Method for generating artificial intelligence model for process control, process control system based on artificial intelligence model, and reactor including same

Abstract

The present invention relates to a method for generating an AI model for process control, a process control system based on the AI ​​model, and a reactor including the same, and includes an AI control model unit 100 including a data storage unit 110 for storing a plurality of preset process data of the reactor, a data correction unit 120 for generating learning data by removing outliers from the stored process data of the reactor, and a data derivation unit 130 for learning from the generated learning data and deriving optimal reactor input conditions for satisfying the reactor operating conditions and the physical property values ​​of the product by the reactor, an input unit 200 for obtaining data including target reactor operating conditions and target physical property values ​​of the product by the reactor and providing the data to the AI ​​control model unit 100, and an output unit 300 for receiving optimal reactor input conditions for satisfying the target reactor operating conditions and the target physical property values ​​of the product from the AI ​​control model unit 100 and controlling the input of the reactor under the optimal reactor input conditions.

Description

[Technical field]

[0001] The present invention relates to a method for generating an artificial intelligence model for process control, a process control system based on an artificial intelligence model, and a reactor including the same. [Background technology]

[0002] In chemical processes, various substances react with each other, and different reactivities appear depending on conditions such as temperature, pressure, composition, etc. Also, since a single factory often produces a variety of products, it is difficult to understand all the chemical processes for the products, and it is either impossible to measure or takes a long time to confirm the measurement results.

[0003] Therefore, controlling chemical processes is difficult, resulting in off-spec products and wasted energy. In addition, if a user arbitrarily changes input conditions to correct the production conditions for such defective products to on-spec production conditions, there is a high possibility of trial and error. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Korean Patent Publication No. 10-2021-0027668 (Publication date: 2021.03.11) Summary of the Invention [Problem to be solved by the invention]

[0005] In order to solve the above-mentioned problems, the present invention provides a method for generating an artificial intelligence model for process control, a process control system based on the artificial intelligence model, and a reactor including the same, thereby reducing the rate of defective products and increasing the rate of normal products generated when producing products using a reactor. [Means for solving the problem]

[0006] As one means for achieving the above object, according to one embodiment of the present invention, there is provided an artificial intelligence model-based process control system including: an artificial intelligence control model unit 100 including a data storage unit 110 for storing preset process data of a plurality of reactors; a data correction unit 120 for generating learning data by removing outliers from the stored process data of the reactor; and a data derivation unit 130 for learning from the generated learning data and deriving optimal reactor input conditions for satisfying physical property values ​​of a product by the reactor; an input unit 200 for obtaining data including target operating conditions of the reactor and target physical property values ​​of a product by the reactor and providing the data to the artificial intelligence control model unit 100; and an output unit 300 for receiving optimal reactor input conditions for satisfying the target operating conditions of the reactor and the target physical property values ​​of the product from the artificial intelligence control model unit 100 and controlling reactor input under the optimal reactor input conditions, wherein the reactor input conditions include the following (a): (a) one or more of the composition, temperature, flow rate, and pressure of the raw material fed into the reactor, or a combination thereof

[0007] In addition, according to one embodiment of the present invention, the reactor charging conditions may further include the following (b): (b) one or more of the composition, temperature, flow rate, and pressure of the catalyst introduced into the reactor, or a combination thereof;

[0008] In addition, according to one embodiment of the present invention, the preset process data of the plurality of reactors may include actual input conditions of the reactors, actual operating conditions of the reactors, and actual physical property values ​​of the products of the reactors.

[0009] In addition, according to one embodiment of the present invention, the data derivation unit 130 derives predicted operating conditions of the reactor and predicted physical property values ​​of the product from the reactor based on the reactor input conditions provided from the input unit 200, and the artificial intelligence control model unit 100 may further include a data analysis unit 140 that compares the actual operating conditions of the reactor and the actual physical property values ​​of the product in the data storage unit 110 or the data correction unit 120 with the predicted operating conditions of the reactor and the predicted physical property values ​​of the product derived by the data derivation unit 130, and a data re-learning unit 150 that re-learns the data derivation unit 130 when the comparison result provided from the data analysis unit 140 satisfies the following condition (1) or condition (2). (1) When the error rate between the actual operating conditions of the reactor and the predicted operating conditions of the reactor exceeds a preset allowable value. (2) If the error rate between the actual physical property value of the product and the predicted physical property value of the product exceeds a preset allowable value.

[0010] In addition, according to an embodiment of the present invention, when the target property value provided from the input unit 200 is changed during the operation of the reactor, the data derivation unit 130 may be characterized in that it analyzes dynamic characteristics of reactor input conditions for reaching the changed target property value from the time when the target property value is changed, and derives new optimal reactor input conditions.

[0011] According to an embodiment of the present invention, the artificial intelligence control model unit 100 may be trained by one or more of the following methods or an ensemble of combinations: linear regression, logistic regression, decision tree, random forest, support vector machine, gradient boosting, convolution neural network, recurrent neural network, long-short term memory, attention model, transformer, generative adversarial network, and reinforcement learning. According to an example, the artificial intelligence control model unit 100 may be a model developed based on the above-mentioned learning methods.

[0012] As another means for achieving the above object, according to one embodiment of the present invention, there is provided a reactor including the above artificial intelligence-based process control system.

[0013] According to an embodiment of the present invention, the reactor may be one of a tubular reactor, a column reactor, a stirred tank reactor, a fluidized bed reactor, and a loop reactor.

[0014] According to an embodiment of the present invention, the reactor may be a plurality of reactors, which may be the same or different from each other, and each reactor may be independently one of a tubular reactor, a column reactor, a stirred tank reactor, a fluidized bed reactor, and a loop reactor.

[0015] As another means for achieving the above-mentioned object, according to one embodiment of the present invention, there is provided a method for generating an artificial intelligence model for process control, comprising the steps of: storing a plurality of reactor process data including actual reactor input conditions, actual reactor operating conditions, and actual physical property values ​​of a product by the reactor; or a calculation result by simulation; removing outliers from the stored reactor process data to generate learning data; and generating an artificial intelligence model including an artificial intelligence algorithm that learns the generated learning data to derive optimal reactor input conditions to satisfy predicted reactor operating conditions, predicted physical property values ​​of a product, and the reactor operating conditions and physical property values ​​of a product by the reactor, wherein the reactor input conditions include the following (a): (a) one or more of the composition, flow rate, and partial pressure of the raw material fed to the reactor, or a combination thereof;

[0016] In addition, according to one embodiment of the present invention, the reactor charging conditions may further include the following (b): (b) one or more of the composition, temperature, flow rate, and pressure of the catalyst introduced into the reactor, or a combination thereof;

[0017] In addition, according to one embodiment of the present invention, the artificial intelligence algorithm may be characterized in that, when the physical property value of the product is changed during the operation of the reactor, the artificial intelligence algorithm may analyze dynamic characteristics of the reactor input conditions to reach the changed physical property value from the time when the physical property value is changed, and derive new optimal reactor input conditions.

[0018] Also, according to an embodiment of the present invention, the step of generating the artificial intelligence model may be characterized by generating the artificial intelligence model by one or more of linear regression, logistic regression, decision tree, random forest, support vector machine, gradient boosting, convolution neural network, recurrent neural network, long-short term memory, attention model, transformer, generative adversarial network, and reinforcement learning, or a combination thereof. Effect of the Invention

[0019] According to one embodiment of the present invention, the present invention can easily derive optimal reactor input conditions for satisfying the target reactor operating conditions and the target physical properties of the product produced by the reactor using an artificial intelligence model.

[0020] According to one embodiment of the present invention, the reactor input can be controlled under the optimal reactor input conditions, thereby reducing the unit cost, time, etc. required for product production.

[0021] According to one embodiment of the present invention, by controlling the process using an artificial intelligence model, the intervention of an external user (such as an administrator) can be minimized, trial and error can be reduced, and a reliable reactor product can be obtained. [Brief description of the drawings]

[0022] [Figure 1] 1 is a flow chart of a method for generating an artificial intelligence model for process control according to an embodiment of the present invention. [Diagram 2] 1 is a block diagram showing an artificial intelligence model process control system according to an embodiment of the present invention; [Diagram 3] FIG. 2 is a re-learning configuration diagram of the artificial intelligence control model unit 100 according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0023] The terms used in this application are merely used to describe specific examples. Thus, for example, singular expressions include plural expressions unless the context clearly requires otherwise. Furthermore, it should be noted that the terms "include" or "comprise" used in this application are used to clearly indicate the presence of features, steps, functions, components, or combinations thereof described in the specification, and are not used to preclude the presence of other features, steps, functions, components, or combinations thereof.

[0024] On the other hand, unless otherwise defined, all terms used herein should be considered to have the same meaning as that commonly understood by a person of ordinary skill in the art to which the present invention pertains. Therefore, unless expressly defined herein, a particular term should not be construed in an overly ideal or formal sense.

[0025] In addition, in this specification, the terms "about," "substantially," and the like are used to mean a numerical value or a close approximation of a numerical value when the manufacturing and material tolerances inherent in the recited meaning are given, and are used to prevent unscrupulous infringers from unfairly taking advantage of the disclosure in which precise or absolute numerical values ​​are recited to facilitate understanding of the present invention.

[0026] In addition, in this specification, the term "system" refers to a collection of components including devices, tools, means, etc. that are organized and interact in an orderly manner to perform a required function.

[0027] In addition, in this specification, the term "part" is used to refer to a component that performs one or more functions or operations, and such a component may be realized by hardware or software, or by a combination of hardware and software.

[0028] In addition, in this specification, the term "predetermined" means that it is preset by an external user (such as an administrator).

[0029] Hereinafter, a process control system based on an artificial intelligence model according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings. The drawings disclosed below are provided as examples to fully convey the concept of the present invention to those skilled in the art. Therefore, the present invention is not limited to the drawings presented below, and may be embodied in other forms. In addition, the same reference numerals refer to the same elements throughout the specification.

[0030] 1 is a flow chart of a method for generating an artificial intelligence model for process control according to an embodiment of the present invention. In the following, the method for generating an artificial intelligence model for process control will be described in detail with reference to FIG. 1.

[0031] A method for generating an AI model for process control according to an embodiment of the present invention may include a step (S100) of storing a plurality of reactor process data including actual reactor input conditions, actual reactor operating conditions, and actual physical property values ​​of a product by the reactor or a calculation result by simulation, a step (S200) of removing outliers from the stored reactor process data to generate learning data, and a step (S300) of generating an AI model including an AI algorithm that learns the generated learning data and derives optimal reactor input conditions to satisfy the reactor operating conditions and the physical property values ​​of a product by the reactor.

[0032] The reactor charging conditions may include: (a)

[0033] (a) One or more of the composition, temperature, flow rate, and pressure of the raw material fed into the reactor, or a combination of these.

[0034] When a catalyst is added to the reactor, the adjusted reactor input conditions may further include the following (b):

[0035] (b) one or more of the composition, temperature, flow rate, and pressure of the catalyst fed into the reactor, or a combination thereof;

[0036] In the reactor feeding condition (a), the composition of the raw material may include the type and content of compounds constituting the raw material, and the content may be an absolute value such as weight or volume, or a relative value such as weight ratio or volume ratio. The temperature of the raw material may mean the temperature of the raw material when fed into the reactor, the flow rate of the raw material may mean the flow rate of the raw material among the total feeds fed into the reactor, and the pressure of the raw material may mean the pressure of the raw material among the total feeds fed into the reactor.

[0037] In terms of more precise control of the reactor, the reactor input condition (a) can be, for example, the feed composition and temperature, or the feed composition and flow rate, or the feed composition and pressure, or the feed temperature and flow rate, or the feed temperature and pressure, or the feed flow rate and pressure; for example, the feed composition, temperature and flow rate, or the feed composition, temperature and pressure, or the feed temperature, flow rate and pressure; for example, the feed composition, temperature, flow rate and pressure.

[0038] In the reactor feeding condition (b), the catalyst composition may include the type and content of the compound constituting the catalyst, and the content may be an absolute value such as weight, volume, etc., or a relative value such as weight ratio, volume ratio, etc. The catalyst temperature may mean the temperature of the catalyst when fed into the reactor, the catalyst flow rate may mean the flow rate of the catalyst among the total feed fed into the reactor, and the catalyst partial pressure may mean the pressure of the catalyst among the total feed fed into the reactor.

[0039] In terms of operating the reactor more precisely, the reactor input condition (b) can be, for example, the catalyst composition and temperature, or the catalyst composition and flow rate, or the catalyst composition and pressure, or the catalyst temperature and flow rate, or the catalyst temperature and pressure, or the catalyst flow rate and pressure, for example, the catalyst composition, temperature and flow rate, or the catalyst composition, temperature and pressure, or the catalyst temperature, flow rate and pressure, for example, the catalyst composition, temperature, flow rate and pressure.

[0040] The reactor charging conditions mentioned below are the same as those described above, and therefore will not be described for the sake of convenience.

[0041] The artificial intelligence model generated by the step of generating the artificial intelligence model (S300) may include an artificial intelligence algorithm A1 for deriving optimal reactor input conditions for satisfying the reactor operating conditions and the physical properties of the product by the reactor, and is not particularly limited in configuration. For example, the artificial intelligence model may include an algorithm for predicting various results of the reactor process.

[0042] According to one embodiment, the artificial intelligence model may include an artificial intelligence algorithm A2 that derives predicted operating conditions of the reactor or predicted physical properties of the product from the reactor based on the input conditions of the reactor.

[0043] According to one embodiment of the present invention, the predicted operating conditions of the reactor or the predicted physical properties of the product from the reactor derived by the artificial intelligence algorithm A2 can be provided to the artificial intelligence algorithm A1, and optimal reactor input conditions can be derived based on the predicted operating conditions of the reactor or the predicted physical properties of the reactor.

[0044] According to one embodiment of the present invention, the artificial intelligence algorithm A1 may be characterized in that, when the physical property value of the product is changed during the operation of the reactor, the dynamic characteristics of the reactor input conditions to reach the changed physical property value from the point at which the physical property value is changed are analyzed, and new optimal reactor input conditions are derived.

[0045] The analysis of dynamic characteristics is performed to predict physical and chemical changes in a process including a reactor that may occur in the process in which a physical property value is changed over time, and an optimal reactor input condition for each process in the process to reach the changed physical property value can be derived. The criteria for dividing the process units by process can be preset by an external user (e.g., an administrator). For example, the process units can be divided based on time, or based on the total amount of products produced. However, it should be noted that the above examples are merely examples for ease of understanding, and the criteria are not limited to the above examples. Based on the above optimal reactor input conditions for each reaction process, the state at the time of the change can be more quickly and stably reached to the changed physical property value.

[0046] The step of generating the artificial intelligence model (S300) may be characterized by generating the artificial intelligence model by one or more of linear regression, logistic regression, decision tree, random forest, support vector machine, gradient boosting, convolution neural network, recurrent neural network, long-short term memory, attention model, transformer, generative adversarial network, and reinforcement learning, or a combination thereof.

[0047] FIG. 2 of the accompanying drawings is a block diagram showing a process control system based on an artificial intelligence model.

[0048] As shown in FIG. 2, a process control system based on an artificial intelligence model according to an embodiment of the present invention may include an artificial intelligence control model unit 100, an input unit 200, and an output unit 300, and may control the input of a reactor based on the reactor input conditions obtained by the artificial intelligence control model unit 100, the input unit 200, and the output unit 300.

[0049] Hereinafter, each component of the process control system based on an artificial intelligence model according to an embodiment of the present invention will be described in detail.

[0050] The artificial intelligence control model unit 100 may include a data storage unit 110, a data correction unit 120, and a data derivation unit .

[0051] The data storage unit 110 may store a plurality of preset process data of the reactors. The plurality of preset process data of the reactors corresponds to raw data for learning an artificial intelligence model as a plurality of experimental data including actual input conditions of the reactor, actual operating conditions of the reactor, and actual physical property values ​​of the product of the reactor collected from a laboratory, a pilot plant, a commercial plant, etc., or calculation results by simulation. The operating conditions of the reactor may mean, for example, but are not limited to, the temperature and pressure in the reactor. The calculation results by simulation are results obtained by simulating the reactor process, and the calculation results may include all data of the process. The calculation results may include, for example, input conditions of the reactor, operating conditions of the reactor, and physical property values ​​of the product of the reactor derived by simulation.

[0052] The data correction unit 120 may remove outliers from the stored experimental data to generate learning data. The learning data is used as data for training an artificial intelligence model. According to another embodiment, the data correction unit 120 may remove outliers and then standardize a scale to generate learning data.

[0053] The data derivation unit 130 can learn from the generated learning data to derive optimal reactor input conditions for satisfying the reactor operating conditions and the physical property values ​​of the product by the reactor. The data derivation unit 130 can be configured as one or more of a physical property prediction AI model, a control optimization AI model, and other AI models, or a combination of these, or an integrated AI model. Here, the derived optimal reactor input conditions can be provided to the output unit 300 by the data derivation unit 130 or a separate data providing unit.

[0054] The optimal reactor input conditions can be derived by an external user (such as a manager) setting in advance a goal to be achieved through the process control system based on an artificial intelligence model. The goal to be achieved through the process control system based on an artificial intelligence model can be, for example, one or more of the following goals or combinations: reduction in input raw materials, reduction in input catalyst, reduction in grade change time, reduction in off-spec, increase in product yield, increase in product production volume, reduction in utility usage, reduction in operating costs, and reduction in production costs, but is not limited thereto.

[0055] According to an example, the data derivation unit 130 may derive predicted operating conditions of the reactor or predicted physical property values ​​of the product of the reactor based on the optimal reactor input conditions as well as the data provided from the input unit 200. According to an example, the derived predicted operating conditions or predicted physical property values ​​may be displayed via a commonly used display device such as a monitor so that an external user (e.g., an administrator) can check them in real time.

[0056] The artificial intelligence control model unit 100 can receive data including target operating conditions of the reactor and target physical property values ​​of the product by the reactor from the input unit 200. According to one example, the data can be all process data of the reactor to be controlled, including reactor input conditions, reactor operating conditions, and physical property values ​​of the product. The target operating conditions of the reactor and the target physical property values ​​of the product by the reactor can be preset by an external user (e.g., an administrator).

[0057] The artificial intelligence control model unit 100 can derive optimal reactor input conditions for satisfying the target reactor operating conditions and the target physical property values ​​of the product by the reactor provided via the data derivation unit 130, and the optimal reactor input conditions derived at this time can be provided to the output unit 300 via the data derivation unit 130 or a separate data providing unit.

[0058] According to an example, when the target property value provided from the input unit 200 is changed during the operation of the reactor, the data derivation unit 130 can analyze the dynamic characteristics of the reactor input conditions to reach the changed target property value from the time when the target property value is changed, and derive new optimal reactor input conditions.

[0059] The analysis of dynamic characteristics is performed to predict not only the reactor input conditions (a) or (b) for reaching the target property value, but also physical and chemical changes in the reactor that may occur in the process of the property value being changed over time, thereby making it possible to more quickly and stably reach the target property value from the state at the time of the change.

[0060] FIG. 3 is a diagram showing a re-learning configuration of the artificial intelligence control model unit 100 according to an embodiment of the present invention.

[0061] According to an embodiment of the present invention, the AI ​​control model unit 100 may further include a data analysis unit 140 and a data re-learning unit 150 to re-learn the AI ​​model.

[0062] According to an example, the data derivation unit 130 may derive predicted operating conditions of the reactor and predicted physical property values ​​of the product from the reactor based on the reactor input conditions provided from the input unit 200. The derived predicted operating conditions of the reactor and predicted physical property values ​​of the product may be provided to the data analysis unit 140 via the data derivation unit 130 or another data providing unit.

[0063] According to an example, the data analysis unit 140 compares the actual operating conditions of the reactor and the actual physical property values ​​of the product (1) provided from the data storage unit 110 or the data correction unit 120 with the predicted operating conditions of the reactor and the predicted physical property values ​​of the product (2) derived by the data derivation unit 130. The comparison result can be provided to the data re-learning unit 150 via the data analysis unit 140 or another data providing unit.

[0064] The actual operating conditions of the reactor and the actual physical property values ​​of the product (1) can be provided from the data storage unit 110 or the data correction unit 120. When provided from the data correction unit 120, the actual operating conditions of the reactor and the actual physical property values ​​of the product can be obtained by removing outliers of the process data of the reactor stored in the data storage unit 110.

[0065] According to an example, the data re-learning unit 150 can cause the data derivation unit 130 to re-learn when the comparison result provided by the data analysis unit 140 satisfies the following condition (1) or condition (2).

[0066] (1) When the error rate between the actual operating conditions of the reactor and the arbitrary operating conditions of the reactor exceeds the allowable value. (2) If the error rate between the actual physical property value of the product and the arbitrary physical property value of the product produced by the reactor exceeds the allowable value.

[0067] Here, the allowable error rate can be set in advance by an external user (such as an administrator) in consideration of the type of process, the type of reactants, the type of controller, etc. The allowable error rate can be, for example, 30% or less, 20% or less, 15% or less, 10% or less, or 5% or less.

[0068] According to the present invention, the above-mentioned re-learning process can further improve the accuracy of the AI ​​control model unit 100. As a result, the reliability of the reactor operating conditions and the optimal reactor input conditions for satisfying the physical properties of the product by the reactor derived from the AI ​​control model unit 100 is further improved.

[0069] According to an embodiment of the present invention, the artificial intelligence control model unit 100 may be trained using one or more of the following methods or an ensemble: linear regression, logistic regression, decision tree, random forest, support vector machine, gradient boosting, convolution neural network, recurrent neural network, long-short term memory, attention model, transformer, generative adversarial network, and reinforcement learning.

[0070] The input unit 200 can obtain data including process operating conditions including reactor input conditions, reactor operating conditions, and product property values, as well as target operating conditions of the reactor and target property values ​​of the product by the reactor, and provide the data to the artificial intelligence control model unit 100. According to an example, the data can include all process data of the reactor to be controlled, including the above-mentioned contents. According to an example, the data can be collected in real time from a reactor in a factory. According to an example, the target operating conditions of the reactor and the target property values ​​of the product by the reactor obtained by the input unit 200 can be preset by an external user (e.g., an administrator).

[0071] The output unit 300 can receive optimal reactor input conditions for satisfying the target operating conditions of the reactor and the target physical property values ​​of the product from the artificial intelligence control model unit 100, and control the input to the reactor under the optimal reactor input conditions.

[0072] According to one embodiment of the present invention, a reactor may be provided that includes the artificial intelligence based process control system described above.

[0073] The reactor can be, for example, one of a tubular reactor, a column reactor, a stirred tank reactor, a fluidized bed reactor, and a loop reactor.

[0074] According to an embodiment of the present invention, the reactor is composed of a plurality of reactors, which may be the same or different from each other, and each of the reactors may be independently one of a tubular reactor, a column reactor, a stirred tank reactor, a fluidized bed reactor, and a loop reactor. When the reactor is composed of a plurality of reactors, the manner of connection between them is not particularly limited. For example, the reactors may be connected in parallel, but are not limited thereto.

[0075] According to one embodiment of the present invention, the present invention can easily derive the target reactor operating conditions and the optimal reactor input conditions for achieving the target physical properties of the product produced by the reactor using an artificial intelligence model.

[0076] According to one embodiment of the present invention, the reactor input can be controlled under the optimal reactor input conditions, thereby reducing the unit cost, time, etc. required for product production.

[0077] According to one embodiment of the present invention, by controlling the process using an artificial intelligence model, the intervention of an external user (e.g., an administrator) can be minimized, trial and error can be reduced, and a reliable reactor product can be obtained.

[0078] The process control system based on artificial intelligence of the present invention can be applied to various petrochemical processes, oil refining processes, chemical processes, etc., in particular, to polymer processes, more particularly, to polyolefin processes using olefin monomers having carbon numbers of C2 to C12, and even more particularly, to polyethylene processes using a solution polymerization method (solution process), and has industrial applicability.

[0079] Although exemplary embodiments of the present invention have been described above, the present invention is not limited thereto, and a person having ordinary knowledge in the art can understand that various changes and modifications are possible without departing from the concept and scope of the claims described below.

Claims

1. an artificial intelligence control model unit (100) including a data storage unit (110) for storing preset process data of a plurality of reactors, a data correction unit (120) for generating learning data from the stored reactor process data by removing outliers, and a data derivation unit (130) for learning from the generated learning data to derive optimal reactor input conditions for satisfying reactor operating conditions and physical property values ​​of products from the reactor; An input unit (200) for obtaining data including target operating conditions of the reactor and target physical property values ​​of the product from the reactor and providing the data to the artificial intelligence control model unit (100); and an output unit (300) that receives optimal reactor input conditions for satisfying the target operating conditions of the reactor and the target physical property values ​​of the product from the artificial intelligence control model unit (100) and controls the input of the reactor under the optimal reactor input conditions, A process control system based on an artificial intelligence model, characterized in that the reactor charging conditions include the following (a): (a) one or more of the composition, temperature, flow rate, and pressure of the raw material fed into the reactor, or a combination thereof

2. 2. The process control system based on an artificial intelligence model of claim 1, wherein the reactor input conditions further include: (b) one or more of the composition, temperature, flow rate, and pressure of the catalyst introduced into the reactor, or a combination thereof;

3. The preset process data of the plurality of reactors includes:

2. The process control system based on the artificial intelligence model of claim 1, including: actual input conditions of the reactor, actual operating conditions of the reactor, and actual physical property values ​​of the product from the reactor; or calculation results from simulation.

4. The data deriving unit (130) derives predicted operating conditions of the reactor and predicted physical property values ​​of the product from the reactor based on the reactor input conditions provided from the input unit (200); The artificial intelligence control model unit (100) includes a data analysis unit (140) that compares the actual operating conditions of the reactor and the actual physical property values ​​of the product in the data storage unit (110) or the data correction unit (120) with the predicted operating conditions of the reactor and the predicted physical property values ​​of the product derived in the data derivation unit (130); 4. The process control system based on the artificial intelligence model of claim 3, further comprising: a data re-learning unit (150) that re-learns the data derivation unit (130) when the comparison result provided by the data analysis unit (140) satisfies the following condition (1) or condition (2): (1) When the error rate between the actual operating conditions of the reactor and the predicted operating conditions of the reactor exceeds a preset allowable value. (2) When the error rate between the actual physical property value of the product and the predicted physical property value of the product exceeds a predetermined allowable value.

5. The data derivation unit (130) 2. The process control system according to claim 1, wherein, when the target property value provided from the input unit (200) is changed during the operation of the reactor, the dynamic characteristics of the reactor input conditions for reaching the changed target property value from the time when the target property value is changed are analyzed, and new optimal reactor input conditions are derived.

6. The artificial intelligence control model unit (100) uses linear regression, logistic regression, and Logistic regression, decision tree, random forest, support vector machine, gradient boosting, convolution neural network, recurrent neural network , long-short term memory, attention model, transformer, generative adversarial network, reinforcement learning, or combinations of these.

10. A process control system based on the artificial intelligence model of claim 1, which is trained by ensemble.

7. A reactor comprising a process control system based on an artificial intelligence model according to any one of claims 1 to 6.

8. The reactor comprises:

8. The reactor of claim 7, which is one of a tubular reactor, a column reactor, a stirred tank reactor, a fluidized bed reactor, and a loop reactor.

9. 8. The reactor according to claim 7, wherein the reactor is composed of a plurality of reactors, which may be identical to or different from each other, and each reactor is independently one of a tubular reactor, a column reactor, a stirred tank reactor, a fluidized bed reactor, and a loop reactor.

10. Storing process data for a plurality of reactors, including actual reactor input conditions, actual reactor operating conditions, and actual physical property values ​​of reactor products; or calculation results from simulations; removing outliers from the stored reactor process data to generate training data; and generating an artificial intelligence model including an artificial intelligence algorithm that learns the generated learning data and derives optimal reactor input conditions that satisfy the reactor operating conditions and the physical properties of the product produced by the reactor, A method for generating an artificial intelligence model for process control, characterized in that the reactor input conditions include the following (a): (a) one or more of the composition, flow rate, and partial pressure of the raw material fed into the reactor, or a combination thereof;

11. 11. The method of claim 10, wherein the reactor input conditions further include the following (b): (b) one or more of the composition, temperature, flow rate, and pressure of the catalyst introduced into the reactor, or a combination thereof;

12. In the artificial intelligence algorithm, 11. The method of claim 10, wherein, when a physical property value of the product is changed during reactor operation, dynamic characteristics of reactor input conditions for reaching the changed physical property value from the time when the physical property value is changed are analyzed, and new optimal reactor input conditions are derived.

13. The step of generating the artificial intelligence model includes: Linear regression, logistic regression, decision tree, random forest, support vector machine, gradient boosting , convolutional neural network, recurrent neural network, long-short term memory 11. The method of generating an artificial intelligence model for process control according to claim 10, wherein the artificial intelligence model is generated by one or more of the following methods: a neural network (NN), an attention model, a transformer, a generative adversarial network, and reinforcement learning, or a combination thereof.

Images