity, and more network access bandwidth than a computer system according to embodiments implementation aimed at using as a client computer. A client computer may include a mouse, keyboard, speakers, and a video card (or graphics accelerator), while there is no need for the server computer to include these items.

Any method (or part of it) according to the implementations described in this application can be implemented based on program instructions. Program instructions may be stored in a computer readable storage medium. Software instructions may be read and executed (by a computer or group of computers) to implement a method (or part of it) according to implementations. See the definition of storage medium given in the list of terms below.

Terminology

The following terms are used in this application.

Algorithm:

The algorithm is a computational method.

Argument:

The input and output of the algorithm are called arguments. The argument may be input to the algorithm, or output from the algorithm, or both input and output. Each argument can have a unique name (related to the algorithm for which it is defined) and also a specific type of object that it can carry (for example, a function that converts an integer to an equivalent floating point number, integer, string, etc. )

Argument Map:

An argument library belonging to an algorithm is called an argument map.

Cycle:

The loop is used to execute the algorithm several times (however, as noted above, not necessarily sequentially). A loop can be formed from an iterator of an algorithm, an algorithm, and an optional dispatcher (that is, a mechanism for distributing iterations of the algorithm). The thread builder gives the user the opportunity to select various algorithm iterators for use in the loop and to select various dispatchers for use in the loop. The loop itself is an algorithm. An argument map of a loop can be defined as a combination of an argument iterator argument map and an algorithm argument map.

Storage medium:

A storage medium is an environment configured to store information. Examples of storage media include various types of magnetic media (e.g., magnetic tape or magnetic disk); optical media of various kinds (for example, a read-only CD); semiconductor random access memory and read-only memory devices of various kinds; various media based on the conservation of electric charge or other physical quantities; etc.

Algorithm Iterator:

The algorithm iterator is used to obtain sets of iterative data intended for the corresponding algorithm within the loop. An algorithm iterator may have a pair of argument maps. The action of one map of arguments appears both inside and outside the loop, and it is used as input and output from an algorithm iterator. The second argument map is used to communicate with the corresponding algorithm within the loop.

Dispatcher:

The dispatcher is used to distribute iterations of the corresponding algorithm.

Script algorithm:

A script algorithm is a script (such as a Jython script or a Perl script) built into the algorithm. The input arguments of the algorithm can be brought into the script, and the output arguments of the script can be the output arguments of the algorithm.

Process algorithm:

The process algorithm is an external process that is embedded in the algorithm. Process algorithms may include process simulation algorithms and optimization algorithms. A process simulation algorithm simulates a physical system, usually based on one or more models specified as input data. One example of a process simulation algorithm is a fluid simulator through porous media called a reservoir simulator.

Flow:

A stream is a collection of interconnected algorithms. To associate the output argument of one algorithm with the input argument of the second algorithm, the thread builder may require matching the name and type of the argument. For example, the output INT (exact numeric data) can be transferred to the input FLOAT (approximate numeric data). However, passing STRING output (string) to INT input may be prohibited. A graphical user interface may be provided to enable manual communication of various algorithms. After the stream is created, if desired, it can be converted into an algorithm.

Although the embodiments have been described in great detail above, numerous changes and modifications will become apparent to those skilled in the art after fully understanding the above disclosure. It is intended that the following claims be interpreted to cover all such changes and modifications.

Claims (29)

1. A method for optimizing decisions regarding a plurality of objects, the objects being oil and / or gas fields, the method comprising the steps of
(a) receive information defining decision variables and uncertainty variables for a plurality of objects and defining for each object an appropriate set of one or more algorithms;
(b) form a solution vector, wherein the solution vector includes a value for each of the decision variables;
(c) performing an estimation process with respect to the solution vector to determine at least a global objective function value for the solution vector in which said execution of the evaluation process includes
(c1) the formation of an uncertainty vector, wherein the uncertainty vector includes a value for each of the uncertainty variables;
(c2) the formation for each object of a data set for the corresponding set of algorithms using the corresponding subset of one or more values in the solution vector and the corresponding subset of one or more values in the uncertainty vector;
(c3) initiating the execution of a set of algorithms for each object based on the corresponding data set to obtain output arguments corresponding to the object;
(c4) calculating, for each object, one or more values of one or more corresponding values from the output arguments corresponding to the object;
(c5) performing operations including (c1) to (c4) several times, resulting in a set of values of each quantity for each object;
(c6) calculating one or more statistical indicators for each object based on the corresponding one of the aggregates; and
(c7) combining the first subset of statistical indicators to determine the value of the global objective function corresponding to the solution vector;
(d) an optimizer is performed, wherein said optimizer implementation includes the repeated implementation of (b) and (c), as a result of which a plurality of solution vectors and corresponding values of the global objective function are generated, and the optimizer uses a plurality of solution vectors and corresponding values of the global objective function for the formation and multiple adjustments of the reference set of decision vectors in order to optimize the global objective function;
(e) storing data including a plurality of decision vectors and their corresponding global objective function values in a storage device;
(f) drilling a number of wells in a plurality of fields, wherein the number of wells is determined by selecting one of the solution vectors of the reference set;
wherein said drilling of a number of wells includes drilling a series of wells at locations determined by selecting one of the solution vectors of the reference set. 2. The method according to claim 1, wherein said initiation of execution includes distributing data sets for deposits to a plurality of remote computers, wherein said execution of a set of algorithms of corresponding data sets is performed on remote computers. 3. The method according to claim 1, wherein said information is received from a client computer via a computer network. 4. The method according to claim 1, in which additionally display a graphical representation of at least a subset of the solution vectors from the reference set. 5. The method according to claim 1, in which the specified information defines one or more functional dependencies between the uncertainty variables, and with the specified formation of the uncertainty vector, take into account the specified one or more functional dependencies between the uncertainty variables. 6. The method according to claim 1, in which the specified information also defines one or more correlation relationships between the uncertainty variables, and with the specified formation of the uncertainty vector take into account these correlation relationships. 7. The method according to claim 6, in which the uncertainty variables include two or more subsets related to two or more objects, respectively, while these correlation relationships include correlation relationships between the uncertainty variables for various objects. 8. The method according to claim 7, in which the specified information also sets one or more restrictions on the decision variables, and with the specified formation of the solution vector, the specified one or more restrictions on the decision variables are observed. 9. The method according to claim 1, in which the decision variables include one or more numeric decision variables and one or more scenario decision variables. 10. The method according to claim 1, in which the value of the global objective function is a statistical indicator of the set of values of the objective function of the object level, while the values of the objective function of the object level are calculated based on the set of executions (c1) to (c4) using various specific values variables of uncertainty. 11. The method according to claim 1, wherein said information also defines one or more functional dependencies between solution variables, and with said formation of a solution vector, said one or more functional dependencies between solution variables are taken into account. 12. The method according to claim 1, in which each of the decision variables has a plurality of attainable values related thereto, representing the possible results of the corresponding solution, each of the uncertainty variables has a plurality of attainable values relating thereto. 13. The method according to claim 1, in which, when the specified evaluation process is performed relative to the solution vector, the corresponding value of the auxiliary function is obtained in addition to the specified value of the global objective function, and the specified execution of the evaluation process relative to the solution vector further includes the step of:
(c8) combine the second subset of statistical indicators to determine the value of the auxiliary function corresponding to the solution vector;
the optimizer uses the values of the global objective function and the values of the auxiliary function corresponding to the set of solution vectors for the specified multiple adjustment of the reference set in order to optimize the objective function subject to one or more restrictions. 14. The method according to item 13, in which one or more of the restrictions include a restriction on the functional combination of the auxiliary function and the global objective function. 15. The method according to item 13, in which one or more restrictions include a restriction on an auxiliary function. 16. The method according to item 13, in which (c8) includes combining a second subset of statistical indicators in accordance with a linear combination, the coefficients of which are specified by the specified information. 17. The method according to item 13, in which additionally
display a graph of the dependence of the value of the global objective function on the value of the auxiliary function for each of the specified set of solution vectors. 18. The method according to item 13, in which the preservation includes the storage of solution vectors from the set of solution vectors together with the corresponding values of the global objective function and the corresponding values of the auxiliary functions in the database in the specified storage device. 19. The method according to p, in which additionally
perform an action on the database to identify a subset of the set of solution vectors that have optimal values of the global objective function for a given value of the auxiliary function;
repeat the specified action for the set of values of the auxiliary function; and
retain the subsets identified at the specified repetition. 20. The method according to p, in which additionally
perform an action on the database to determine the geometric location of the optimal values of the global objective function relative to the value of the auxiliary function; and
display the geometric location of the optimal values. 21. The method according to claim 20, in which additionally
display a graphical representation of a subset of the set of solution vectors that correspond to a point at the geometric location of the optimal values, in response to a user selecting a specified point. 22. The method according to claim 1, in which a set of one or more algorithms for each object is selected by the user. 23. The method according to claim 1, in which a set of one or more algorithms for at least one of the objects includes an algorithm for estimating oil and gas production over time. 24. The method according to claim 1, in which a set of one or more algorithms for at least one of the objects includes an algorithm for evaluating the transportation of oil and gas on the surface. 25. The method according to claim 1, in which the set of one or more algorithms for at least one of the objects includes a first algorithm for estimating the transportation of oil, gas and water to the surface, said first algorithm being an analytical model of a ground network pipelines; or a set of one or more user-defined algebraic expressions. 26. The method according to claim 1, in which the set of algorithms for at least one of the objects includes a first algorithm for estimating the volume of oil and gas and three-dimensional properties of the reservoir. 27. The method according to claim 1, in which the set of algorithms for at least one of the objects includes a first algorithm for evaluating the economic performance of the object. 28. A computer-readable storage medium that stores program instructions, wherein the program instructions are executable for implementing any one of the methods of claims 1 to 27. 29. A computer system for optimizing decisions regarding a variety of objects, the objects being oil and / or gas fields, the system contains:
a storage device that stores program instructions;
at least one processor configured to read and execute program instructions, wherein said program instructions are executable for implementing any one of the methods of claims 1 to 27.

Images