RU2750060C2 - Method and system for industrial process control using trained machine learning algorithm (mla) - Google Patents

Abstract

FIELD: industrial process control.SUBSTANCE: invention relates to means of controlling an industrial process based on a predictive model. A method for predicting the chemical composition of a substance used in the process of industrial processing is proposed, and the process of industrial processing is carried out at the processing plant. The method uses an electronic device designed to implement a physical model to create a predictable forecast of the total chemical composition on at least some intermediate output channels; a thermodynamic model to simulate processing within an industrial process, and a machine learning algorithm (MLA), that is designed to predict the future total chemical composition of a substance at any stage of the industrial process based on the current process parameter at any stage of the processing within an industrial process.EFFECT: expanding the arsenal of tools for processing parameters of an industrial process using machine learning.21 cl, 6 dwg

Description

FIELD OF TECHNOLOGY

[1] The present technology relates to systems and methods for creating a machine learning algorithm (MLA) based on a predictive model to control an industrial process. More specifically, the present technology relates to a method and system for using a trained MLA to predict the chemical composition of a substance used in an industrial processing process.

LEVEL OF TECHNOLOGY

[2] Natural gas derivatives are used in many areas. Examples of the use of natural gas derivatives: as a fuel for industrial heating and drying processes, as a fuel for the operation of public and industrial stations, as a domestic fuel for cooking, heat and water heating, as a fuel for vehicles, as a raw material for chemical synthesis, large-scale production of fuels using a gas-liquid conversion (GTL) process, etc.

[3] To obtain derivative natural gas products, it is processed, which is a complex industrial process designed to purify raw natural gas by separating impurities (and some non-methane hydrocarbons and liquids) to produce, for example, dry natural gas that meets transportation requirements. through the pipeline. Typical natural gas processing plants (also sometimes called gas processing plants) purify raw natural gas by recovering common pollutants such as water, carbon dioxide (CO ₂ ) and hydrogen sulfide (H ₂ S). Some of the substances that pollute natural gas and that are removed during natural gas processing can be used on their own or as feedstock for another process.

[4] One example of a natural gas processing process is gas fractionation. The gas fractionation process is carried out in a gas fractionation plant, which is also sometimes called a gas processing plant. The gas processing plant uses raw natural gas as an input stream and produces processed natural gas as a product of a predetermined composition and / or quality (eg dry natural gas of a pipeline quality).

[5] A typical gas processing plant contains many distillation columns ("working chambers"). In a broad sense, industrial processing is based on the physical premise that different boiling points correspond to different fractions of a gas. Thus, in a particular column from a series of columns, the gas is heated to a temperature predetermined for a given column, and the material fed to this column is separated as it heats up to a predetermined temperature (light fractions rise, heavy fractions descend). These separated fractions are individually withdrawn from this column each to its own "next stage" distillation column (or to the final product outlet).

[6] The process is repeated through a series of columns until the outlet from the column is the desired product (the desired product has a predetermined chemical composition), for example, dry natural gas of a quality that meets pipeline transportation requirements.

[7] A typical gas processing plant has a variety of sensors that detect various parameters. In a broad sense, parameters can be categorized into "process parameters" (i.e. related to the industrial processing itself, such as temperature and / or pressure) and " substance parameters "(which represent the parameters of a substance that goes through a given stage of the industrial processing process in a particular distillation column). In one embodiment, the parameters of the substance can represent the chemical composition of some or all of the gas fractions of the substance that is undergoing industrial processing.

[8] Various methods are known for determining some of these parameters. For example, sensors can be located within the distillation columns to determine the pressure and / or temperature associated with an industrial processing process that occurs in some or all of the distillation columns of a gas processing plant. The chemical composition can be determined, for example, using a chromatograph. The chemical composition process is usually done "offline" in the sense that laboratory tests may be necessary to determine the chemical composition of a substance undergoing industrial processing. Thus, we can say that the parameters of a substance are determined "retrospectively" for a certain point in time in the past.

[9] It is a well-known problem that some of the sensors or processes that determine process parameters or material parameters may be expensive, impractical and / or may not provide real-time data (which may be "too late" for specific monitoring tasks. process). A "virtual sensor" is a field of technology that predicts specific parameters for specific processes without the need for physical sensors. Typically, virtual gauges are based on statistical models, machine learning algorithm (MLA) models, etc.

[10] An example of a virtual sensor is presented in patent application US 7,142,975, issued to Wang et al. On November 28, 2006. Wang et al. Describe a virtual in-cylinder pressure sensor for use in an internal combustion engine. The transmitter has one or more independent modules, each of which is used to evaluate different cylinder pressure variables. Each module is trained using measurement data from a physical cylinder pressure sensor and a real engine. Once trained, the modules can be inserted into the ECU of the same type and used to evaluate and predict the values for which they were trained.

DISCOVERY TECHNOLOGY

[11] Embodiments of the present technology have been developed taking into account the developers' understanding of at least one technical problem associated with existing approaches to controlling the operating parameters of industrial processing processes, for example, gas fractionation. Certain embodiments of the present technology have been developed in response to the developers' understanding of at least one problem associated with "virtual sensors" in the art for use in determining specific process parameters. Embodiments of the present technology aim to provide a trained MLA that acts as a virtual sensor to monitor industrial processing. Some of the embodiments of the present technology aim to provide an MLA that acts as a virtual sensor to predict at least one material parameter, for example, the chemical composition of a substance that is being industrially processed. When creating the invention, the problem of expanding the arsenal of technical means for a specific purpose was solved, namely, technical means known as virtual sensors. The technical result consists in the implementation of this purpose.

[12] Without setting any limits to a specific theory, the developers of this technology have developed options for its implementation based on the following premises. The chemical composition of the gas fraction that enters a given distillation column can vary over time (i.e., from batch to batch). Because of this variability, there is a need to refine the settings for the parameters of the industrial processing process, for example, the temperature and / or pressure of the corresponding distillation column (so that the substance entering the distillation column is optimally processed).

[13] One of the difficulties in refining processing parameters is that it takes time to change. For example, an increase in the temperature of a control plate of a given distillation column by a couple of degrees (for example) cannot occur instantaneously, and it takes time to reach a new target temperature (between when the operator enters the appropriate command and when the temperature actually reaches the new target value).

[14] This, in turn, imposes on the operator the obligation to know in advance that there is a need to change the processing parameter in such a way that the change actually occurs before or simultaneously with when the processed gas composition requires a change in the processing parameter. This is necessary to carry out the most optimal (or as close to optimal) process of industrial processing.

[15] Additionally, rapid changes in the processing parameters of a given distillation column can make the distillation column environment unstable, which in turn can lead to unstable product production. Under certain conditions, such a rapid change in the processing parameters can lead to the shutdown of the distillation column, for example, to the known in the art "flooding of the distillation column" - i. a situation in which a liquid fraction is "stuck" due to a rapid increase in vapor flow. Thus, embodiments of the present technology aim to provide control to the gas processing plant that allows for gradual and timely adjustment of process parameters based on “future needs” that are determined based on the predicted chemical composition of the material to be processed through the industrial process. To this end, embodiments of the present technology are aimed at optimizing the parameters of the industrial processing process by predicting the gas composition at any given stage of the industrial processing process at any given future point in time, which allows for gradual and timely adjustment of the process parameters.

[16] In order to achieve the above results, embodiments of the present technology provide a computer executable method that performs or has access to: (i) a physical model; (ii) a thermodynamic model; and (ii) a machine learning algorithm (MLA). In accordance with non-limiting embodiments of the present technology, the MLA acts as a "virtual sensor" and is configured to predict the chemical composition of a substance at any given stage of the industrial processing process without using a physical sensor to determine the chemical composition of the substance within a given distillation column. Thus, MLA can be used to control the industrial processing process, for example, by creating the proposed process parameters for various distillation columns based on the predicted chemical composition of a substance at the appropriate stage of the industrial processing process.

[17] Naturally, embodiments of the present technology can be used in addition to or in conjunction with physical sensors within some distillation columns. For example, MLA can be used to "validate" and / or calibrate physical sensor readings. Therefore, embodiments of the present technology are not intended to completely eliminate the use of physical sensors (although this is possible).

[18] MLA is configured to generate such predictions based on current process parameters in a given distillation column, the process parameters being determined in real time (or near real time) using one or more process sensors located within some or all of the distillation columns.

[19] The process parameters, for example, can be the following: temperature and / or pressure in a given distillation column, and so on. In some non-limiting embodiments of the present technology, the MLA, acting as a "virtual sensor", can determine the chemical composition for sufficiently short time intervals, for example, every minute, and the like. In any case, the time interval over which the MLA can predict the overall chemical composition is comparatively shorter than, for example, the time intervals over which the chemical composition can be analyzed and determined on a chromatograph.

[20] To train the MLA, it needs to "feed" the training dataset. In accordance with non-limiting embodiments of the present technology, the training dataset is generated at least in part based on the results of the physical model and the thermodynamic model. More specifically, in accordance with at least some embodiments of the present technology, the result of the physical model is used, at least in part, to create a thermodynamic model. In turn, the result of the thermodynamic model is used, at least in part, to create a training dataset for MLA training.

[21] Known process parameters and substances

[22] An electronic device that acts as a control element for a gas processing plant receives an indication of one or more parameters of the industrial processing process and / or an indication of the partial composition of the substance (gas) that is processed in the distillation column of the gas processing plant.

[23] Broadly speaking, an electronic device receives process parameters from one or more process variable sensors located within one or more distillation columns. By way of example, process parameters (and associated process parameter sensors) may be, without limitation: temperature readings obtained from thermostats located in some or all of the distillation columns (or at some other stage of the industrial processing process); pressure readings from pressure gauges located in some or all of the distillation columns (or at some other stage in the industrial processing process), etc. In general, these process sensors are configured to provide real-time (or near real-time) readings to an electronic device using a wired or wireless connection (or, alternatively, the readings can be read from one or more sensors and manually entered by the operator of the electronic device. ).

[24] The electronic device also receives a partial chemical composition (ie, gas fractions) as it passes through the distillation columns. For example, an indication of the partial chemical composition can be obtained from a chromatograph that analyzes samples of gas fractions taken from one or more points of the material flow through a series of distillation columns. In some embodiments of the present technology, chromatographs are located on one or more outlet channels of one or more of a number of rectification columns.

[25] In some embodiments of the present technology, chromatographs do not analyze the entire chemical composition of the gas fractions, but only a partial chemical composition of the gas fractions. The specific components of the gas fractions, measured as part of the partial chemical composition of the substance, are determined in advance by the operator of the gas processing plant and / or the electronic device. It should be noted that in some embodiments of the present technology, the analysis of the partial composition is performed "offline" in relation to the time when the sample of the gas fraction was obtained for the partial chemical analysis.

[26] In a specific example, a subset of the chemical components of the gas fractions that significantly affect the quality and / or composition of the final product are selected as part of the partial chemical analysis performed on the chromatograph.

[27] In some embodiments of the present technology, the chromatograph measures the partial chemical composition of gas fraction samples that are obtained at predetermined time intervals. The predefined time slots can be every 5 minutes, every 60 minutes, and so on. In some embodiments of the present technology, the partial chemistry reading may be delayed (i.e., not in real time). This is especially the case when the partial chemistry is obtained at a remote test laboratory (i.e. the chromatograph is at a remote location).

[28] The electronic device also receives an indication of the lag time between the time when a specific portion of the incoming raw material enters the gas processing plant, and when the corresponding portion of the final product leaves the gas processing plant. In other words, the time lag is the time it takes for the raw material to "pass" through the gas processing plant to process and reach the final product stage.

[29] Physical model

[30] The physical model, when executed by an electronic device, makes it possible to create, based on at least an indication of the lag time and a partial chemical composition of the substance, a predicted indication of the total chemical composition of the substance in at least some intermediate input channels.

[31] More specifically, based on the lag time and the resulting partial chemistry index, the physics model "recovers" the complete chemistry of the substance as it "passes" through the gas processing plant. As a specific example (but not limitation), the resulting partial composition can be the partial chemical composition of the material as it leaves each (or some) of the distillation columns, and the recovered total chemical composition can be the chemical composition of the material as as it enters each (or some) respective distillation columns.

[32] Thermodynamic model

[33] The thermodynamic model, when executed by an electronic device, is configured to simulate a processing process performed by a gas processing plant. More specifically, the thermodynamic model allows some of the undetected parameters of the industrial processing process to be generated based on specific discovered or otherwise known parameters of the industrial processing process and / or the substance that is the subject of the industrial processing process.

[34] Broadly speaking, a thermodynamic model can have some of the following properties.

[35] The thermodynamic model provides a model that defines (or describes) all of the distillation columns in a plant and the chemical composition of the material as it moves through the distillation columns.

[36] The thermodynamic model takes into account the interdependencies between the processes in the various distillation columns of the plant in accordance with the "sequential" nature of the process in the plant (where the outlet of one distillation column is the input stream directly to the next distillation column).

[37] The thermodynamic model is generated from historical data (actual measured process parameters, actual measured partial chemical composition of a substance) as well as data generated by the physical model (ie simulated data). After creating and validating a thermodynamic model (as described below), the thermodynamic model allows you to create any missing information about the industrial processing process and / or the substance that is undergoing industrial processing. Additionally or alternatively, the thermodynamic model allows the identification of those process variable sensors that are out of order and give incorrect readings.

[38] Machine learning algorithm (virtual sensor)

[39] In some embodiments of the present technology, MLA (after training) can predict the chemical composition of a substance only at those stages where a partial chemical composition is obtained (using chromatographs or the like). In other embodiments of the present technology, the MLA (after training) can predict the chemical composition of a substance at any stage in the path that a substance takes in a gas processing plant.

[40] In its broadest sense, the MLA is configured to predict the chemical composition of a substance at any given stage in an industrial processing process for a given point in time in the future. In a particular embodiment of the technology, the horizon for predicting the future depends on the frequency of readings of the process parameters and / or the frequency of readings of the partial chemical composition that was used to train the MLA.

[41] Models of verification and adjustment

[42] Embodiments of the present technology use one or more 5 validation models and / or one or more correction models.

[43] In some embodiments of the present technology, an adjustment model for the thermodynamic model is implemented. In a particular embodiment, the adjustment model for the thermodynamic model is a linear adjustment model. The linear adjustment model can be based on predetermined values of various processing factors.

[44] In some embodiments of the present technology, the MLA validation metric is used. In some embodiments of the present technology, the mean absolute error (MAPE) metric is used as follows:

[45] where:

[46] c - analyte;

[47] n is the number of measurements;

[48] i is the number of this dimension;

[49] t is the point in time when the forecast is performed;

[50] h - forecast time interval;

[51] A (s, t + h) _i , actual measurement at time t + h;

[52] F (c, t) _{i is the} predicted value for measurement generated by the thermodynamic model for time t.

[53] Broadly, the MAPE metric refers to the mean prediction error expressed as a percentage. First, the MAPE metric calculates the variance between each predicted value and the corresponding actual reading; the MAPE metric further expresses the percent variance of the actual reading and finally the percentages are averaged to arrive at the final MAPE metric.

[54] Based on the MAPE metric, a quality assessment vector can be obtained. In accordance with non-limiting embodiments of the present technology, a given quality score vector is associated with MLA prediction errors for a given analyte. More specifically, this vector of quality assessment can be expressed as follows:

[55] In which:

[56] VA is the estimated MLA;

[57] ci - substance component;

[58] MAPE (ci) - forecast error for component ci.

[59] Thus, the first subject of this technology is a method for predicting the chemical composition of a substance used in the process of industrial processing, and the process of industrial processing is carried out in a processing plant that has a main input channel of raw materials and a main output channel of the final processed substance, which are connected to each other. the other by means of a number of working chambers, each working chamber having an intermediate inlet channel and an intermediate outlet channel, and the intermediate outlet channel of the series of chambers supplies the input stream for this intermediate inlet channel. The method is carried out by an electronic device communicatively connected to the processing plant. The method includes: receiving from a processing plant an indication of a partial chemical composition of a substance determined at least at some predetermined area of the processing plant; receiving from the processing plant an indication of the delay time related to the industrial processing process when the substance passes through the processing plant from the input channel of the raw material to the main output channel of the processed substance; performing a physical model to create, based on at least an indication of the lag time and a partial chemical composition of the substance, a predicted indication of the total chemical composition of the substance in at least some intermediate inlets; based on the predicted indication of the complete chemical composition of the substance, creating a thermodynamic model for modeling the industrial processing process, and the thermodynamic model is configured to predict at least: a process parameter at any given stage and at any given time of the industrial processing process; and the complete chemical composition of the substance at any stage and at any given point in time in the industrial processing process; using the result of a thermodynamic model as input for a training set for training a machine learning algorithm (MLA) to predict the future total chemical composition of a substance at any given stage of the industrial processing process based on the current process parameter at this stage of the industrial processing process.

[60] In some embodiments of the method, the method further includes: using a physical model to create, based on the lag time, a time subdivision of the industrial processing process for each of a number of working chambers, the time subdivision being used to generate a predictable indication of the overall chemical composition substances in at least some intermediate inlet channels.

[61] In some embodiments, the method further includes checking the thermodynamic model.

[62] In some embodiments of the method, the learning step is performed only in response to a positive test of the thermodynamic model.

[63] In some embodiments of the method, the method further includes iteratively calibrating at least one of the physical model and the thermodynamic model until a positive test result is obtained.

[64] In some embodiments of the method, checking the thermodynamic model includes applying the MAPE formula:

[65] where:

[66] c - analyte;

[67] n is the number of measurements;

[68] i is the number of this dimension;

[69] t is the point in time when the forecast is performed;

[70] h - forecast time interval;

[71] A (s, t + h) _i actual measurement at time t + h;

[72] F (c, t) _{i is the} predicted value for measurement generated by the thermodynamic model for time t.

[73] In some embodiments of the method, the process parameter is a plurality of parameters of an industrial processing process, and in which: creating a thermodynamic model includes creating a thermodynamic model using a subset of the process parameters, excluding the given process parameter; and in which checking the thermodynamic model includes: comparing the generated value for the given process variable with the actual value of the given process variable.

[74] In some embodiments of the method, this process parameter is either a pressure parameter or a temperature parameter.

[75] In some embodiments of the method, the physical model creates a predicted indication of the overall chemical composition of a substance based on: an indication of the lag time; partial chemical composition of the substance; at least one parameter of the industrial processing process.

[76] In some embodiments of the method, in which creating a thermodynamic model includes creating a thermodynamic model using a subset of process parameters, excluding the given process parameter; and in which the given process variable is used to validate the thermodynamic model.

[77] In some embodiments of the method, the current process variable is either a pressure parameter or a temperature parameter.

[78] In some embodiments of the method, the current process variable further includes an indication of the future partial chemical composition of the material at a given stage of the industrial process.

[79] In some embodiments of the method, the method further includes, in an operating phase that occurs during operation, using the MLA to predict future overall chemistry based on process parameters during operation.

[80] In some embodiments of the method, the method further includes obtaining an indication of a process parameter during operation and using the parameter to predict the overall chemical composition.

[81] In some embodiments of the method, the future overall chemistry is used to control at least one parameter of the industrial processing process.

[82] In some embodiments of the method, the future gross chemistry at the time of operation includes a plurality of future gross chemistries predicted for multiple time intervals, and in which a plurality of future gross chemistries are periodically predicted for time intervals that are shorter than time intervals for determining the partial chemical composition.

[83] In some embodiments of the method, at least one predetermined portion of the processing plant is one of the intermediate outlet channels.

[84] In some embodiments of the method, at least one of the intermediate output channels is all intermediate output channels.

[85] In some embodiments of the method, a predicted indication of the total chemical composition of at least some intermediate inlets is a predicted indication of the total chemical composition of all intermediate inlets.

[86] In some embodiments of the method, MLA training includes using the result of a thermodynamic model to create a training set for MLA, the training set includes: as an input, at least one process parameter associated with an industrial processing process on a given stage of the industrial processing process; as the target of the forecast, the actual total chemical composition of a substance at a given stage of the industrial process.

[87] Another object of this technology is an electronic device, communicatively connected to a processing plant carrying out the industrial processing process, and the processing plant has a main input channel of raw materials and a main output channel of the final processed substance, which are connected to each other through a series of working chambers, and each working chamber has an intermediate input channel and an intermediate output channel, and the intermediate output channel of the series of chambers supplies the input stream for this intermediate input channel. The electronic device includes a processor and a memory connected to the processor, the memory stores instructions executed on the computer, the execution of which causes the processor to execute: receiving from the processing plant an indication of the partial chemical composition of a substance determined at least in some predetermined area of the processing plant; receiving from the processing plant an indication of the delay time related to the industrial processing process when the substance passes through the processing plant from the input channel of the raw material to the main output channel of the processed substance; performing a physical model to create, based on at least an indication of the lag time and a partial chemical composition of the substance, a predicted indication of the total chemical composition of the substance in at least some intermediate inlets; based on the predicted indication of the complete chemical composition of the substance, creating a thermodynamic model for modeling the industrial processing process, and the thermodynamic model is configured to predict at least: a process parameter at any given stage and at any given time of the industrial processing process; and the complete chemical composition of the substance at any given stage and at any given time in the industrial processing process; using the result of a thermodynamic model as input for a training set for training a machine learning algorithm (MLA) to predict the future total chemical composition of a substance at any given stage of the industrial processing process based on the current process parameter at this stage of the industrial processing process.

[88] In the context of the present description, unless clearly indicated otherwise, "electronic device", "user device", "server", "remote server" and "computer system" means hardware and / or system software suitable for solving the corresponding task. Thus, some non-limiting examples of hardware and / or software include computers (servers, desktops, laptops, netbooks, etc.), smartphones, tablets, network equipment (routers, switches, gateways, and so on) and / or their combination.

[89] In the context of the present description, unless clearly indicated otherwise, "computer-readable medium" and "memory" means absolutely any type and nature of media, and examples without limiting the present technology include RAM, ROM, discs (compact disks, DVDs, floppy disks, hard disks, etc.), USB keys, flash cards, solid state drives, and magnetic tape drives.

[90] In the context of the present description, unless clearly indicated otherwise, an "indication" of an information element may be the information element itself or a pointer, reference, link or other indirect method that allows the recipient of the indication to find a network, memory, database or other computer-readable medium from which the information item can be retrieved. For example, a reference to a document may include the document itself (i.e., its content), or it may be a unique document descriptor that identifies a file with respect to a particular file system, or by some other means convey an indication of a network folder to the recipient. , a memory address, a table in a database, or some other location where the file can be accessed. As will be appreciated by those skilled in the art, the degree of accuracy required for such an indication depends on the degree of initial understanding of how the information exchanged between the recipient and sender of the pointer is to be interpreted. For example, if, before establishing a connection between the sender and the receiver, it is clear that the information element attribute takes the form of a database key for an entry in a specific table of a predetermined database containing the information element, then the transfer of the database key is all that is necessary for the efficient transmission of the information item. element to the recipient, even though the information element itself was not transferred between the sender and the recipient of the indication.

[91] In the context of the present description, unless specifically indicated otherwise, the words "first", "second", "third", etc. are used as adjectives solely to distinguish the nouns to which they refer from each other, and not for the purpose of describing any specific relationship between these nouns. So, for example, it should be borne in mind that the use of the terms "first server" and "third server" does not imply any ordering, assignment to a particular type, history, hierarchy or ranking (for example) of servers / between servers, as well as their use (in itself) does not imply that a "second server" must necessarily exist in a given situation. Hereinafter, as indicated here in other contexts, the mention of the "first" element and the "second" element does not exclude the possibility that they are the same actual real element. So, for example, in some cases, the "first" server and the "second" server may be the same software and / or hardware, and in other cases, they may be different software and / or hardware.

[92] Each embodiment of the present technology pursues at least one of the aforementioned purposes and / or objects, but all are optional. It should be borne in mind that some of the objects of this technology, obtained as a result of attempts to achieve the above-mentioned goal, may not meet this goal and / or may satisfy other purposes, not separately specified here. Additional and / or alternative characteristics, aspects and advantages of embodiments of the present technology will become apparent from the following description, the accompanying drawings, and the appended claims.

BRIEF DESCRIPTION OF DRAWINGS

[93] For a better understanding of this technology, as well as its other aspects and characteristics, reference is made to the following description, which should be used in conjunction with the accompanying drawings, where:

[94] FIG. 1 is a schematic diagram of a system constructed in accordance with a non-limiting embodiment of the present technology.

[95] FIG. 2 shows the electronic device of the system shown in FIG. 1, the electronic device is configured to perform three procedures — a physical model procedure, a thermodynamic model procedure, and a machine learning algorithm (MLA) procedure; they are all performed in accordance with non-limiting embodiments of the present technology.

[96] FIG. 3 is a schematic diagram of a physical model procedure performed in accordance with a non-limiting embodiment of the present technology.

[97] FIG. 4 is a schematic diagram of a thermodynamic model procedure performed in accordance with a non-limiting embodiment of the present technology.

[98] FIG. 5 is a schematic diagram of a machine learning algorithm (ML A) procedure performed in accordance with a non-limiting embodiment of the present technology.

[99] FIG. 6 is a flowchart of a method implemented in accordance with a non-limiting embodiment of the present technology.

[100] It should also be noted that the drawings are not to scale unless otherwise indicated.

IMPLEMENTATION

[101] FIG. 1 is a schematic diagram of a system 100 in accordance with non-limiting embodiments of the present technology. It is important to keep in mind that the following description of system 100 is a description of exemplary embodiments of the present technology. Thus, all of the following description is presented only as a description of an illustrative example of the present technology. This description is not intended to define the scope or delineate the boundaries of this technology. Some useful examples of modifications to system 100 may also be covered by the following description. The purpose of this is also solely to help understand, and not to define the scope and boundaries of this technology. These modifications are not an exhaustive list, and those skilled in the art will appreciate that other modifications are possible. In addition, it should not be interpreted to mean that where it has not already been done, i.e. where no examples of modifications have been set forth, no modifications are possible, and / or that what is described is the only embodiment of this element of the present technical solution. As one skilled in the art will appreciate, this is most likely not the case. In addition, it should be borne in mind that the system 100 is in some specific manifestations a fairly simple implementation of the present technology, and in such cases, this option is presented here for the purpose of ease of understanding. As one skilled in the art will appreciate, many embodiments of the present technology will be of much greater complexity.

[102] In general, system 100 includes a gas processing plant 102 and a supervisory electronic device 104 (or simply "electronic device" 104). It should be noted that in FIG. 1, an electronic device 104 is connected to a gas processing plant 102 via a communications network 105. The connection to the data network 105 can be realized directly with a cable, a local area network (LAN), or a wide area network (WAN) such as the Internet. Therefore, it is possible that the electronic device 104 and the gas processing plant 102 are located at the same location. It is also possible that the electronic device 104 and the gas processing plant 102 are under the control and / or control and / or supervision of the same person. In alternative embodiments of the technology, it is possible that the electronic device 104 and the gas processing plant 102 are located at different locations. It is also possible that the electronic device 104 and the gas processing plant 102 are under the control and / or control and / or supervision of different persons.

[103] The implementation of the gas processing facility 102 is not particularly limited and may include some or all of the following: condensate and water removal unit (s), acid gas removal unit (s), sulfur unit (s), unit ( i) tail gas cleaning, dehydration unit (s), mercury removal unit (s), nitrogen separation unit (s); fractionation unit (s), etc.

[104] The embodiments of the electronic device 104 are not particularly limited, but personal computers (desktop computers, laptops, netbooks, etc.), wireless communication devices (smartphones, mobile phones, tablets, etc.) can be used as an example of the electronic device 104. as well as network equipment (routers, switches or gateways). In alternative embodiments of the present technical solution, the electronic device 104 may be implemented as a server. In an exemplary embodiment of the present technology, the server may be a Dell ™ PowerEdge ™ server running a Microsoft ™ Windows Server ™ operating system. Needless to say, the server can be any other suitable hardware, application software, and / or system software, or a combination thereof. In the presented non-limiting embodiments of the present technology, the electronic device 104 is a single electronic device 104. In other non-limiting embodiments of the present technology, the functionality of the electronic device 104 may be decoupled and performed by multiple electronic devices.

[105] The gas processing plant 102 is configured to perform an industrial processing process. To this end, the gas fractionation chamber (not separately numbered) of the gas processing plant 102 has a main feedstock inlet 106 and a final processed material main outlet 108. The main input channel of the raw material 106 is configured to receive, during operation, the raw material (not separately numbered) intended for processing. The feedstock can be raw natural gas recovered from the natural gas production process.

[106] As shown in FIG. 1, the final processed material main outlet 108 is configured to discharge the final processed product satisfying specific requirements (eg, chemical composition). A plurality of additional processed matter outlets 110 are also provided, the plurality of additional processed matter outlets 110 being configured to separate one or more additional products from the industrial processing process. These other one or more additional products can be considered "by-products" of the final processed product, but they can be useful on their own or as an input stream for another / further processing (as shown in Fig. 1).

[107] The main raw material inlet 106 and the final processed material main outlet 108 (as well as a plurality of additional processed material outlets 110) are connected to each other by a series of working chambers 112. It should be noted that a number of working chambers 112 are also sometimes referred to as a "series of distillation columns" 112. The configuration of a series of working chambers 112 is well known to those skilled in the art and therefore will not be described in detail here.

[108] Each working chamber of the row of working chambers 112 has an intermediate inlet and an intermediate outlet, the intermediate outlet of the row of working chambers providing an input stream for a particular intermediate inlet.

[109] FIG. 1, the first working chamber 114 and the second working chamber 116 are marked. The intermediate outlet 118 of the first working chamber 114 supplies a gas fraction (i.e., the outlet from the first working chamber 114), which is the inlet stream of the intermediate inlet 120 of the second working chamber 116 (i.e. e. inlet flow into the second working chamber 116).

[110] The feedstock entering the main inlet 106 is associated with a chemical composition 119. The chemical composition 119 is not particularly limited as shown in FIG. 1, chemical composition 119 has the plurality of atoms and / or molecules shown in FIG. 1 letters in the Latin alphabet (ie C ₁ , C ₂ , C ₃ , etc.), which are not separately numbered.

[111] When the raw material enters the main inlet channel 106, it enters the first working chamber 114. The first working chamber 114 performs part of the industrial processing, and the raw material that enters the first working chamber 114 is separated into lighter and heavier fractions. so that the lighter fraction rises and is separated through the corresponding intermediate outlet 118 into the second processing chamber 116, and the heavier fraction is lowered and separated through the corresponding intermediate outlet 118 into another subsequent processing chamber (not numbered for simplicity).

[112] The process is repeated through the row of working chambers 112 until reaching the stage where the outlet of a given row of working chambers 112 is the desired product (the desired product has a predetermined chemical composition), for example, dry natural gas of a quality that meets the requirements of transportation through a pipeline that may be the outlet stream of the main outlet 108 of the final processed material. Simultaneously, one or more of the plurality of additional processed material outlets 110 may provide one or more corresponding additional products.

[113] In accordance with non-limiting embodiments of the present technology, one or more process sensors 117 are provided in the architecture of system 100. FIG. 1, the first process sensor 121 and the second process sensor 122 are numbered. This one of the first process sensor 121 and the second process sensor 122 may be implemented as one or more of the following: a thermostat and / or a pressure gauge.

[114] Although FIG. 1, only two process sensors are shown — the first process sensor 121 and the second process sensor 122 — which are required in every embodiment of the present technology. For example, the first process sensor 121 and the second process sensor 122 may be located in some or each of the row of working chambers 112. For example, in a specific non-limiting embodiment of the present technology, each of the plurality of working chambers 112 includes a first process sensor 121 (configured as a thermostat) and a second process sensor 122 (configured as a pressure gauge). Naturally, any structure of the first process sensor 121 and the second process sensor 122 are possible.

[115] In accordance with embodiments of the present technology, the electronic device 104 may receive readings from the first process sensor 121 and the second process sensor 122 (as well as other sensors potentially present in other non-limiting embodiments of the present technology).

[116] In such a way that a given first process sensor 121 and a second process sensor 122 may be located in each of the plurality of working chambers 112, the electronic device 104 may be configured to receive a real-time or near-real-time indication of the process parameters in each of the row of working chambers 112. Alternatively, in these alternative embodiments of the technology, the first process sensor 121 and the second process sensor 122 may be located at respective points (i.e., locations) of the row of working chambers 112, the electronic device 104 may be configured with the ability to receive in real time or almost real time an indication of the process parameters at the corresponding data points of a number of working chambers 112.

[117] Analysis of chemical composition

[118] In the architecture of system 100, a chromatograph unit 130 is represented. In the embodiment of the present technology shown in FIG. 1, the chromatograph is connected to a communications network 105, for example, to transmit and / or receive data from or to an electronic device 104 (as well as any other devices connected to a communications network 105).

[119] The chromatograph unit 130 is configured to perform the chromatography process. In a broad sense, chromatography is a laboratory method for separating a mixture. A mixture dissolved in a liquid is called a mobile phase, which is carried through a structure containing another substance, called a stationary phase. Different constituents of the mixture move at different speeds, which initiates their separation. The separation is based on the difference in separation between the mobile and stationary phases. Substances thus separated can then be analyzed for partial chemical composition.

[120] Thus, the electronic device 104 is also configured to obtain the partial chemical composition of the substance (i.e., gas fractions as they move through the row of working chambers 112) as it passes through the row of working chambers 112. For example , an indication of the partial chemistry can be obtained from a chromatograph unit 130, which analyzes gas fraction samples taken from one or more points in the flow of a substance through a series of working chambers 112 as the material is processed into fractions in a series of working chambers 112.

[121] In some embodiments of the present technology, the chromatograph unit 130 receives samples collected from one or more intermediate output channels 118 of one or more of a number of working chambers 112 and transfers their partial chemical composition to the electronic device 104. An indication of the partial chemical composition may be received by electronic device 104 via data network 105.

[122] In some embodiments of the present technology, the chromatograph unit 130 does not analyze the entire chemical composition of the gas fractions, but only a partial chemical composition of the gas fractions. Specific constituents of the gas fractions are measured as part of the partial chemical composition of the substance determined in advance by the operator of the gas processing plant 102. In a specific example, a subset of the chemical constituents of the gas fractions that significantly affect the quality and / or composition of the final product are selected as part of the partial chemical analysis performed on block 130 of the chromatograph.

[123] In some embodiments of the present technology, the chromatograph unit 130 measures the partial chemistry of the gas fractions at predetermined time intervals. The predefined time slots can be every 5 minutes, every 60 minutes, and so on. In some embodiments of the present technology, the partial chemistry reading may be delayed (i.e., not in real time). This is especially important when the partial chemistry is obtained from a remote test laboratory (ie, when the chromatograph unit 130 is remote from the gas processing plant 102).

[124] With reference to FIG. 2, electronic device 104 is configured to perform three procedures — physical model procedure 202, thermodynamic model procedure 204, and machine learning algorithm (MLA) procedure 206. Broadly speaking, the electronic device 104 can be said to be programmed to execute the physical model routine 202, the thermodynamic model routine 204, and the machine learning algorithm (MLA) routine 206.

[125] In some embodiments of the present technology, the physical model routine 202, the thermodynamic model routine 204, and the machine learning algorithm (MLA) routine 206 are software routines and thus the electronic device 104 can store (in memory 210) those executable on the machine. instructions that cause processor 212 of electronic device 104 to execute physical model routine 202, thermodynamic model routine 204, and machine learning algorithm (MLA) routine 206, as will be described later.

[126] In alternative embodiments of the technology, each of the physical model routine 202, thermodynamic model routine 204, and machine learning algorithm (MLA) routine 206 may be executed by appropriate electronic devices (not shown) that have processors and machine-readable codes to perform routines that will be described below.

[127] Naturally, the physical model routine 202, thermodynamic model routine 204, and machine learning algorithm (MLA) routine 206 may be implemented in any combination of application software, hardware, system software, or a combination thereof.

[128] Also within the architecture shown in FIG. 2, external memory 240 is illustrated. Note that in alternative embodiments of the technology, external memory 240 may be implemented as part of memory 210. In additional alternative embodiments of the present technology, external memory 240 may be implemented in a distributed fashion.

[129] External memory 240 may store data representing one or more of the following:

• historical and current indicators of one or more sensors 117 of the process;

• historical partial composition of the gas fraction at this stage of the industrial process in the past;

• results of analysis of the chemical composition of the gas (actual results obtained in the past);

• meta-description of system 100 and its components.

[130] A description of each of the physical model procedure 202, thermodynamic model procedure 204, and machine learning algorithm (MLA) procedure 206 will now be presented.

[131] Procedure 202 physical model

[132] In a broad sense, the purpose of the physical model procedure 202 can be described as follows. The actual measured chemistry is determined at each stage. Additionally, block 130 chromatograph analyzes the partial composition of the samples, but does not analyze the output stream (upper and lower intermediate outlet channels 118 of the corresponding working chamber). However, embodiments of the present technology require, as input to the thermodynamic model routine 204, an indication of the fractions exiting both the upper and lower intermediate exit channels 118. In some embodiments of the present technology, the physical model routine 202 additionally uses in advance certain heuristic analysis.

[133] Without being limited to any particular theory, the physical model procedure 202 is based on the following premises. The time it takes for the gas fraction to pass one of the row of working chambers 112 is constant (lag_time). In other words, the time it takes for the gas fraction to travel from the inlet to the upper outlet and the outer outlet remains the same (lag_time).

[134] The path that the gas fraction passes through one of the row of working chambers 112 can be divided into three stages. In the first step, all of the gas fraction that enters one of the row of working chambers 112 is concentrated at one point (close to the intersection of the intermediate inlet 120 and the inlet pipe that leads to it). At the second stage, the gas fraction is equally distributed in one of the number of working chambers 112. At the last stage, the concentration of a particular gas fraction is reduced by a factor of (1 + conc), where conc is the ratio of the mass of a particular gas fraction to the total mass of the gas fraction in the row of working chambers 112.

[135] In some embodiments of the present technology, the thermodynamic model routine 204 is generated based on known (measured) parameters. As a non-limiting example, the thermodynamic model routine 204 is generated based on the known (measured) values of the processing parameters. The delay_time parameter can be determined based on the number of hot plates. Alternatively, the delay_time parameter can be created by multiplying the number of hot plates (which intersect with the inlet pipe) and the upper intermediate outlet 118 multiplied by the time it takes for the gas fraction to pass through the hot plate. Additionally, the parameter conc is used, which is created as the average mass of the gas fraction (for example, entering one of the series of working chambers 112 for a predetermined time period, for example, per second) from the total mass of the gas fraction in this one of the number of working chambers 112.

[136] The physical model routine 202 uses mathematical models to create a vector V, where a given iteration i of vector V is the mass of a gas fraction (predetermined chemical composition) that exits a particular intermediate outlet 118 i minutes after that batch is triggered. Once the vector V has been generated (and knowing the values of the upper and lower output channels), heuristic analysis can be applied to determine, at every minute (or any other time interval), the mass of the gas fraction exiting through the upper and lower intermediate outputs 118.

[137] The generated vectors V can be organized into a vector matrix M. As an example, assume that a given gas fraction, which is included in a given

intermediate input channel 120 at time t, which was divided through the upper intermediate output channel 118, is equal to the sum of all elements of the matrix M located on the diagonal, which is defined by the index (t, 0), (t + 1, 1), ..., ( t + i, i) etc.

This expression represents the specific average (weighted) values of all incoming gas fractions in such a way that all the weighted values are added to one. The output of each gas fraction at a given 118 has a lag coefficient that correlates with the time delay of a given outlet channel relative to the beginning of the process.

[138] With reference to FIG. 3, in accordance with non-limiting embodiments of the present technology, the physical model routine 202 is configured to obtain an indication of a time lag 302 and an indication of a partial chemical composition of a substance 304.

[139] The reference to the delay time 302 is the delay time related to the industrial processing process as the substance passes through the gas processing plant 102 from the main feed inlet 106 to the main processed substance outlet 108.

[140] The physical model routine 202 performed by the electronic device 104 is configured to generate, based on at least the lag time and the partial chemical composition of the material fraction, a predicted indication of the total chemical composition 306 of the gas fraction in at least some intermediate inlets 120.

[141] In some embodiments of the present technology, the physical model routine 202 is configured to determine the overall chemical composition 306 of the gas fraction at each intermediate channel 120. The latter is especially characteristic (but not limited) to those embodiments of the technology where an indication of the partial chemical composition substances 304 are obtained from each of the intermediate exit channels 118).

[142] In some additional embodiments of the present technology, the physical model routine 202 is configured to determine the overall composition 306 of the gas fraction of the feed when it enters the main inlet 106 of the feed (in addition to or instead of the total chemical composition 306 of the gas fraction on some or all intermediate output channels 120).

[143] Broadly, based on the time lag and the resulting partial chemistry, the physics model procedure 202 "restores" the complete chemistry of the gas fraction as it "passes" through the gas processing plant 102. As a specific example (but not limitations), the resulting partial composition may be the partial chemical composition of the gas fraction as it exits each (or some) of the series of working chambers 112 (i.e., the chemical composition of the gas fraction measured at the intermediate outlet 118), and the recovered total the chemical composition can be the chemical composition of the gas fraction when it enters the respective each (or some) of the number of working chambers 112 (ie, each of the intermediate channels 120).

[144] Procedure 204 of the thermodynamic model

[145] Broadly, the thermodynamic model procedure 204 is used to simulate the refining process at the gas refinery 102. Without being bound by any particular theory, the refining process simulations are performed in an environment where specific measurements are not / are not available. More specifically, the thermodynamic model procedure 204 creates some of the undetected parameters of the industrial processing process based on specific discovered or otherwise known parameters of the industrial processing process and / or the total chemical composition of the substance (or fraction of the substance) that is the subject of the industrial processing process.

[146] The thermodynamic model routine 204 is generated from historical data (actual measured process parameters, actual measured chemical gas fraction), as well as data generated by the physical model routine 202 (ie, simulated data).

[147] FIG. 4, the thermodynamic model routine 204 uses as input: the total chemistry 306 of the gas fraction in at least some of the intermediate inlets 120 (generated by the physical model routine 202) and the measured historical data 308. In some embodiments of the present technology, the thermodynamic routine 204 the model can also use, as ancillary input, additional data 310 (eg, process parameters, and so on).

[148] The procedure 204 of the thermodynamic model is configured to create, based on the input information, a reconstructed process parameter 412. The recreated process parameter 412 is one or more parameters that are not measured and / or not available from the physical model routine 202.

[149] More specifically, the thermodynamic model routine 204 is configured to create, as the reconstructed process parameter 412, one or more of: a process parameter at any given stage or at any given time in an industrial processing process; and the complete chemical composition of the substance at any given stage and at any given time in the industrial processing process.

[150] After creating and validating the thermodynamic model routine 204 (as described in more detail below), the thermodynamic model routine 204 is configured to generate any missing information about the industrial processing process and / or the substance (gas fraction) that is being industrialized. Additionally or alternatively, the thermodynamic model procedure 204 allows the identification of those process variable sensors that are out of order and give incorrect readings.

[151] Procedure 206 MLA

[152] In a broad sense, MLA 260, after training (as will be described in more detail below), is configured to generate a prediction of the chemical composition of a substance (or gas fraction) at any given stage of the industrial processing process for a given point in time in the future based on current at that point in the process parameters. In other words, the MLA 260 is configured to predict a process variable for a future point in time, and future points in time represent the same or later point in time than the "current process variable" measurement.

[153] In a particular embodiment of the technology, the horizon for predicting the future depends on the frequency of readings of the process parameters and / or the frequency of readings of the partial chemistry that was used to train the MLA.

[154] FIG. 5 first describes the training process of procedure 206 MLA. In accordance with non-limiting embodiments of the present technology, the MLA procedure 206 is “fed” a training object 502, wherein the training object 502 is a plurality of training objects 504. A training object 502 (an example of a plurality of training objects 504) comprises a target component 506 and a feature component 508.

[155] Target component 506 is the actual (in real time) total chemical composition of a substance (or gas fraction) at a given point in the industrial processing. Feature component 508 represents process parameters associated with the target component 506 (ie, taken at the same time / stage of the industrial processing process). The content of the feature component 508 may represent one or more process sensors 117, with one or more process sensors 117 present at a given stage in the industrial process associated with the overall chemical composition of the substance (or gas fraction).

[156] Thus, it can be said that in accordance with non-limiting embodiments of the present technology, a learning object 502 from a plurality of learning objects 504 is created based on information received from one or more process sensors 117 and the total chemistry generated by the thermodynamic procedure 204. models. In some embodiments of the present technology, the result of the thermodynamic model routine 204 is used to train the MLA routine 206 only after verifying the result of the thermodynamic model routine 204. How the result of the thermodynamic model procedure 204 is verified will be described later.

[157] As part of the training process, the MLA procedure 206 is fed a plurality of training objects 504 to create the MLA formula 510. The MLA Formula 510, in a sense, learns the correlation and / or latent relationship of information received from one or more process sensors 117 and the corresponding total chemistry generated by the thermodynamic model routine 204.

[158] After training (and validation), procedure 206 MLA (using the 510 MLA formula) is performed to predict the total chemical composition of a substance (or gas fraction) at any given stage of the industrial processing process based on the then current industrial processing parameter.

[159] More specifically, the MLA formula 510 receives a work object 512 (eg, an actual process variable at a given time in the future) and the results of a work forecast 514. A work object 512, for example, a process variable metric from one or more process sensors 117 and a work forecast 514 represents the predicted total gas fraction chemistry (at a given time in the future or at any time thereafter, within the predicted range of Formula 510 MLA). In other words, it should be noted that the work forecast 514 can be made for any current moment in time (i.e., "now") with respect to a given moment in the future (i.e., the time when the metric is taken against the work object 512) or for a specific horizon in the future. Work forecast 514 can be used to monitor industrial processing with electronic device 104.

[160] For example, the electronic device 104, based on the working forecast 514 of the total chemical composition of the gas fraction at a given stage of the industrial processing process, can determine the most optimal setting of the process parameters and adjust the process parameters to this optimal setting.

[161] Formula 510 MLA can be created using one or more models. The following are examples of some of the models that can be used to create a 510 MLA formula.

[162] Lasso regression, i.e. linear regression with L1-regularization, which allows you to ignore clearly irrelevant features.

[163] A sequence (pipeline) using standard out-of-order processing that replaces abnormally high or abnormally low values with a predetermined threshold value. The predefined threshold can be based on historical extremes as well as Lasso regression.

[164] LightGBM, decision tree model using light gradient boosting.

[165] Verification and Adjustment Procedures

[166] In accordance with non-limiting embodiments of the present technology, electronic device 104 is further configured to perform one or more model checking procedures and / or one or more model updating procedures. In some non-limiting embodiments of the present technology, one or more model checking procedures and / or one or more model adjusting procedures are performed taking into account the temporal nature of parameters associated with the industrial processing process. For example, training objects sorted by time can be divided into k parts. Education model used to predict the value of j ^th (where 1 <j ≤ k) represents the training using the (j-1) of the k parts.

[167] In accordance with non-limiting embodiments of the present technology, the electronic device 104 is configured to check the result of the thermodynamic model routine 204. In some embodiments of the present technology, the result of the thermodynamic model procedure 204 is validated before the result of the thermodynamic model procedure 204 is used to train the MLA procedure 206. In some embodiments of the present technology, the result of the thermodynamic model routine 204 is used to train the MLA routine 206 in response to a positive result verification result.

[168] In some embodiments of the present technology, verification of the thermodynamic model routine 204 may be performed as follows. It will be appreciated that the thermodynamic model routine 204 is generated based on historical data (actual measured process parameters, actual measured chemical gas fraction), as well as data generated by the physical model routine 202 (i.e., simulated data).

[169] In some embodiments of the present technology, the thermodynamic model routine 204 is then generated based on the auxiliary data 310. In some embodiments of the present technology, the auxiliary data 310 includes actual process parameters. In some embodiments of the present technology, creating a thermodynamic model routine 204 includes creating a thermodynamic model routine 204 using a subset of process parameters excluding that process parameter; and checking the thermodynamic model procedure 204 includes: comparing the generated value for the given process variable with the actual value of the given process variable.

[170] For example, training a thermodynamic model routine 204 may use a temperature reading as ancillary data 310, in which case the generated pressure value is used for comparison with the actual pressure reading. If the generated value and the actual value are within a predetermined error, the thermodynamic model procedure 204 will be considered valid. If the generated value and the actual value are not within a predetermined error, the thermodynamic model routine 204 is corrected. In some embodiments of the present technology, the adjustment may be performed as an iterative calibration of at least one of the physical model procedure 202 and the thermodynamic model procedure 204 until a positive test is obtained for the thermodynamic model procedure 204.

[171] In some non-limiting embodiments of the present technology, the electronic device 104 is configured to apply the correction model 5 to the thermodynamic model procedure 204. In a particular embodiment, the adjustment model for the thermodynamic model is a linear adjustment model. The linear adjustment model can be based on predetermined values of various processing factors.

[172] In some embodiments of the present technology, electronic device 106 is configured to use an MLA validation metric to validate the MLA procedure 206. In some embodiments of the present technology, the mean absolute error (MAPE) metric is used as follows:

[173] where:

[174] c - analyte;

[175] n is the number of measurements;

[176] i is the number of this dimension;

[177] t is the point in time when the forecast is performed;

[178] h - forecast time interval;

[179] A (s, t + h) _i actual measurement at time t + h;

[180] F (c, t) _{i is the} predicted value for measurement generated by the thermodynamic model for time t.

[181] Broadly, the MAPE metric indicates the mean prediction error expressed as a percentage. First, the MAPE metric calculates the variance between each predicted value and the corresponding actual reading; the MAPE metric further expresses the percent variance of the actual reading and finally the percentages are averaged to arrive at the final MAPE metric.

[182] Based on the MAPE metric, a quality score vector can be obtained. In accordance with non-limiting embodiments of the present technology, a given quality score vector is associated with MLA prediction errors for a given analyte. More specifically, this vector of quality assessment can be expressed as follows:

[183] In which:

[184] VA - assessed MLA;

[185] c _i is a component of a substance;

[186] MAPE (ci) - forecast error for component ci.

[187] In alternative embodiments of the technology, the MAPE formula may be expressed as follows.

[188] ude

[189] c - analyte;

[190] n is the number of measurements;

[191] Ai - actual measurement value;

[192] Fi is the predicted value for the measurement generated by the thermodynamic model for time t.

[193] In another non-limiting embodiment of the present technology, the CORR metric may be used. The CORR metric can be expressed as follows.

[194] where

[195] c - analyte;

[196] n is the number of measurements;

[197] i is the number of this dimension;

[198] Ai - actual measurement value;

[199] Fi is the predicted value for the measurement generated by the thermodynamic model for time t.

[200] The CORR metric indicates the correlation and / or the presence (and strength) of a linear 15 correlation between the MLA's prediction and actual performance. The higher the CORR metric, the stronger the linear correlation.

[201] In another non-limiting embodiment of the present technology, the RMSE metric may be used. The RMSE metric can be expressed as follows:

[202] where:

[203] c - analyte;

[204] n is the number of measurements;

[205] i is the number of this dimension;

[206] Ai - actual measurement value;

[207] Fi is the predicted value for the measurement generated by the thermodynamic model for time t.

[208] The RMSE metric indicates a prediction error. Compared to the MAPE metric, the RMSE metric penalizes the model used by the MLA for large deviations (in absolute values) of the forecast from the actual readings.

[209] Taking into account the architecture described above, it is possible to implement a method for predicting the chemical composition of a substance used in the process of industrial processing, and the process of industrial processing is carried out in a gas processing plant 102, which has a main input channel of raw materials 106 and a main output channel 108 of the final processed substance, which connected to each other by a series of working chambers 112, each working chamber having an intermediate inlet 120 and an intermediate outlet 118, and the intermediate outlet 118 of the row of working chambers supplies the input stream for this intermediate inlet 120, to which it is directly connected.

[210] FIG. 6 is a flow diagram of a method 600 that is performed in accordance with non-limiting embodiments of the present technical solution. The method 600 may be executed by an electronic device 104 communicatively coupled to the gas processing plant 102. It should be noted that although the method 600 has been described with reference to the gas processing plant 102 and its industrial processing process, this is not necessarily a limitation in every an alternative embodiment of the present technology. Thus, method 600 may be adopted for other types of industrial processing processes (eg, biological treatment, chemical processing, water filtration systems, etc.). Method 600 can also be applied to an oil refining process.

[211] It should be noted that in alternative embodiments of the technology, method 600 may be applied to other industrial processes. An example of another industrial process is the separation of liquid air into oxygen and nitrogen.

[212] Step 602 - obtaining from the processing plant an indication of the partial chemical composition of a substance determined at least at some predetermined site of the processing plant

[213] The method 600 begins at block 602, where the electronic device 104 receives from the processing plant 102 an indication of the partial chemical composition of the substance determined at least at some predetermined location of the processing plant 102.

[214] In some embodiments of the present technology, electronic device 104 can retrieve information from external memory 240 (which can store data representing a partial chemical composition that has been determined in the past and stored in external memory 240).

[215] As mentioned earlier, the partial chemical composition can be determined, for example, using a chromatograph. The chemical composition process is usually done "offline" in the sense that laboratory tests may be necessary to determine the chemical composition of a substance undergoing industrial processing.

[216] In alternative embodiments of the technology, the electronic device 104 may receive an indication of the partial chemical composition from another element other than the external memory 240. For example, the electronic device 104 may receive an indication of the partial chemical composition directly from a laboratory using chromatographic analysis or the like. like that.

[217] Step 604 - Obtaining from the processing plant an indication of the lag time associated with the industrial processing process as the substance passes through the processing plant from the feed inlet to the main processed material outlet

[218] In step 604, the electronic device 104 receives from the processing plant 102 an indication of the lag time associated with the industrial processing process as the substance passes through the processing plant from the raw material inlet 106 to the main processed material outlet 108. In some non-limiting example, data that represents the delay time may be entered by an operator of electronic device 104 based on information stored in external memory 240 (information obtained from gas processing plant 102 and stored in external memory 240).

[219] Step 606 - Executing a physical model to create, based at least an indication of the lag time and a partial chemical composition of the substance, a predicted indication of the total chemical composition of the substance in at least some intermediate input channels

[220] In step 606, electronic device 104 performs a physical model routine 202 to create, based on at least the lag time and partial chemistry indication, a predicted total chemistry indication in at least some intermediate inlets 120.

[221] In some embodiments of the method 600, the method further includes: using a physical model to create, based on the lag time, a time division of the industrial processing process into each of a number of working chambers, the time division being used to generate a predictable indication of the total chemical the composition of the substance in at least some intermediate inlet channels.

[222] Step 608 - based on the predicted indication of the full chemical composition of the substance, creating a thermodynamic model for modeling the industrial processing process, and the thermodynamic model is configured to predict at least: a process parameter at any given stage and at any given time of the industrial process processing; and the complete chemical composition of the substance at any given stage and at any given time in the industrial processing process

[223] At block 608, the electronic device generates, based on the predicted indication of the total chemical composition of the substance, a thermodynamic model procedure 204 for simulating an industrial processing process, wherein the thermodynamic model procedure 204 is configured to predict at least: a process parameter at any given step and at any given time in the industrial processing process; and the complete chemical composition of the substance at any given stage and at any given time in the industrial processing process.

[224] In some embodiments of method 600, method 600 further includes checking a thermodynamic model. As previously mentioned, in some embodiments of the technology, the MLA 260 training step is performed only in response to a positive test of the thermodynamic model procedure 204. Prior to a positive verification result, method 600 may further include the step of iteratively calibrating at least one of the physical model and the thermodynamic model until a positive verification result is obtained.

[225] It will be recalled that in some embodiments of the present technology, the thermodynamic model routine 204 is then generated based on the auxiliary data 310. In some embodiments of the present technology, the auxiliary data 310 includes actual process parameters. In some embodiments of the present technology, creating a thermodynamic model routine 204 includes creating a thermodynamic model routine 204 using a subset of process parameters excluding that process parameter; and checking the thermodynamic model procedure 204 includes: comparing the generated value for the given process variable with the actual value of the given process variable.

[226] Testing the thermodynamic model procedure 204 may further include applying the MAPE formula:

[227] where:

[228] c - analyte;

[229] n is the number of measurements;

[230] i is the number of this dimension;

[231] t is the point in time when the prediction is made;

[232] h - forecast time interval;

[233] A (s, t + h) _i actual measurement at time t + h;

[234] F (c, t) _{i is the} predicted value for measurement generated by the thermodynamic model for time t.

[235] Step 610 - Using the result of the thermodynamic model as input to a training set for training a machine learning algorithm (MLA) to predict the future total chemical composition of a substance at any given stage of the industrial processing process based on the current process parameter at this stage of the industrial processing process

[236] At block 610, the electronic device 104 uses the result of the thermodynamic model procedure 204 as input to a training set for training machine learning algorithm (MLA) 206 to predict the future total chemical composition of a substance at any given stage of the industrial processing process based on the current parameter. process at this stage of the industrial processing process.

[237] It should be noted that the forecast of the future total chemical composition of a substance during operation can be performed for various points in the future. The technical effect of some of these embodiments of the technology provides the ability to calculate / predict future total chemistry for multiple time intervals, multiple future total chemistries are periodically predicted at time intervals to determine the partial chemistry using a chromatograph.

[238] It is important to keep in mind that not all of the technical results mentioned herein may be manifested in every embodiment of the present technology. For example, the embodiments of the present technology can be performed without showing some technical results, others can be performed with the manifestation of other technical results or not at all.

[239] Some of these steps, as well as signal transmission / reception processes, are well known in the art and have therefore been omitted in some parts of this specification for simplicity. Signals can be transmitted-received by optical means (eg, fiber optic connection), electronic means (eg, wired or wireless), and mechanical means (eg, based on pressure, temperature, or other suitable parameter).

[240] Modifications and improvements to the above described embodiments of the present technology will be apparent to those skilled in the art. The foregoing description is provided by way of example only and does not establish any limitation. Thus, the scope of the present technology is limited only by the scope of the appended claims.

Claims (54)

1. A method for training a machine learning algorithm (MLA) acting as a virtual sensor to predict the chemical composition of a substance used in an industrial processing process carried out in a processing plant that has a main input channel of raw material and a main output channel of the final processed substance, which are connected with each other by means of a number of working chambers, each of which has an intermediate inlet channel and an intermediate outlet channel, and this intermediate outlet channel of the series of chambers supplies the input stream for this intermediate inlet channel, a method implemented on an electronic device communicatively connected to a processing plant, a method including: receiving from the processing plant an indication of the partial chemical composition of the substance, determined at least in some predetermined area of the processing plant; receiving from the processing plant an indication of the delay time related to the industrial processing process when the substance passes through the processing plant from the input channel of the raw material to the main output channel of the processed substance; performing a physical model to create, based on at least an indication of the lag time and a partial chemical composition of the substance, a predicted indication of the total chemical composition of the substance in at least some intermediate inlets; based on the predicted indication of the complete chemical composition of the substance, the creation of a thermodynamic model for modeling the process of industrial processing, and the thermodynamic model is configured to predict at least: a process parameter at any given stage and any given point in time of the industrial processing process; and full chemical composition of the substance at any stage and at any time of the industrial processing process; using the result of a thermodynamic model as input for a training set for learning to predict the future full chemical composition of a substance at any given stage of the industrial processing process based on the current process parameter at this stage of the industrial processing process. 2. The method according to claim 1, further comprising: the use of a physical model to create, based on the delay time, the time breakdown of the industrial processing process for each of a number of working chambers, and the temporal disaggregation is used to generate a predicted indication of the overall chemical composition of a substance in at least some intermediate inlets. 3. The method of claim 1, wherein the method further includes checking the thermodynamic model. 4. The method according to claim 3, wherein the learning step is performed only in response to a positive result of checking the thermodynamic model. 5. The method of claim 4, wherein the method further includes iteratively calibrating at least one of the physical model and the thermodynamic model until a positive test result is obtained. 6. The method of claim 3, wherein checking the thermodynamic model includes applying the mean absolute error (MAPE) formula:

where c is the analyte; n is the number of measurements; i is the number of this measurement; t is the moment in time when the forecast is performed; h - forecast time interval; A (c, t + h) _i - actual measurement at time t + h; F (c, t) _i is the predicted value for measurement, generated by the thermodynamic model for time t. 7. The method according to claim 3, in which the process parameter is a set of parameters of the industrial processing process, and in which: creating a thermodynamic model includes creating a thermodynamic model using a subset of process parameters, excluding this process parameter; and in which verification of the thermodynamic model includes: compares the generated value for this process variable with the actual value for this process variable. 8. The method of claim 7, wherein the process variable is either a pressure parameter or a temperature parameter. 9. The method of claim 1, wherein the physical model generates a predicted indication of the total chemical composition of a substance based on: time delay indications; partial chemical composition of the substance; at least one parameter of the industrial processing process. 10. The method according to claim 9, in which creating a thermodynamic model includes creating a thermodynamic model using a subset of process parameters, excluding the given process parameter; and in which the given process variable is used to validate the thermodynamic model. 11. The method of claim 1, wherein the current process variable is either a pressure parameter or a temperature parameter. 12. The method of claim 11, wherein the current process variable further includes an indication of a future partial chemistry at a given stage in the industrial process. 13. The method of claim 1, wherein the method further includes, in an operation phase that occurs during operation, using the MLA to predict future overall chemistry based on process parameters during operation. 14. The method of claim 13, wherein the method further includes obtaining an indication of a process parameter during operation and using the parameter to predict the overall chemistry. 15. The method of claim 14, wherein the future overall chemistry is used to control at least one parameter of the industrial processing process. 16. The method of claim 13, wherein the future gross chemistry at time of operation includes a plurality of future gross chemistries predicted for multiple time intervals, and wherein a plurality of future gross chemistries are periodically predicted for time intervals that are shorter, than the time intervals for determining the partial chemical composition. 17. The method of claim 1, wherein the at least one predetermined portion of the processing plant is one of the intermediate outlet channels. 18. The method of claim 17, wherein at least one of the intermediate output channels is all of the intermediate output channels. 19. The method of claim 1, wherein the predicted indication of total chemical composition at at least some intermediate inlets is a predictive indication of total chemical composition of all intermediate inlets. 20. The method of claim 1, wherein training the MLA includes using the result of the thermodynamic model to create an MLA training set, the training set including: as an input feature, at least one process parameter associated with the industrial processing process at this stage of the industrial processing process; as the target of the forecast, the actual total chemical composition of a substance at a given stage of the industrial process. 21. An electronic device for training a machine learning algorithm (MLA), acting as a virtual sensor for predicting the chemical composition of a substance used in the industrial processing process, a device communicatively connected to a processing plant carrying out the industrial processing process, and the processing plant has a main input channel raw material and the main outlet channel of the final processed substance, which are connected to each other by means of a number of working chambers, and each working chamber has an intermediate inlet channel and an intermediate outlet channel, and the intermediate outlet channel of the series of chambers supplies the input stream for this intermediate inlet channel; an electronic device that includes a processor and memory connected to the processor, and the memory stores instructions executable on the computer, the execution of which initiates the processor to perform: receiving from the processing plant an indication of the partial chemical composition of the substance, determined at least in some predetermined area of the processing plant; receiving from the processing plant an indication of the delay time related to the industrial processing process when the substance passes through the processing plant from the input channel of the raw material to the main output channel of the processed substance; performing a physical model to create, based on at least an indication of the lag time and a partial chemical composition of the substance, a predicted indication of the total chemical composition of the substance in at least some intermediate inlets; based on the predicted indication of the complete chemical composition of the substance, the creation of a thermodynamic model for modeling the process of industrial processing, and the thermodynamic model is configured to predict at least: a process parameter at any stage and at any time of the industrial processing process; and full chemical composition of the substance at any stage and at any time of the industrial processing process; using the result of the thermodynamic model as input for a training set for training to predict the future total chemical composition of a substance at any stage of the industrial processing process based on the current process parameter at this stage of the industrial processing process.

Images