Learning apparatus, inference apparatus, learning method, inference method, learning program, and inference program

Abstract

To implement an operation support system of a batch plant using a machine learning model designed for imitation learning.SOLUTION: A processing apparatus 210 includes: a collection unit 1331 which collects histories of production processes, which are combinations of an explanatory variable representing a status in a production process of a product and an objective variable representing operation of a device in the production process; a global model learning unit 1313 which optimizes a parameter of a multi-process model configured to output an objective variable from an explanatory variable, using the histories of the multiple production processes as learning data; and a differential model learning unit 2133 which trains, for each production process, a process-specific differential model configured to output a difference from an explanatory variable, using a difference between an objective variable in the learning data and the objective variable output by the multi-process model, as learning data.SELECTED DRAWING: Figure 8

Description

[Technical Field]

[0001] The present invention relates to a learning device, an inference device, a learning method, an inference method, a learning program, and an inference program. [Background technology]

[0002] BACKGROUND ART A technique known as imitation learning is known in the past, in which a machine learning model (hereinafter, referred to as a model) learns human behavior and then uses the model to teach behavior to humans, robots, or the like.

[0003] Here, an operation support system for continuous plants using a model that performs imitation learning has been proposed. For example, in a continuous plant such as a chemical plant, imitation learning is performed by having a model learn the operations actually performed by an operator (plant operation imitation learning). Note that a continuous plant, for example, a heat treatment furnace, is a system in which the heat treatment furnace is operated at all times, and products are fed into the heat treatment furnace to perform heat treatment continuously.

[0004] Using a model that performs this type of imitation learning, it is possible not only to assist plant operation with recommended values, but also to autopilot operations. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Patent No. 7335414 Summary of the Invention [Problem to be solved by the invention]

[0006] In contrast to continuous plants, there is a system called a batch plant, which is a plant in which each production process, such as material input, heating, and cooling, is carried out in sequence in one furnace.

[0007] There are some problems when applying plant operation imitation learning to build a model for a batch plant.

[0008] In batch plants, the operating method differs depending on the production process. For example, the required operations differ depending on the production process, such as the cooling process and the production process, so it is difficult to perform consistent control using a single model.

[0009] Furthermore, because batch plants have multiple production processes, creating separate models for each production process in a batch plant would result in insufficient data corresponding to each production process, resulting in inaccuracy of the model.

[0010] The present invention has been made in consideration of the above, and aims to provide a learning device, an inference device, a learning method, an inference method, a learning program, and an inference program that realize an operation support system for a batch plant using a machine learning model that performs imitation learning. [Means for solving the problem]

[0011] In order to solve the above-mentioned problems and achieve the object, the learning device of the present invention is characterized by having: a collection unit that collects a history of each production process, which is a combination of explanatory variables that represent a situation in the production process of a manufactured product and a target variable that represents the operation of equipment in the production process; a first learning unit that uses the histories of a plurality of the production processes as learning data and optimizes parameters of a multi-process model that outputs the target variable from the explanatory variables; and a second learning unit that, for each of the production processes, uses the difference between the target variable in the learning data and the target variable output by the multi-process model as learning data and learns a process-specific difference model that outputs the difference from the explanatory variables.

[0012] In order to solve the above-mentioned problems and achieve the object, an inference device according to the present invention includes a storage unit that stores: a multi-process model that is trained using, as training data, histories of a plurality of production processes among histories that are combinations of explanatory variables that represent a state of a product production process and objective variables that represent operations of equipment in the production process, and that outputs the objective variable from the explanatory variables; and a plurality of process-specific difference models that are provided for each of the production processes, and that, when the explanatory variables are input, output the difference between the objective variable corresponding to the input explanatory variable and the objective variable output by the multi-process model; a discrimination unit that discriminates the production process corresponding to the explanatory variables acquired by the acquisition unit; a selection unit that selects, from the plurality of process-specific differential models, the process-specific differential model that corresponds to the production process discriminated by the discrimination unit; and an inference unit that adds together a first output of the multi-process model obtained by inputting the explanatory variables of the inference target into the multi-process model and a second output of the process-specific differential model obtained by inputting the explanatory variables of the inference target into the process-specific differential model selected by the second selection unit, and infers the sum as the objective variable of the explanatory variables of the inference target. [Effects of the Invention]

[0013] According to the present invention, an operation support system for a batch plant is realized using a machine learning model that performs imitation learning. [Brief explanation of the drawings]

[0014] [Figure 1] FIG. 1 is a diagram illustrating a plant operation system. [Figure 2] FIG. 2 is a diagram illustrating an outline of model learning according to the first embodiment. [Figure 3] FIG. 3 is a diagram illustrating an example of the configuration of the processing device according to the first embodiment. [Figure 4-1] FIG. 4-1 is a diagram showing an example of data registered in the history database (DB) shown in FIG. [Figure 4-2] FIG. 4-2 is a diagram illustrating an example of data registered in the history DB illustrated in FIG. [Figure 4-3] FIG. 4-3 is a diagram illustrating an example of data registered in the history DB illustrated in FIG. [Figure 4-4] FIG. 4-4 is a diagram illustrating an example of data registered in the history DB illustrated in FIG. [Figure 5] FIG. 5 is a flowchart of the learning process according to the first embodiment. [Figure 6] FIG. 6 is a flowchart of the inference process according to the first embodiment. [Figure 7] FIG. 7 is a diagram illustrating an outline of model learning in the second embodiment. [Figure 8] FIG. 8 is a diagram illustrating an example of the configuration of a processing device according to the second embodiment. [Figure 9] FIG. 9 is a flowchart of the learning process according to the second embodiment. [Figure 10] FIG. 10 is a flowchart of the inference process according to the second embodiment. [Figure 11] FIG. 11 is a diagram illustrating an example of the configuration of a processing device according to the third embodiment. [Figure 12] FIG. 12 is a flowchart of the learning process according to the third embodiment. [Figure 13] FIG. 13 is a flowchart of the inference process according to the third embodiment. [Figure 14] FIG. 14 is a diagram illustrating an example of a computer that implements a processing device by executing a program. DETAILED DESCRIPTION OF THE INVENTION

[0015] Hereinafter, embodiments of a learning device, an inference device, a learning method, an inference program, and an inference program according to the present application will be described in detail with reference to the accompanying drawings. Note that the present invention is not limited to the embodiments described below.

[0016] [Embodiment 1] [Configuration of the First Embodiment] First, the plant operation system will be described with reference to Fig. 1. Fig. 1 is a diagram for explaining the plant operation system.

[0017] The plant operation system 1 is a system for managing and controlling a product production process (hereinafter referred to as a production process) in a plant. The plant includes a chemical plant for producing chemical products. In this embodiment, a plant operation system applied to a batch plant will be described. A batch plant is a plant in which, for example, each production process, such as material input, heating, and cooling, is performed in sequence for one piece of equipment (e.g., a furnace).

[0018] As shown in FIG. 1, the plant operation system 1 includes a processing device 10, a terminal device 20, and a plant system 30.

[0019] The processing device 10 performs processing related to a model (machine learning model) for performing imitation learning. The processing device 10 can function as a learning device and an inference device.

[0020] The processing device 10 and the plant system 30 are connected to each other via a network so that they can communicate data with each other. For example, the network is the Internet and an intranet. The processing device 10 uses the model to present recommended operation values ​​to an operator of the plant system 30, thereby supporting plant operation. Furthermore, when the processing device 10 satisfies predetermined autopilot conditions set by the operator or the like and receives an autopilot command from the operator, it uses the model to perform autopilot control of the plant system 30.

[0021] The plant system 30 may include equipment used in the production process and a distributed control system (DCS). For example, the equipment is a furnace or a reactor, and each production process, such as material input, heating, and cooling, is performed in sequence for one furnace.

[0022] The terminal device 20 is an information processing device such as a personal computer, a tablet terminal, or a smartphone, or a dedicated terminal for operating equipment in the plant.

[0023] The operator is a user who operates the devices included in the plant system 30 via the terminal device 20. The operator also instructs the processing device 10 to start or stop the autopilot. The models used in the processing device 10 are managed appropriately by a system administrator or the like.

[0024] The processing of each device in the plant operation system 1 will be described with reference to FIG.

[0025] The terminal device 20 operates the equipment of the plant system 30 in response to an operation by an operator (step S1). For example, the terminal device 20 operates to set the temperature inside one piece of equipment (e.g., a furnace) of a batch plant, the pressure inside the equipment, the target value of the production amount in the production process, the amount of raw materials to be input into the equipment, etc.

[0026] The plant system 30 operates in accordance with the operation from the terminal device 20 (step S2). Then, the plant system 30 transmits the operation history to the processing device 10 (step S3).

[0027] For example, the history includes sensor values ​​of various sensors installed in the furnace of the plant system 30 and setting values ​​set by operation from the terminal device 20. The history may also be time-series data in which a time (time stamp) is assigned to each record. The history may also be tagged with process information indicating multiple production processes that were sequentially executed on the equipment, for example.

[0028] The processing device 10 registers the history of each production process collected from the plant system 30 in a history database (DB) for each production process (step S3).

[0029] Next, the processing device 10 acquires learning data to be used for learning (for example, machine learning) of the model (machine learning model) from the history, learns the model, and performs inference using the model (step S4).

[0030] The processing device 10 uses predetermined autopilot conditions to determine, for example, whether or not autopilot is possible by the processing device 10. Details of each process of the processing device 10 will be described later.

[0031] Furthermore, the processing device 10 presents a guidance screen 21 showing the inference result and the autopilot implementation determination result to the operator's terminal device 20 (step S6). When the autopilot conditions are satisfied, the processing device 10 displays on the guidance screen 21 that it is now possible to start the autopilot. When the terminal device 20 instructs the processing device 10 to start the autopilot, the processing device 10 performs autopilot control of the plant system 30 using the model (step S5).

[0032] Furthermore, if the autopilot conditions are not satisfied while the autopilot is in operation, the processing device 10 displays an instruction to stop the autopilot on the guidance screen 21. When an instruction to stop the autopilot is received from the terminal device 20, the processing device 10 stops the autopilot control of the plant system 30. Then, the terminal device 20 operates the equipment of the plant system 30 in response to an operation by the operator (step S1).

[0033] The model learns the operations of the operators through imitation learning, so other operators can imitate the operations by following the operations obtained as inference results from the model.

[0034] [Model Training] In a batch plant to which the first embodiment is applied, the operation method differs depending on the production process. In this batch plant, the required operations, such as a cooling process and a heating process, differ depending on the production process. In this embodiment, the following model learning is performed to enable consistent control of a batch plant with a single model. Figure 2 is a diagram illustrating an overview of model learning in the first embodiment.

[0035] As shown in FIG. 2, the processing device 10 first generates a global model (multiple process model) corresponding to all production processes of a batch plant ((1) in FIG. 2). At this time, the processing device 10 performs learning to optimize the parameters of the global model so that an operation value Y is output when an operation execution time t and a process variable X, which is an explanatory variable, are input using all history n (1≦n≦N) of all production processes as learning data (arrow Y1). Note that as learning data for the model, the operation execution time t n , process variable X n , operation value Y n is used.

[0036] The process variable X has items such as the temperature at one or more locations of an equipment (e.g., a furnace), the flow rate at one or more locations, the pressure at one or more locations, and the concentration of gases generated in the production process. The values ​​of these items are detected by sensors installed in the equipment. The operation execution time t is a timestamp indicating the date and time when the temperature, pressure, flow rate, and gas concentration were acquired.

[0037] The objective variable is the operation to be performed, and is, for example, a set value set by an operation from the terminal device 20. The operations are various operations performed in the production process, such as material input operations, heating operations, cooling operations, valve operations, etc. The set value may be a normalized value of an actually set value.

[0038] Next, the processing device 10 performs fine tuning (additional learning) on ​​the global model using the histories allocated to each production process. For each production process, the processing device 10 sets the parameters of the global model to initial values ​​(arrow Y2), and performs additional learning on a process-specific model that outputs a target variable from an explanatory variable, using only the history of the target production process as learning data.

[0039] For example, when constructing a production process D model, the processing device 10 sets the parameters of the global model as initial values, and uses only the history of production process D as learning data (arrow Y3D) to perform additional learning of the process D model corresponding to production process D, thereby constructing the process D model ((2) in FIG. 2). Similarly, when constructing process-specific models for production processes A to C, the processing device 10 sets the parameters of the global model as initial values, and uses only the history of the corresponding production processes A, B, and C as learning data to perform additional learning of the process A model, process B model, and process C model. This learning method leads to high-quality local optima, and is therefore considered to be particularly effective as a learning method using gradient descent.

[0040] During inference, the processing device 10 uses different process-specific models depending on the production process of the explanatory variable to be inferred.

[0041] [Processing equipment] Next, the processing device 10 will be described in detail. Fig. 3 is a diagram showing an example of the configuration of the processing device 10 according to embodiment 1. As shown in Fig. 3, the processing device 10 includes a communication unit 11, a storage unit 12, and a control unit 13.

[0042] The communication unit 11 performs data communication with other devices via a network. For example, the communication unit 11 is a network interface card (NIC).

[0043] The storage unit 12 is a storage device such as a hard disk drive (HDD), a solid state drive (SSD), an optical disk, etc. Note that the storage unit 12 may also be a data-rewritable semiconductor memory such as a random access memory (RAM), a flash memory, or a non-volatile static random access memory (NVSRAM).

[0044] The storage unit 12 stores an OS (Operating System) and various programs executed by the processing device 10. The storage unit 12 includes a history DB 121 and model information 122.

[0045] The history DB 121 is information including history provided from the plant system 30. For example, in the history 1210 of the history DB 121, records are accumulated every minute, divided into production processes.

[0046] In the history 1210, for example, a process A history 121A, a process B history 121B, a process C history 121C, and a process D history 121D are registered. In the history 1210, the history is divided into production processes and registered based on tags or process information indicating the production processes assigned to the history. Alternatively, the processing device 10 may compare the production process schedule of the applicable batch plant with the timestamp of the history, determine the production process that was performed at the stamped date and time, tag the determined production process to the history, and register it in the history DB 121. Alternatively, the processing device 10 may determine which process this history belongs to based on explanatory variables (sensor values) included in the history.

[0047] 4-1 to 4-4 are diagrams showing an example of data registered in the history DB 121 shown in Fig. 3. As shown in Fig. 4-1 to 4-4, the history DB 121 includes a list of explanatory variables such as time, situation, operation, etc., and a set value (operation) which is a target variable. Fig. 4-1 illustrates a process A history 121-A, Fig. 4-2 illustrates a process B history 121-B, Fig. 4-3 illustrates a process C history 121-C, and Fig. 4-4 illustrates a process D history 121-D.

[0048] The time indicates the time when the operation is performed. The explanatory variable list includes, for example, a first temperature, a second temperature, and a first flow rate. The explanatory variable list may also include pressure at one or more locations, the concentration of gas generated in the production process, etc.

[0049] The first temperature, the second temperature, the first pressure, the second pressure, and the first flow rate are sensor values ​​of sensors installed at various locations in the plant system 30, respectively.

[0050] The first temperature, the second temperature, the first pressure, the second pressure, and the first flow rate are explanatory variables of the model, and are examples of explanatory variables that represent the status of the production process of the product.

[0051] The time is a timestamp indicating the date and time when the first temperature, the second temperature, the first pressure, the second pressure, the flow rate, and the gas concentration were acquired.

[0052] The operation is, for example, a setting value set by an operation from the terminal device 20. The setting value may be a normalized value of an actually set value. The setting value corresponds to the objective variable of the model.

[0053] The set value is an objective variable of the model, and is an example of an objective variable that represents the operation of equipment in the production process.

[0054] The model information 122 is information such as parameters for constructing each process model. For example, if the process model is a neural network, the model information 122 is the weight and bias of each layer.

[0055] The model information 122 stores model information of a process-specific model 1220. The process-specific model 1220 includes, for example, parameters of a process A model 122A used when inferring data for production process A, a process B model 122B used when inferring data for production process B, a process C model 122C used when inferring data for production process C, and a process D model 122D used when inferring data for production process D.

[0056] The control unit 13 controls the entire processing device 10. The control unit 13 is, for example, an electronic circuit such as a CPU (Central Processing Unit), an MPU (Micro Processing Unit), or a GPU (Graphics Processing Unit), or an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

[0057] The control unit 13 also has an internal memory for storing programs that define various processing procedures and control data, and executes each process using the internal memory. The control unit 13 also functions as various processing units by running various programs.

[0058] For example, the control unit 13 has a learning unit 131 (learning device) and an inference unit 132 (inference device). The learning unit 131 executes learning of a global model and a process-specific model. The inference unit 132 uses a process-specific model of an explanatory variable of an inference target to infer a target variable from the explanatory variable of the inference target, and applies the target variable to the plant system 30.

[0059] The learning unit 131 includes a collection unit 1311, a learning data reading unit 1312, a global model learning unit 1313 (first learning unit), and an additional learning unit 1314 (third learning unit). For example, the additional learning unit 1314 employs sequential learning using the JIT method.

[0060] The collection unit 1311 collects the history of each production process in the plant system 30. The history is a combination of explanatory variables that represent the status of the product in the production process and objective variables that represent the operation of equipment in the production process. The collection unit 1311 registers each history in the history DB 121 for each production process.

[0061] The learning data reading unit 1312 reads out the history of all production processes from the history DB 121 as learning data, and outputs it to the global model learning unit 1313 .

[0062] The global model learning unit 1313 uses the history of all production processes as learning data to optimize the parameters of the global model that outputs the objective variable from the explanatory variable.

[0063] The additional learning unit 1314 performs additional learning of a process-specific model that outputs a target variable from an explanatory variable, for each production process, using the parameters of the global model as initial values ​​and only the history of the target production process as learning data. The additional learning unit 1314 stores the model parameters of each additionally learned process-specific model (e.g., process A model 122A, process B model 122B, process C model 122C, process D model 122D) in model information 122. After each process-specific model is constructed, the model parameters of the global model are deleted.

[0064] The inference unit 132 includes a data acquisition unit 1321, a process discrimination unit 1322 (discrimination unit), a first model selection unit 1323 (process-specific model selection unit), a first inference unit 1324 (process-specific inference unit), and a system application unit 1325.

[0065] The data acquisition unit 1321 acquires explanatory variables of the inference target. The explanatory variables of the inference target are explanatory variables (corresponding to each sensor value) at a predetermined time.

[0066] The process determination unit 1322 determines a production process corresponding to the explanatory variable acquired by the data acquisition unit 1321. The process determination unit 1322 determines the production process, for example, by referring to a tag indicating the production process attached to the explanatory variable. The process determination unit 1322 may determine the production process corresponding to the explanatory variable by comparing the production process schedule of the batch plant to which the explanatory variable is applied with the timestamp of the explanatory variable and determining the production process that was performed at the stamped date and time.

[0067] The first model selection unit 1323 selects a production process model corresponding to the production process identified by the process identification unit 1322 from among the multiple process models 1220. For example, if the production process identified by the process identification unit 1322 is production process D, the first model selection unit 1323 selects the process D model 122D.

[0068] The first inference unit 1324 infers a response variable corresponding to an explanatory variable of an inference target, using the process-specific model selected by the first model selection unit 1323. For example, when the model selected by the first inference unit 1324 is the process D model 122D, the explanatory variable of the inference target is input to this process D model 122D, and the inferred value output from the process D model 122D is acquired as a response variable corresponding to the explanatory variable of the inference target.

[0069] The system application unit 1325 applies the objective variable inferred by the first inference unit 1324 as an operation value of the plant system 30. Alternatively, the system application unit 1325 presents the objective variable inferred by the first inference unit 1324 to an operator.

[0070] In Figure 3, an example is explained in which the processing device 10 performs both the learning of the process-specific models and the inference using the process-specific models corresponding to each process, but the learning of the process-specific models and the inference using the process-specific models corresponding to each process may also be performed by a different device.

[0071] [Learning process] Next, a description will be given of the learning process according to Embodiment 1. Fig. 5 is a flowchart of the learning process according to Embodiment 1.

[0072] 5, the processing device 10 collects the history of each production process in the plant system 30 (step S11) and registers each history for each production process in the history DB 121. The processing device 10 reads out the histories of all production processes from the history DB 121 as learning data (step S12).

[0073] The processing device 10 performs learning of the global model using the history of all production processes as learning data (step S13), and optimizes the parameters of the global model.

[0074] The processing device 10 performs additional learning of each process-specific model by setting the parameters of the global model as initial values ​​for each production process and using only the history of the target production process as learning data (step S14). The processing device 10 stores the optimized parameters of each process-specific model in the model information 122.

[0075] [Inference processing] Next, a description will be given of the inference processing according to embodiment 1. Fig. 6 is a flowchart of the inference processing according to embodiment 1.

[0076] 6, the processing device 10 acquires explanatory variables of an inference target (step S21). The processing device 10 determines a production process corresponding to the explanatory variables acquired in step S21 (step S22).

[0077] The processing device 10 performs a first model selection process to select a process-specific model corresponding to the production process identified by the process identification unit 1322 from the plurality of process-specific models 1220 (step S23).

[0078] The processing device 10 performs a first inference process of inferring a response variable corresponding to an explanatory variable of an inference target using the process-specific model selected in step S23 (step S24).

[0079] The processing device 10 applies the response variable inferred in step S24 as an operation value of the plant system 30 or a value presented to an operator (step S25).

[0080] [Effects of the First Embodiment] In this way, in the first embodiment, a process-specific model is constructed for a batch plant with multiple production processes. This process-specific model is generated by additionally learning only the history of each production process, using the parameters of a global model that has learned the history of all production processes as initial values.

[0081] Therefore, according to embodiment 1, even if there is little data corresponding to one production process, the accuracy of the model for each process can be ensured, and an operation support system for a batch plant using a machine learning model that performs imitation learning can be realized.

[0082] Furthermore, because batch-based plants are plants in which each production process is executed sequentially on a single piece of equipment, it is believed that the history of other production processes also contains globally useful information. For example, if the logic of "preventing the furnace temperature from exceeding a certain level" is consistently set across all production processes, operations are performed to prevent the furnace temperature from exceeding the certain level, and common information can be expected to be included. Therefore, a global model trained using the history of all production processes in a batch-based plant is considered to be a model that has learned information common across production processes. Furthermore, in the first embodiment, the accuracy of each process model can be improved by additionally training each process model based on this global model.

[0083] In consideration of the above, when training a global model, instead of the history of all production processes in a batch plant, the history of multiple production processes with similar processing contents may be used as training data, thereby training only the history of multiple similar production processes. This allows training of the global model to be performed more efficiently. Examples of multiple production processes with similar processing contents include heating processes, cooling processes, and processes with common equipment operation methods.

[0084] [Embodiment 2] Next, a description will be given of a processing device 210 according to the second embodiment. Fig. 7 is a diagram for explaining an outline of model learning in the second embodiment.

[0085] In the second embodiment, similarly to the first embodiment, the processing device 210 generates a global model (multiple process model) corresponding to all production processes ((1) in FIG. 7). The processing device 10 inputs the entire history n (1≦n≦N) of all production processes of a batch plant as learning data (arrow Y21), an operation execution time t, and a process variable X, which is an explanatory variable, and generates an operation value Y g Learning is performed to optimize the parameters of the global model so as to output the objective variable.

[0086] Then, the processing device 210 compares the objective variable (operation value Y) in the learning data with the objective variable (operation value Y g ) is used as training data, and a process-specific difference model (for example, process A difference model, process B difference model, process C difference model, process D difference model) that outputs this difference R is trained from the explanatory variables.

[0087] For example, when constructing a process D differential model ((2) in Figure 7), only the history of production process D in the learning data is used. Specifically, the operation execution time t and the explanatory variables (process variables X n ) and the output of the global model (operating value Y g ), operation execution time t and explanatory variables (process variables X nThe difference between the operation time t and the corresponding objective variable (operation value Y) is used as the learning data for the process D difference model. Then, the process D difference model is n The parameters of the process D difference model are optimized so that the output (difference R ((Equation (1))) when

[0088]

number

[0089] Similarly, for the process-specific differential models of the production processes A to C, the processing device 210 executes machine learning using the differences between the objective variables in the history of the corresponding production processes A, B, and C and the objective variables output by the global model as learning data, thereby constructing a process A differential model, a process B differential model, and a process C differential model.

[0090] During inference, the processing device 210 performs inference using a global model and a process-specific difference model corresponding to the production process of the explanatory variable of the inference target. The processing device 210 uses the first output (operation value Y g ) and the second output (R) of the process-specific difference model, which is obtained by inputting the explanatory variables to be inferred into the process-specific difference model corresponding to the production process of the explanatory variables to be inferred, are added together and output as the objective variable Y (equation (2)) of the explanatory variables to be inferred.

[0091]

number

[0092] [Processing equipment] A description will be given of a processing device 210 according to the second embodiment. Fig. 8 is a diagram showing an example of the configuration of the processing device 210 according to the second embodiment.

[0093] The processing device 210 shown in Fig. 8 has a control unit 213 instead of the control unit 13 of the processing device 10 shown in Fig. 3. The processing device 210 also includes parameters of a global model 2123 and a process-specific difference model 2124 as model information 122. The global model 2123 is a model trained to output a target variable from the explanatory variables using the history of all production processes of a batch plant as learning data. The global model 2123 is used during inference for each production process.

[0094] The process-specific difference model 2124 is a model that has been trained to output, for each production process (for example, production processes A to D), the difference between the objective variable in the training data and the objective variable output by the global model as training data, from the explanatory variables. The process-specific difference model 2124 includes, for example, parameters of a process A difference model 2124A used when inferring data from production process A, a process B difference model 2124B used when inferring data from production process B, a process C difference model 2124C used when inferring data from production process C, and a process D difference model 2124D used when inferring data from production process D.

[0095] The control unit 213 has a learning unit 2131 instead of the learning unit 131 shown in Fig. 3. The control unit 213 has an inference unit 2132 instead of the inference unit 132 shown in Fig. 3. The learning unit 2131 executes learning of the global model and the process-specific difference model. The inference unit 2132 infers a target variable from the explanatory variable of the inference target based on the global model and the process-specific difference model corresponding to the production process of the explanatory variable of the inference target, and applies the target variable to the plant system 30.

[0096] The learning unit 2131 has a differential model learning unit 2133 (second learning unit) instead of the additional learning unit 1314 shown in FIG.

[0097] The differential model learning unit 2133 uses, for each production process, the difference between the objective variable in the learning data and the objective variable output by the global model as learning data, and learns a process-specific differential model that outputs the difference between the objective variable and the objective variable output by the global model from the explanatory variables.

[0098] The inference unit 2132 has a second model selection unit 2134 (selection unit) and an inference unit 2135 (inference unit) instead of the first model selection unit 1323 and the first inference unit 1324 shown in FIG.

[0099] The second model selection unit 2134 selects, from the plurality of process-specific differential models 2124, a process-specific differential model that corresponds to the production process identified by the process identification unit 1322.

[0100] The second inference unit 2135 adds together the first output of the global model obtained by inputting the explanatory variables to be inferred into the global model 2123 and the second output of the process-specific differential model obtained by inputting the explanatory variables to be inferred into the process-specific differential model selected by the second model selection unit 2134, and infers the sum as the objective variable of the explanatory variables to be inferred.

[0101] [Learning process] Next, a description will be given of the learning process according to the second embodiment. Fig. 9 is a flowchart of the learning process according to the second embodiment.

[0102] Steps S211 to S213 shown in Fig. 9 are the same processes as steps S11 to S13 in Fig. 5. The parameters of the global model 2123 optimized in step S213 are stored in the model information 122 and are also used during inference.

[0103] The processing device 210 uses the difference between the objective variable in the learning data and the objective variable output by the global model as learning data for each production process, and performs learning of each process-specific differential model that outputs the difference between the objective variable and the objective variable output by the global model from the explanatory variables (step S214). The processing device 10 stores the optimized parameters of each process-specific differential model in the model information 122.

[0104] [Inference processing] Next, a description will be given of the inference processing according to the embodiment 2. Fig. 10 is a flowchart of the inference processing according to the embodiment 2.

[0105] Steps S221 and S222 shown in FIG. 9 are the same processes as steps S21 and S22 shown in FIG.

[0106] The processing device 10 performs a second model selection process to select, from the plurality of process-specific differential models 2124, a process-specific differential model corresponding to the production process identified by the process identifying unit 1322 (step S223).

[0107] The processing device 210 performs a second inference process to infer a response variable corresponding to the explanatory variable of the inference target using the global model 2123 and the process-specific differential model selected in step S23 (step S224). Specifically, the processing device 210 acquires a first output of the global model obtained by inputting the explanatory variable of the inference target into the global model 2123. The processing device 210 acquires a second output of the process-specific differential model obtained by inputting the explanatory variable of the inference target into the process-specific differential model selected by the second model selection unit 2134. The processing device 210 adds the first output and the second output and outputs the sum as a response variable of the explanatory variable of the inference target.

[0108] Step S225 in FIG. 10 is the same process as step S25 shown in FIG.

[0109] [Effects of the second embodiment] The processing device 210 according to the second embodiment infers a response variable corresponding to an explanatory variable of an inference target for a batch plant having multiple production processes, using a global model corresponding to all production processes and a process-specific differential model. This process-specific differential model is a model that receives an explanatory variable of an inference target as input and outputs the difference between the response variable corresponding to the explanatory variable of the inference target and the output of the global model.

[0110] During inference, the processing device 210 outputs the sum of the first output of the global model obtained by inputting the explanatory variables of the inference target into the global model and the second output of the process-specific differential model obtained by inputting the explanatory variables of the inference target into the process-specific differential model corresponding to the production process of the explanatory variables of the inference target as the objective variable of the inference target.

[0111] In a batch plant, each production process is executed sequentially on a single piece of equipment, and therefore, it is believed that data from other processes also contains globally useful information. A global model trained using the history of all production processes is considered to be a model that has learned information common to all processes, and a certain degree of inference accuracy can be ensured using only the output of the global model. Furthermore, in the second embodiment, a process-specific differential model trained using only the history of the corresponding process is used for each production process. These process-specific differential models output the inference results of the global model, which do not fully express the differences in the production processes, for each production process in the form of the difference between the objective variable corresponding to the explanatory variable to be inferred and the output of the global model.

[0112] Therefore, it is considered that the processing device 210 can infer with high accuracy the objective variable of the explanatory variable to be inferred in each production process by adding the first output of the global model and the second output of the process-specific difference model.

[0113] As described above, according to the second embodiment, by using the global model and the differential model for each process, it is possible to realize an operation support system for a batch plant using a machine learning model that performs imitation learning.

[0114] In consideration of the above, when training a global model, instead of the history of all production processes in a batch plant, the history of multiple production processes with similar processing contents may be used as training data, thereby training only the history of the similar multiple production processes. This allows training of the global model to be performed more efficiently. Examples of multiple production processes with similar processing contents include heating processes, cooling processes, and processes with a common equipment operation method.

[0115] In the second embodiment, it is believed that inference accuracy can be ensured even when various machine learning methods other than the gradient descent method are used.

[0116] [Embodiment 3] Next, a description will be given of embodiment 3. In embodiment 3, a process-specific model, or a global model and a process-specific differential model, is selected according to a predetermined rule, and inference is performed.

[0117] [Processing equipment] A description will be given of a processing device according to embodiment 3. Fig. 11 is a diagram showing an example of the configuration of the processing device according to embodiment 3.

[0118] 11, a processing device 310 according to the third embodiment has a control unit 313 instead of the control unit 13 of the processing device 10 shown in Fig. 3. The processing device 310 also includes parameters of a global model 2123, a process-specific model 1220, and a process-specific difference model 2124 as model information 122.

[0119] The control unit 313 has a learning unit 3131 instead of the learning unit 131 shown in Fig. 3. The control unit 313 has an inference unit 3132 instead of the inference unit 132 shown in Fig. 3. The learning unit 3131 has a configuration in which the differential model learning unit 2133 of Fig. 8 is further added to the learning unit 131. The inference unit 3132 has a configuration in which a determination unit 3136, and the second model selection unit 2134 and second inference unit 2135 of Fig. 8 are further added to the inference unit 132.

[0120] The determination unit 3136 determines whether to have the first inference unit 1324 or the second inference unit 2135 perform the inference in accordance with a predetermined rule. The inference process performed by the first inference unit 1324 is the first inference process. The inference process performed by the second inference unit 2135 is the second inference process.

[0121] The predetermined rule is set, for example, based on the inference accuracy of the first inference unit 1324 and the inference accuracy of the second inference unit 2135 in each production process, and associates whether to execute the first inference process or the second inference process for each production process. Alternatively, the predetermined rule associates whether to execute the first inference process or the second inference process for each production process according to the characteristics of each production process. For example, a production process that is known to require more detailed operations or operations using unique logic than other processes is set to execute the first inference process. Furthermore, the predetermined rule may be dynamically changed according to the operational performance of the plant operation system 1.

[0122] [Learning process] Next, a description will be given of the learning process according to Embodiment 3. Fig. 12 is a flowchart of the learning process according to Embodiment 3.

[0123] Steps S311 to S314 shown in Fig. 12 are the same processes as steps S11 to S14 in Fig. 5. Note that the model parameters of the global model are not deleted but are stored in the model information 122. Step S315 shown in Fig. 12 is the same process as step S214 shown in Fig. 9.

[0124] [Inference processing] Next, a description will be given of the inference processing according to the embodiment 3. Fig. 13 is a flowchart of the inference processing according to the embodiment 3.

[0125] Steps S321 and S322 shown in FIG. 13 are the same processes as steps S21 and S22 shown in FIG.

[0126] The processing device 310 determines whether to execute the first inference process or the second inference process according to a predetermined rule (step S323).

[0127] When the first inference process is executed (step S323: first inference process), the processing device 310 executes a first model selection process (step S324) and a first inference process (step S325). Steps S324 and S325 are the same processes as steps S23 and S24 shown in FIG. 6.

[0128] When executing the second inference process (step S323: second inference process), the processing device 310 executes a second model selection process (step S326) and a second inference process (step S327). Steps S326 and S327 are the same processes as steps S224 and S225 shown in Fig. 10. Step S328 shown in Fig. 13 is the same process as step S25 shown in Fig. 6.

[0129] [Effects of the Third Embodiment] In this way, in embodiment 3, for each production process, the inference process that is more suitable for that production process is selected from the first inference process and the second inference process, and inference is performed, thereby improving the inference accuracy.

[0130] [System configuration of the embodiment] The processing device 10 is a functional concept and does not necessarily have to be physically configured as shown in the figure. In other words, the specific form of distribution and integration of the functions of the processing device 10 is not limited to that shown in the figure, and all or part of it can be functionally or physically distributed or integrated in any unit depending on various loads, usage conditions, etc.

[0131] Furthermore, all or any part of the processes performed by the processing device 10 may be realized by a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), and a program analyzed and executed by the CPU and the GPU. Furthermore, each process performed by the processing device 10 may be realized as hardware using wired logic.

[0132] Furthermore, among the processes described in the embodiments, all or part of the processes described as being performed automatically can be performed manually. Alternatively, all or part of the processes described as being performed manually can be performed automatically using a known method. In addition, the processing procedures, control procedures, specific names, and information including various data and parameters described above and illustrated can be changed as appropriate unless otherwise specified.

[0133] [program] 14 is a diagram showing an example of a computer in which a program is executed to implement the processing device 10. The computer 1000 includes, for example, a memory 1010 and a CPU 1020. The computer 1000 also includes a hard disk drive interface 1030, a disk drive interface 1040, a serial port interface 1050, a video adapter 1060, and a network interface 1070. These components are connected by a bus 1080.

[0134] The memory 1010 includes a ROM 1011 and a RAM 1012. The ROM 1011 stores a boot program such as a BIOS (Basic Input Output System). The hard disk drive interface 1030 is connected to a hard disk drive 1090. The disk drive interface 1040 is connected to a disk drive 1100. A removable storage medium such as a magnetic disk or optical disk is inserted into the disk drive 1100. The serial port interface 1050 is connected to a mouse 1110 and a keyboard 1120, for example. The video adapter 1060 is connected to a display 1130, for example.

[0135] The hard disk drive 1090 stores, for example, an OS (Operating System) 1091, an application program 1092, a program module 1093, and program data 1094. That is, a program that defines each process of the processing device 10 is implemented as a program module 1093 in which code executable by the computer 1000 is written. The program module 1093 is stored, for example, in the hard disk drive 1090. For example, a program module 1093 for executing the same process as the functional configuration of the processing device 10 is stored in the hard disk drive 1090. Note that the hard disk drive 1090 may be replaced with an SSD (Solid State Drive).

[0136] Furthermore, setting data used in the processing of the above-described embodiment is stored as program data 1094, for example, in memory 1010 or hard disk drive 1090. Then, CPU 1020 reads program module 1093 and program data 1094 stored in memory 1010 or hard disk drive 1090 into RAM 1012 as necessary and executes them.

[0137] The program module 1093 and program data 1094 are not limited to being stored in the hard disk drive 1090, but may also be stored in, for example, a removable storage medium and read by the CPU 1020 via the disk drive 1100 or the like. Alternatively, the program module 1093 and program data 1094 may be stored in another computer connected via a network (such as a local area network (LAN) or a wide area network (WAN)). The program module 1093 and program data 1094 may then be read by the CPU 1020 from the other computer via the network interface 1070.

[0138] Although the present invention has been described above as an embodiment, the present invention is not limited to the descriptions and drawings that form part of the disclosure of the present invention. In other words, other embodiments, examples, and operational techniques that can be made by those skilled in the art based on the present invention are all included in the scope of the present invention. [Explanation of symbols]

[0139] 1 Plant operation system 10,210,310 Processing equipment 20 Terminal equipment 30 Plant Systems 11 Communications Department 12 Storage section 13,213 Control Unit 121 History DB 122 Model Information 131,2131,3131 Learning Department 132,2132,3132 Reasoning part 1210 History 1220 Process Model 1311 Collection Department 1312 Learning data reading unit 1313 Global Model Learning Department 1314 Additional Learning Department 1321 Data Acquisition Department 1322 Process discrimination section 1323 First Model Selection Section 1324 First Inference Section 1325 System Application Department 2123 Global Model 2124 Process-specific difference model 2133 Differential Model Learning Unit 2134 Second Model Selection Section 2135 Second Inference Section 3136 Judgment section

Claims

1. a collection unit that collects a history of each production process, which is a combination of explanatory variables that represent the status of the product production process and objective variables that represent the operation of equipment in the production process; a first learning unit that optimizes parameters of a multi-process model that outputs the objective variable from the explanatory variables using the histories of the plurality of production processes as learning data; a second learning unit that uses, for each of the production processes, a difference between a response variable in the learning data and a response variable output by the multi-process model as learning data, and learns a process-specific difference model that outputs the difference from the explanatory variables; A learning device comprising:

2. 2. The learning device according to claim 1, wherein the product is manufactured in a batch plant in which a plurality of production processes are executed in sequence using one manufacturing device.

3. 2. The learning device according to claim 1, wherein the plurality of production processes are production processes having similar processing contents.

4. 2. The learning device according to claim 1, wherein the plurality of production processes are heating processes, cooling processes, or processes that share a common equipment operation method.

5. a third learning unit that performs additional learning of a process-specific model that outputs the objective variable from the explanatory variables, for each of the production processes, using parameters of the multi-process model as initial values ​​and the history of the target production process as learning data; 2. The learning device according to claim 1, further comprising:

6. a storage unit that stores: a multi-process model that is trained using, as training data, histories of a plurality of production processes among histories that are combinations of explanatory variables that represent the status of a product production process and objective variables that represent the operation of equipment in the production process, and that outputs the objective variable from the explanatory variables; and a plurality of process-specific differential models provided for each of the production processes, that, when the explanatory variables are input, output a difference between an objective variable corresponding to the input explanatory variable and the objective variable output by the multi-process model; an acquisition unit that acquires the explanatory variables of the inference target; a discrimination unit that discriminates the production process corresponding to the explanatory variable acquired by the acquisition unit; a selection unit that selects, from the plurality of process-specific differential models, the process-specific differential model that corresponds to the production process identified by the identification unit; an inference unit that adds together a first output of the multi-step model obtained by inputting explanatory variables of the inference target into the multi-step model and a second output of the process-specific difference model obtained by inputting explanatory variables of the inference target into the process-specific difference model selected by the selection unit, and infers the sum as the objective variable of the explanatory variables of the inference target; An inference device comprising:

7. the storage unit stores a plurality of process models corresponding to the production processes, each of which outputs the objective variable from the explanatory variable; a process-specific model selection unit that selects, from the plurality of process-specific models, the process-specific model that corresponds to the production process identified by the identification unit; a step-by-step inference unit that infers the objective variable corresponding to the explanatory variable of the inference target using the step-by-step model selected by the step-by-step model selection unit; a determination unit that determines, for each production process, whether to cause the inference unit or the process-specific inference unit to execute inference in accordance with a predetermined rule; and 7. The inference device according to claim 6, wherein the inference unit or the process-specific inference unit executes inference in accordance with the determination of the determination unit.

8. A learning method executed by a learning device, a collection step of collecting a history of each production process, which is a combination of explanatory variables representing the status of the product production process and objective variables representing the operation of equipment in the production process; a first learning step of optimizing parameters of a multi-step model that outputs the objective variable from the explanatory variables using the histories of the plurality of production steps as learning data; a second learning step of learning, for each of the production processes, a difference between a response variable in the learning data and a response variable output by the multi-process model as learning data, to learn a process-specific difference model that outputs the difference from the explanatory variables; A learning method comprising:

9. An inference method executed by an inference device, comprising: the inference device has a storage unit that stores: a multi-process model that is trained using, as training data, histories of a plurality of production processes among histories that are combinations of explanatory variables that represent the status of a product production process and objective variables that represent operations of equipment in the production process, and that outputs the objective variable from the explanatory variables; and a plurality of process-specific difference models provided for each of the production processes, that, when the explanatory variables are input, output the difference between the objective variable corresponding to the input explanatory variable and the objective variable output by the multi-process model; an acquisition step of acquiring the explanatory variables of the inference target; a determining step of determining the production process corresponding to the explanatory variables acquired in the acquiring step; a selection step of selecting, from the plurality of process-specific differential models, the process-specific differential model corresponding to the production process identified in the identification step; an inference step of adding together a first output of the multi-step model obtained by inputting explanatory variables of the inference target into the multi-step model and a second output of the process-specific difference model obtained by inputting explanatory variables of the inference target into the process-specific difference model selected in the selection step, and inferring the sum as the objective variable of the explanatory variables of the inference target; An inference method comprising:

10. a collection step of collecting a history of each production process, which is a combination of explanatory variables representing a situation in the production process of the product and objective variables representing the operation of equipment in the production process; a first learning step of optimizing parameters of a multi-process model that outputs the objective variable from the explanatory variables using the histories of the plurality of production processes as learning data; a second learning step of learning, for each of the production processes, a difference between a response variable in the learning data and a response variable output by the multi-process model as learning data, to learn a process-specific difference model that outputs the difference from the explanatory variables; A learning program that allows a computer to execute the above.

11. an acquisition step of acquiring explanatory variables that represent a situation in a production process of a product and are to be inferred; a determining step of determining the production process corresponding to the explanatory variables acquired in the acquiring step; a selection step of selecting, from among a plurality of process-specific differential models provided for the respective production processes, the process-specific differential models outputting the difference between a response variable corresponding to the input explanatory variable and a response variable output by a multi-process model trained using training data of histories of a plurality of production processes among histories that are combinations of response variables representing device operations in the production process, when the explanatory variables are input, the process-specific differential model corresponding to the production process identified in the discrimination step; an inference step of adding together a first output of the multi-step model obtained by inputting explanatory variables of the inference target into the multi-step model and a second output of the process-specific difference model obtained by inputting explanatory variables of the inference target into the process-specific difference model selected in the selection step, and inferring the sum as the objective variable of the explanatory variables of the inference target; An inference program that allows a computer to execute the above.

Images