Apparatus, method, and program

Abstract

To provide an apparatus, a method, and a program for performing various types of control using a learning model in a facility such as a plant.SOLUTION: In a system 10 including a plurality of plants configured in a mesh network, a processing apparatus installed in a plant 20A comprises: a receiving unit for receiving model parameters related to a first model generated by learning so as to output an output value corresponding to a state value in transmission source plants 20B to 20D; a data generation unit for generating a data set of the state value and the output value using the first model corresponding to the received model parameters; and a model generation unit for generating, by learning, a second model that outputs the output value corresponding to the state value in the plant 20A by using the data set as learning data.SELECTED DRAWING: Figure 1

Description

[Technical field]

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

[0002] In facilities such as plants, various types of control are performed using learning models. Summary of the Invention

[0003] In a first aspect of the present invention, there is provided an apparatus including a receiving unit that receives model parameters for a first model generated by learning so as to output an output value corresponding to a state value in a first plant, a data generating unit that generates a data set of state values ​​and output values ​​using the first model corresponding to the received model parameters, and a model generating unit that uses the data set as learning data to generate by learning a second model that outputs an output value corresponding to a state value in a second plant.

[0004] In the above device, the model generation unit may generate the second model using a data set of state values ​​and output values ​​acquired in the second plant and a data set of state values ​​and output values ​​generated from the first model as learning data.

[0005] In the above device, the model generation unit may generate the second model by machine learning, weighting a data set of state values ​​and output values ​​acquired in the second plant more heavily than the data set of state values ​​and output values ​​generated from the first model.

[0006] In the above device, the model generation unit may generate the second model by machine learning a data set of state values ​​and output values ​​acquired in the second plant at a learning rate greater than a learning rate for the data set of state values ​​and output values ​​generated from the first model.

[0007] In any of the above devices, a first calculation unit may be provided that calculates an evaluation value of the first model when a state value acquired in the second plant is input, and the model generation unit may generate the second model without using the data set as learning data when the evaluation value of the first model is equal to or less than a threshold value.

[0008] In the above device, a request unit may be provided that requests transmission of model parameters for a third model generated by learning so as to output an output value corresponding to a state value in the third plant when the evaluation value of the first model is equal to or less than a threshold value, and the receiving unit may receive the model parameters transmitted in response to the request.

[0009] Any of the above devices may include a second calculation unit that calculates an evaluation value of the second model generated by the model generation unit, and a request unit that requests the transmission of new model parameters for the first model when the evaluation value of the second model is equal to or less than a threshold, and the receiving unit may receive the model parameters transmitted in response to the request.

[0010] In any of the above devices, the first model may have a smaller amount of model parameters than a model used to output an output value according to a state value in the first plant.

[0011] In any of the above apparatus, the second plant may differ from the first plant in at least one of its configuration or location.

[0012] In a second aspect of the present invention, there is provided a method comprising the steps of receiving model parameters for a first model generated by learning so as to output an output value corresponding to a state value in a first plant, generating a data set of state values ​​and output values ​​using the first model corresponding to the received model parameters, and generating by learning a second model using the data set as learning data so as to output an output value corresponding to a state value in a second plant.

[0013] In a third aspect of the present invention, there is provided a program that is executed by a computer and causes the computer to function as a receiving unit that receives model parameters for a first model generated by learning so as to output an output value corresponding to a state value in a first plant, a data generating unit that generates a data set of state values ​​and output values ​​using the first model corresponding to the received model parameters, and a model generating unit that uses the data set as learning data to generate a second model by learning that outputs an output value corresponding to a state value in a second plant.

[0014] The above summary of the invention does not list all of the necessary features of the present invention. Also, subcombinations of these features may also be inventions. [Brief description of the drawings]

[0015] [Figure 1] 1 shows a schematic diagram of a system 10 according to the present embodiment. [Diagram 2] An example of the configuration of a treatment device 100 installed in one of a plurality of plants 20 is shown. [Diagram 3] 2 shows a model generation flow of the processing device 100 according to the present embodiment. [Figure 4] 1 shows a model transmission flow of the processing device 100 according to the present embodiment. [Diagram 5] 22 illustrates an example computer 2200 in which aspects of the present invention may be embodied, in whole or in part. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] The present invention will be described below through embodiments of the invention, but the following embodiments do not limit the scope of the invention according to the claims. Furthermore, not all of the combinations of features described in the embodiments are necessarily essential to the solution of the invention.

[0017] FIG. 1 shows a schematic diagram of a system 10 in this embodiment. The system 10 includes a plurality of plants 20A, 20B, 20C, and 20D (hereinafter, simply referred to as plants 20) configured in a mesh network. The plants 20 may be, for example, industrial plants such as chemical or metal plants, plants that manage and control wellheads and surrounding areas of gas fields and oil fields, plants that manage and control power generation such as hydroelectric power, thermal power, and nuclear power, plants that manage and control environmental power generation such as solar power and wind power, and plants that manage and control water supply and sewage, dams, etc. At least one of the multiple plants 20 is different from at least one of the other multiple plants 20 in at least one of the configurations or installation locations. The multiple plants 20 may be managed and controlled to have the same or different objects.

[0018] The multiple plants 20 are connected to each other by wire or wirelessly and transmit and receive data. In each of the multiple plants 20, various controls are performed using a model that outputs an output value according to a state value acquired from a sensor or the like installed in equipment in the plant 20. The models of the multiple plants 20 are each generated by autonomous distributed machine learning. In each of the multiple plants 20, learning data is generated from a model of another plant 20, and a model generated using the learning data is used.

[0019] Here, the state value may be a value indicating at least one of the temperature, pressure, flow rate, set value, position information, and voice data in the equipment of the plant 20, and the same applies below. The output value may be a value indicating at least one of the presence or absence of an abnormality in the equipment of the plant 20, the location of the abnormality, the quality of the product, the production amount, and the control result, and the same applies below.

[0020] Furthermore, the model used in the plant 20 may be a model trained by any other machine learning algorithm including, for example, a neural network, statistical learning, random forest, gradient boosting, logistic regression, or a support vector machine (SVM). The model may be a model that outputs an output value indicating an abnormality prediction result of the equipment in the plant 20, a product quality estimation result, a production volume prediction, an abnormality cause identification result, an optimal control value, or the like, in response to an input of a state value. For example, the model may output at least one of an output value indicating an abnormality occurring in the equipment or product in the plant 20 or an output value indicating no abnormality occurring, when one or more state values ​​(for example, temperature, pressure, etc.) detected in the plant 20 are input. The model may output an output value indicating an optimal control amount of a control target such as a valve in the plant 20, when one or more state values ​​(for example, temperature, pressure, etc.) detected in the plant 20 are input.

[0021] FIG. 2 shows a configuration example of a processing device 100 that performs data processing for one plant 20A among a plurality of plants 20. The processing device 100 is installed inside or outside the plant 20A, and is connected to each piece of equipment in the plant 20A by wire or wirelessly, and is connected to the other plants 20B, 20C, and 20D by wire or wirelessly. The processing device 100 acquires various data from N pieces of equipment 25-1 to 25-N (N indicates an integer of 1 or more) installed in the plant 20A, and processes data related to the control of the plant 20A. Each of the plants 20B, 20C, and 20D may include a device having a configuration and operation similar to that of the processing device 100, and the device in the plant 20B, 20C, and 20D may be connected to the processing device 100.

[0022] Here, the device 25 of the plant 20A may be a field device, and may be, for example, a sensor device such as a pressure gauge, a flow meter, or a temperature sensor, a valve device such as a flow control valve or an on-off valve, an actuator device such as a fan or a motor, an imaging device such as a camera or a video camera that captures the situation of the plant 20A or an object such as the equipment, an audio device such as a microphone or a speaker that collects abnormal sounds of the equipment of the plant 20A or emits alarm sounds, a position detection device that outputs position information of a device in the equipment of the plant 20A, or other devices. Also, the device 25 of the plant 20A may be a computer such as a PC (personal computer), a tablet computer, or a smartphone used by an operator or manager of the plant 20A.

[0023] The processing device 100 may be realized by a computer such as a PC, a tablet computer, a smartphone, a workstation, a server computer, or a general-purpose computer, or may be realized by a computer system in which multiple computers are connected. Such a computer system is also a computer in the broad sense. The processing device 100 may be a dedicated computer designed for controlling a plant, or may be dedicated hardware realized by a dedicated circuit.

[0024] The processing device 100 includes an acquisition unit 110, a model generation unit 120, a second calculation unit 130, a model storage unit 140, a control unit 150, a communication unit 160, a first calculation unit 170, a data generation unit 180, and a transmission model generation unit 190.

[0025] The acquisition unit 110 is connected to the equipment 25 of the plant 20A, the model generation unit 120, the first calculation unit 170, and the data generation unit 180. The acquisition unit 110 may acquire at least one of a state value in the plant 20A detected by the equipment 25 of the plant 20A and an output value corresponding to the state value. The acquisition unit 110 may acquire, as an output value, a value indicating at least one of the presence or absence of an abnormality in the equipment, the location of the abnormality, the quality of the product, the production amount, or the control result input by an operator or manager of the plant 20A, from the equipment 25. The acquisition unit 110 may acquire the state value input by an operator or manager of the plant 20A. The acquisition unit 110 may acquire a data set generated in the data generation unit 180.

[0026] The model generation unit 120 is connected to the second calculation unit 130. The model generation unit 120 uses the data acquired by the acquisition unit 110 as learning data to generate, by learning, a model (hereinafter also referred to as a second model) that outputs an output value corresponding to a state value in the plant 20A.

[0027] The second calculation unit 130 is connected to the model storage unit 140 and the communication unit 160. The second calculation unit 130 calculates an evaluation value of the second model generated by the model generation unit 120. When the second model is a classification model, the second calculation unit 130 may calculate at least one of the following as the evaluation value of the second model: accuracy rate, precision rate, recall rate, F-measure, specificity, or false negative rate. When the second model is a regression model, the second calculation unit 130 may calculate at least one of the following as the evaluation value of the second model: MAE (mean absolute error), MSE (mean square error), RMSE (mean squared error), or R-squared coefficient of determination. As an example, the second calculation unit 130 may input a state value acquired from the equipment 25 of the plant 20A by the acquisition unit 110 to the second model, and calculate the ratio (probability) of the output value corresponding to the input being correct as the accuracy rate. Here, the evaluation value may be an index for evaluating the prediction accuracy of the second model, and is not particularly limited to the above example.

[0028] The model storage unit 140 is connected to the control unit 150 and the transmission model generation unit 190 , and stores the second model generated by the model generation unit 120 .

[0029] The control unit 150 is connected to the equipment of the plant 20A. The control unit 150 controls the operation of the plant 20A by using the second model stored in the model storage unit 140. The control unit 150 inputs the state value acquired by the acquisition unit 110 from the equipment 25 of the plant 20A to the second model, and generates an output value according to the input. The control unit 150 outputs control data according to the generated output value to the plant 20A. The control unit 150 may output control data to the plant 20A for displaying the output value on the equipment 25.

[0030] The communication unit 160 is wirelessly connected to the plants 20B, 20C, and 20D, and transmits and receives data to and from the plants 20B, 20C, and 20D. The communication unit 160 includes a receiving unit 162 and a request unit 164.

[0031] The receiver 162 receives, from at least one of the plants 20B, 20C, and 20D, model parameters related to a model (hereinafter also referred to as a first model) generated by learning so as to output an output value corresponding to a state value in the transmission source plant 20. The receiver 162 may receive model parameters related to the first model used for operating the transmission source plants 20B, 20C, and 20D.

[0032] Here, the model parameters may be parameters for constructing, in the processing device 100, a model having the same input / output relationship (relationship between state values ​​and output values) as the first model of the source plant 20. The model parameters include, for example, at least one of the weights used in the first model of the source plant 20, the hyperparameters, the type of state value input to the first model, the type of output value of the first model, or the type of the first model. For example, when the first model of the source plant 20 is a neural network model, the model parameters may include at least one of the type of state value input to the input layer of the first model, the type of output value output at the output layer, the number of neurons, the number of layers, or the weights between neurons.

[0033] The request unit 164 transmits request data requesting transmission of new model parameters to at least one of the plants 20B, 20C, and 20D when at least one of the evaluation value (e.g., accuracy rate) of the second model generated by the model generation unit 120 and the evaluation value (e.g., accuracy rate) of the first model according to the model parameters received by the receiving unit 162 is equal to or lower than a threshold value. In this way, the receiving unit 162 may receive model parameters transmitted in response to a request from the other plants 20.

[0034] The first calculation unit 170 calculates an evaluation value of the first model when the state value acquired in the plant 20A is input. When the first model is a classification model, the first calculation unit 170 may calculate at least one of the following as the evaluation value of the first model: a correct answer rate, a precision rate, a recall rate, an F value, a specificity rate, or a false negative rate. When the first model is a regression model, the first calculation unit 170 may calculate at least one of the following as the evaluation value of the first model: a mean absolute error (MAE), a mean square error (MSE), a mean square error (RMSE), or an R-squared coefficient of determination. As an example, the first calculation unit 170 may input a state value acquired from the device 25 of the plant 20A by the acquisition unit 110 to the first model according to the model parameters received by the reception unit 162, and calculate the rate at which the output value according to the input is correct as the correct answer rate. Here, the evaluation value may be an index for evaluating the prediction accuracy of the first model, and is not particularly limited to the above example.

[0035] The data generating unit 180 generates a data set of state values ​​and output values ​​by using a first model according to the model parameters received by the receiving unit 162. The data generating unit 180 may supply the generated data set to the acquiring unit 110. When the evaluation value of the first model calculated by the first calculating unit 170 is equal to or less than a threshold, the model generating unit 120 may generate a second model without using the data set as learning data.

[0036] The transmission model generation unit 190 is connected to the communication unit 160, and generates a transmission model for transmitting model parameters to another plant 20, based on the second model stored in the model storage unit 140. The transmission model generation unit 190 may generate a transmission model having a smaller amount of model parameters than the second model, based on the second model stored in the model storage unit 140, and may generate a distillation model as an example. The transmission model generation unit 190 transmits the model parameters of the transmission model to the other plant 20 via the communication unit 160.

[0037] 3 shows a model generation flow of the processing device 100 according to this embodiment. In S300, the receiving unit 162 receives model parameters of a first model of the external plants 20B, 20C, and 20D from at least one of the plants 20B, 20C, and 20D. The receiving unit 162 may receive model parameters of a model (first model) used in operation to output an output value corresponding to a state value in the plant 20 that is the transmission source.

[0038] Furthermore, the receiving unit 162 may receive model parameters of a model (first model) having a smaller amount of model parameters compared to a model used in operation in the plant 20 of the transmission source. For example, the receiving unit 162 may receive model parameters of a model having fewer types or fewer number of model parameters required to construct the model compared to a model used in operation. When the model used in operation in the plant 20 of the transmission source is a neural network model, the receiving unit 162 may receive model parameters of a model having fewer types of state values ​​input to the input layer, types of output values ​​output in the output layer, number of neurons, or number of layers compared to a model used in operation. As an example, the receiving unit 162 may receive model parameters of a distillation model (first model) generated from a model used in operation in the plant 20 of the transmission source. The distillation model is generated based on the relationship between state values ​​and output values ​​of the operation model, and since the distillation model has fewer model parameters than the operation model, the communication load for transmitting and receiving the model parameters can be further reduced.

[0039] Before the model generation flow, the processing device 100 may transmit request data requesting transmission of model parameters to at least one of the plants 20B, 20C, and 20D via the request unit 164. The receiving unit 162 may receive the model parameters transmitted in response to the request data. The receiving unit 162 may also receive model parameters periodically transmitted from at least one of the plants 20B, 20C, and 20D.

[0040] In S310, the first calculation unit 170 may calculate the accuracy rate of the first model as the evaluation value. The first calculation unit 170 may construct a first model according to the received model parameters, input a state value acquired in the plant 20A to the first model, and determine whether or not an output value of the first model according to the state value is correct. The first calculation unit 170 may determine whether or not an output value of the first model is correct, using a state value acquired by the acquisition unit 110 from the device 25 of the plant 20A and an output value at the state value. The first calculation unit 170 may compare the output value of the first model according to the input state value with the output value acquired by the acquisition unit 110 from the plant 20A, and determine that the answer is correct if they match or the difference is equal to or less than a threshold. On the other hand, the first calculation unit 170 may compare the output value of the first model according to the input state value with the output value acquired by the acquisition unit 110 from the plant 20A, and determine that the answer is not correct (wrong answer) if they do not match or the difference exceeds a threshold. The first calculation unit 170 may determine whether the output value of the first model is correct or incorrect for all or some of the data sets of the state values ​​and their output values ​​acquired by the acquisition unit 110 from the plant 20A, and calculate the rate of correct answers to the determined total number (total number of data sets used) as the accuracy rate. When the first calculation unit 170 receives model parameters from multiple plants 20B, 20C, and 20D, the first calculation unit 170 may construct a first model and calculate the accuracy rate for each of the plants 20 that are the transmission sources.

[0041] The first calculation unit 170 may compare the calculated correct answer rate with a threshold value. If the first calculation unit 170 determines that the correct answer rate exceeds the threshold value (Yes in FIG. 3), the flow proceeds to S320, and if the first calculation unit 170 determines that the correct answer rate is equal to or less than the threshold value (No in FIG. 3), the flow returns to S300. The threshold value used in this step may be preset by the user.

[0042] When returning to S300, that is, when the evaluation value (e.g., accuracy rate) of the first model is equal to or less than a threshold, the request unit 164 requests the transmission of model parameters related to a model generated by learning so as to output an output value corresponding to a state value in the other plant 20. For example, when the accuracy rate of the first model corresponding to the model parameters received from the plant 20B is equal to or less than a threshold, the request unit 164 may transmit request data requesting the transmission of the model parameters to at least one of the other plants 20C and 20D. The request unit 164 may transmit the request data to the multiple plants 20 in order, and the processing device 100 may execute steps 300 and 310 each time the processing device 100 receives the model parameters corresponding to the request data. In this case, the request unit 164 may transmit the request data to the multiple plants 20 in an order predetermined by the user. In step 300, the receiving unit 162 receives the model parameters transmitted in response to the request, and the processing device 100 proceeds to the following step S310.

[0043] In S320, the data generating unit 180 generates a data set of state values ​​and output values ​​by inputting state values ​​to the constructed first model and outputting output values ​​according to the input state values. The data generating unit 180 may generate a plurality of data sets by changing the state values. The data generating unit 180 may input a set of a plurality of types of state values ​​corresponding to the input of the first model to the first model and generate one or a plurality of output values ​​according to the input. The data generating unit 180 may randomly generate state values ​​within a predetermined range and input the generated state values ​​to the first model. The predetermined range may be set by a user for each type of state value, or may be a range in which the state values ​​in the plant 20A acquired by the acquiring unit 110 are distributed. Alternatively, a data set of the acquired state values ​​and the output values ​​of the first model may be generated each time the acquiring unit 110 acquires a new state value from the plant 20A. The data generating unit 180 supplies a plurality of data sets of the generated state values ​​and output values ​​to the acquiring unit 110. The data generation unit 180 does not need to generate a data set for the first model for which the accuracy rate is equal to or less than the threshold in S310, and in this case does not supply the data set to the acquisition unit 110.

[0044] In S330, the acquisition unit 110 acquires a data set from the data generation unit 180. The acquisition unit 110 may acquire a data set of state values ​​and output values ​​corresponding to the state values ​​from the devices 25 in the plant 20A. The acquisition unit 110 supplies the acquired data set to the model generation unit 120.

[0045] In S340, the model generation unit 120 uses the data set supplied from the acquisition unit 110 as learning data to generate a second model to be used in the operation of the plant 20A by learning. The model generation unit 120 may generate the second model by using, as learning data, the data set of state values ​​and output values ​​acquired in the plant 20A by the acquisition unit 110 and the data set of state values ​​and output values ​​generated from the first model.

[0046] The model generating unit 120 may generate the second model by performing machine learning while weighting the data set of state values ​​and output values ​​acquired in the plant 20A more heavily than the data set of state values ​​and output values ​​generated from the first model. The model generating unit 120 may weight the data set by at least one of a learning rate in the machine learning or the number of data sets used as learning data.

[0047] The model generating unit 120 may generate the second model by machine learning the data set of state values ​​and output values ​​acquired in the plant 20A at a learning rate greater than that of the data set of state values ​​and output values ​​generated from the first model. In the iterative learning, the model generating unit 120 may adjust the weights of the second model with a larger amount of change in learning using the data set acquired in the second plant 20, and may adjust the weights of the second model with a smaller amount of change in learning using the data set generated by the data generating unit 180.

[0048] The model generating unit 120 may generate the second model by performing machine learning using, as learning data, a number of data sets of state values ​​and output values ​​acquired in the second plant 20 that is greater than the number of data sets of state values ​​and output values ​​generated from the first model. The model generating unit 120 may generate the second model by performing machine learning using, as learning data, a predetermined ratio of the number of data sets acquired in the second plant 20 and the data set generated by the data generating unit 180. The predetermined ratio may be set by a user.

[0049] If the acquisition unit 110 has not been able to acquire a data set of state values ​​and output values ​​in the second plant 20, the model generation unit 120 may generate the second model by using only the data set generated by the data generation unit 180 as learning data. The acquisition unit 110 may acquire a data set of state values ​​and output values ​​from a simulation model that reproduces the state and behavior of the plant A, and the model generation unit 120 may generate the second model by further using the data set acquired from the simulation model as learning data. In this case, the acquisition unit 110 may acquire a data set from an external simulator device.

[0050] In S350, the second calculation unit 130 calculates the accuracy rate of the second model generated by the model generation unit 120 as an evaluation value. The second calculation unit 130 may input a state value acquired in the plant 20A to the second model generated by the model generation unit 120, and determine whether an output value corresponding to the state value is correct. The second calculation unit 130 may use the state value and the output value acquired by the acquisition unit 110 in the plant 20A. The second calculation unit 130 may compare the output value of the second model corresponding to the input of the state value with the output value corresponding to the state value acquired by the acquisition unit 110 in the plant 20A, and determine that the answer is correct if they match or the difference is equal to or less than a threshold. On the other hand, the second calculation unit 130 may compare the output value of the second model corresponding to the input of the state value with the output value corresponding to the state value acquired by the acquisition unit 110 in the plant 20A, and determine that the answer is not correct (wrong answer) if they do not match or the difference exceeds a threshold. The second calculation unit 130 may determine whether the output value of the second model is correct or incorrect for all or some of the data sets of state values ​​and output values ​​acquired by the acquisition unit 110 in the plant 20A, and calculate the proportion of correct answers to the total number determined (the total number of data sets used) as the accuracy rate.

[0051] The second calculation unit 130 may compare the calculated correct answer rate with a threshold value. If the second calculation unit 130 determines that the correct answer rate exceeds the threshold value (Yes in FIG. 3), the flow proceeds to S360, and if the correct answer rate is determined to be equal to or less than the threshold value (No in FIG. 3), the flow returns to S300. The threshold value used in this step may be preset by the user.

[0052] When returning to S300, that is, when the evaluation value (e.g., accuracy rate) of the second model is equal to or less than a threshold, the request unit 164 requests the transmission of new model parameters related to the first model. For example, when the accuracy rate of the second model generated using the model parameters received from the plant 20B is equal to or less than a threshold, the request unit 164 may transmit request data requesting the transmission of new model parameters to the plant 20B. The request unit 164 may transmit request data requesting the transmission of new model parameters after a predetermined period has elapsed since the previous reception of the model parameters. When the accuracy rate of the second model is equal to or less than a threshold, the request unit 164 may transmit request data requesting the transmission of model parameters to at least one of the other plants 20C and 20D. In step 300, the receiving unit 162 receives the model parameters transmitted in response to the request, and the processing device 100 proceeds to the following step S310. Note that the predetermined period in this step may be set in advance by the user.

[0053] In S360, the model storage unit 140 stores the generated second model. When the second model is stored, the control unit 150 can control the operation of the plant 20A using the second model. The model generation flow may be executed periodically or in response to reception of model parameters, and may update the second model in the model storage unit 140.

[0054] The model parameters of the second model generated by such a model generation flow may be transmitted via the communication unit 160 in response to request data from the other plants 20B, 20C, and 20D.

[0055] 4 shows a model transmission flow of the processing device 100 according to this embodiment. In S400, the processing device 100 generates a second model. The processing device 100 generates the second model in the model generation flow of FIG.

[0056] In S410, the transmission model generation unit 190 generates a transmission model having a smaller amount of model parameters compared to the second model stored in the model storage unit 140. The transmission model generation unit 190 may generate a distilled model from the second model stored in the model storage unit 140 by machine learning. The transmission model generation unit 190 inputs a state value to the second model stored in the model storage unit 140, generates an output value according to the state value, and generates a data set of the state value and the output value. The transmission model generation unit 190 may input a set of multiple types of state values ​​corresponding to multiple inputs of the second model to the model, and generate one or multiple output values ​​according to the input. The transmission model generation unit 190 may input a state value of the plant 20A acquired by the acquisition unit 110 to the second model. In addition, the transmission model generation unit 190 may generate a state value randomly within a predetermined range, and input the generated state value to the second model. The predetermined range may be set by the user for each type of state value, or may be a range in which the state values ​​in the plant 20A acquired by the acquisition unit 110 are distributed.

[0057] The transmission model generation unit 190 generates a distilled model by machine learning using a data set generated using the second model as learning data. The transmission model generation unit 190 may generate the distilled model by learning the relationship between the state value and the output value of the generated data set. The transmission model generation unit 190 may learn the distilled model so that the same output value as the second model is output when the input state value is the same. The transmission model generation unit 190 outputs model parameters of the generated distilled model to the communication unit 160.

[0058] The transmission model generation unit 190 may execute S410 in response to the communication unit 160 receiving request data from another plant 20 requesting the transmission of model parameters.

[0059] In S420, the communication unit 160 transmits the model parameters supplied from the transmission model generation unit 190 to another plant 20 or the requesting plant 20.

[0060] According to the above embodiment, learning data sets for machine learning can be efficiently acquired from other plants, and the accuracy of the model can be improved. For example, the number of data when the plant 20 operates for one year with 1000 sensors and a collection cycle of 1 minute is about 520 million cells (1000 columns and 525,600 rows). If such raw plant data is communicated, the communication volume becomes large, about 3.9 GB in double type (8 bytes). By transmitting and receiving model parameters between multiple plants 20, the communication volume can be dramatically reduced compared to transmitting and receiving raw plant data acquired in the plant 20. Furthermore, by transmitting and receiving model parameters of a distillation model created from an operation model, the communication volume can be reduced to about several to several tens of MB. This allows the size of the data to be transmitted and received to be dramatically reduced compared to data compression. Furthermore, by transmitting and receiving model parameters, the leakage of raw plant data can be prevented.

[0061] The processing device 100 may not have the transmission model generation unit 190. In this case, the processing device 100 may transmit the model parameters of the second model stored in the model storage unit 140 to the other plant 20 via the communication unit 160. The control unit 150 may not control the plant 20A using the second model, and may control the plant 20A using, for example, a black box control algorithm. Even in this case, the processing device 100 can contribute to the generation of models in the other plants 20B, 20C, and 20D of the system 10 by supplying the model parameters related to the second model or its distillation model stored in the model storage unit 140 to the other plants 20B, 20C, and 20D.

[0062] Various embodiments of the present invention may also be described with reference to flow charts and block diagrams, where the blocks may represent (1) stages of a process in which operations are performed or (2) sections of an apparatus responsible for performing the operations. Particular stages and sections may be implemented by dedicated circuitry, programmable circuitry provided with computer readable instructions stored on a computer readable medium, and / or a processor provided with computer readable instructions stored on a computer readable medium. Dedicated circuitry may include digital and / or analog hardware circuitry and may include integrated circuits (ICs) and / or discrete circuits. Programmable circuitry may include reconfigurable hardware circuitry including logical AND, logical OR, logical XOR, logical NAND, logical NOR, and other logical operations, memory elements such as flip-flops, registers, field programmable gate arrays (FPGAs), programmable logic arrays (PLAs), and the like.

[0063] A computer-readable medium may include any tangible device capable of storing instructions that are executed by a suitable device, such that the computer-readable medium having instructions stored thereon comprises an article of manufacture that includes instructions that can be executed to create means for performing the operations specified in the flowcharts or block diagrams. Examples of computer-readable media may include electronic storage media, magnetic storage media, optical storage media, electromagnetic storage media, semiconductor storage media, and the like. More specific examples of computer-readable media may include floppy disks, diskettes, hard disks, random access memories (RAMs), read-only memories (ROMs), erasable programmable read-only memories (EPROMs or flash memories), electrically erasable programmable read-only memories (EEPROMs), static random access memories (SRAMs), compact disk read-only memories (CD-ROMs), digital versatile disks (DVDs), Blu-ray (RTM) disks, memory sticks, integrated circuit cards, and the like.

[0064] The computer readable instructions may include either assembler instructions, Instruction Set Architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk®, JAVA®, C++, etc., and conventional procedural programming languages ​​such as the “C” programming language or similar programming languages.

[0065] Computer readable instructions may be provided to a processor or programmable circuitry of a general purpose computer, special purpose computer, or other programmable processing device, either locally or over a wide area network (WAN) such as a local area network (LAN), the Internet, etc., to execute the computer readable instructions to create means for performing the operations specified in the flowcharts or block diagrams. Examples of processors include computer processors, processing units, microprocessors, digital signal processors, controllers, microcontrollers, etc.

[0066] 5 illustrates an example of a computer 2200 in which aspects of the present invention may be embodied in whole or in part. Programs installed on the computer 2200 may cause the computer 2200 to function as or perform operations associated with an apparatus or one or more sections of the apparatus according to an embodiment of the present invention, and / or to perform a process or steps of the process according to an embodiment of the present invention. Such programs may be executed by the CPU 2212 to cause the computer 2200 to perform specific operations associated with some or all of the blocks of the flowcharts and block diagrams described herein.

[0067] The computer 2200 according to this embodiment includes a CPU 2212, a RAM 2214, a graphic controller 2216, and a display device 2218, which are connected to each other by a host controller 2210. The computer 2200 also includes input / output units such as a communication interface 2222, a hard disk drive 2224, a DVD-ROM drive 2226, and an IC card drive, which are connected to the host controller 2210 via an input / output controller 2220. The computer also includes legacy input / output units such as a ROM 2230 and a keyboard 2242, which are connected to the input / output controller 2220 via an input / output chip 2240.

[0068] The CPU 2212 operates according to a program stored in the ROM 2230 and the RAM 2214, thereby controlling each unit. The graphic controller 2216 retrieves image data generated by the CPU 2212 into a frame buffer or the like provided in the RAM 2214 or into the graphic controller 2216 itself, and causes the image data to be displayed on the display device 2218.

[0069] The communication interface 2222 communicates with other electronic devices via a network. The hard disk drive 2224 stores programs and data used by the CPU 2212 in the computer 2200. The DVD-ROM drive 2226 reads programs or data from the DVD-ROM 2201 and provides the programs or data to the hard disk drive 2224 via the RAM 2214. The IC card drive reads programs and data from an IC card and / or writes programs and data to an IC card.

[0070] The ROM 2230 stores therein a boot program or the like executed by the computer 2200 upon activation, and / or a program that depends on the hardware of the computer 2200. The input / output chip 2240 may also connect various input / output units to the input / output controller 2220 via a parallel port, a serial port, a keyboard port, a mouse port, and the like.

[0071] The programs are provided by a computer-readable medium such as a DVD-ROM 2201 or an IC card. The programs are read from the computer-readable medium, installed in the hard disk drive 2224, the RAM 2214, or the ROM 2230, which are also examples of computer-readable media, and executed by the CPU 2212. Information processing described in these programs is read by the computer 2200, and brings about cooperation between the programs and the various types of hardware resources described above. An apparatus or method may be constructed by realizing information manipulation or processing according to the use of the computer 2200.

[0072] For example, when communication is performed between the computer 2200 and an external device, the CPU 2212 may execute a communication program loaded in the RAM 2214 and instruct the communication interface 2222 to perform communication processing based on the processing described in the communication program. Under the control of the CPU 2212, the communication interface 2222 reads transmission data stored in a transmission buffer processing area provided in the RAM 2214, the hard disk drive 2224, the DVD-ROM 2201, or a recording medium such as an IC card, and transmits the read transmission data to the network, or writes reception data received from the network to a reception buffer processing area or the like provided on the recording medium.

[0073] Furthermore, the CPU 2212 may cause all or a necessary portion of a file or database stored in an external recording medium such as the hard disk drive 2224, the DVD-ROM drive 2226 (DVD-ROM 2201), an IC card, etc. to be read into the RAM 2214, and perform various types of processing on the data on the RAM 2214. The CPU 2212 then writes back the processed data to the external recording medium.

[0074] Various types of information, such as various types of programs, data, tables, and databases, may be stored in the recording medium and undergo information processing. The CPU 2212 may perform various types of processing on the data read from the RAM 2214, including various types of operations, information processing, conditional judgment, conditional branching, unconditional branching, information search / replacement, etc., as described throughout this disclosure and specified by the instruction sequence of the program, and write back the results to the RAM 2214. The CPU 2212 may also search for information in a file, database, etc. in the recording medium. For example, when a plurality of entries each having an attribute value of a first attribute associated with an attribute value of a second attribute are stored in the recording medium, the CPU 2212 may search for an entry that matches a condition, in which an attribute value of the first attribute is specified, from among the plurality of entries, read the attribute value of the second attribute stored in the entry, and thereby obtain the attribute value of the second attribute associated with the first attribute that satisfies a predetermined condition.

[0075] The above-described programs or software modules may be stored in a computer-readable medium on the computer 2200 or in the vicinity of the computer 2200. Also, a recording medium such as a hard disk or a RAM provided in a server system connected to a dedicated communication network or the Internet can be used as a computer-readable medium, thereby providing the programs to the computer 2200 via the network.

[0076] Although the present invention has been described above using the embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It is clear to those skilled in the art that various modifications and improvements can be made to the above embodiments. It is clear from the description of the claims that such modifications and improvements can also be included in the technical scope of the present invention.

[0077] It should be noted that the order of execution of each process, such as operations, procedures, steps, and stages, in the devices, systems, programs, and methods shown in the claims, specifications, and drawings is not specifically stated as "before," "prior to," etc., and may be realized in any order unless the output of a previous process is used in a later process. Even if the operational flow in the claims, specifications, and drawings is explained using "first," "next," etc. for convenience, it does not mean that it is essential to perform the process in this order. [Explanation of symbols]

[0078] 10. System 20 Plant 25 Equipment 100 Processing equipment 110 Acquisition Department 120 Model Generation Unit 130 Second Calculation Section 140 Model Memory Section 150 Control section 160 Communications Department 162 Receiving section 164 Request Section 170 First Calculation Section 180 Data Generation Division 190 Transmission model generation unit 2200 Computer 2201 DVD-ROM 2210 Host Controller 2212 CPU 2214 RAM 2216 Graphics Controller 2218 Display Device 2220 Input / Output Controller 2222 Communication Interface 2224 Hard Disk Drive 2226 DVD-ROM drive 2230 ROM 2240 Input / Output Chip 2242 Keyboard

Claims

1. a receiving unit that receives model parameters related to a first model generated by learning so as to output an output value corresponding to a state value in the first plant; a data generator for generating a data set of state values ​​and output values ​​using the first model in response to the received model parameters; a model generation unit that generates, by learning, a second model that outputs an output value corresponding to a state value in a second plant by using the data set as learning data. Device.

2. The model generation unit generates the second model by using a data set of state values ​​and output values ​​acquired in the second plant and a data set of state values ​​and output values ​​generated from the first model as learning data.

2. The apparatus of claim 1.

3. The model generation unit generates the second model by machine learning by weighting a data set of state values ​​and output values ​​acquired in the second plant more heavily than a data set of state values ​​and output values ​​generated from the first model.

3. The apparatus of claim 2.

4. The model generation unit generates the second model by performing machine learning on a data set of state values ​​and output values ​​acquired in the second plant at a learning rate greater than a learning rate of the data set of state values ​​and output values ​​generated from the first model.

4. The apparatus of claim 3.

5. a first calculation unit that calculates an evaluation value of the first model when a state value acquired in the second plant is input, The model generation unit generates the second model without using the data set as learning data when the evaluation value of the first model is equal to or less than a threshold.

2. The apparatus of claim 1.

6. a request unit that requests transmission of model parameters for a third model generated by learning so as to output an output value corresponding to a state value in a third plant when the evaluation value of the first model is equal to or less than a threshold value; The receiving unit receives the model parameters transmitted in response to the request.

6. The apparatus of claim 5.

7. a second calculation unit that calculates an evaluation value of the second model generated by the model generation unit; a request unit that requests transmission of new model parameters for the first model when the evaluation value of the second model is equal to or less than a threshold value; The receiving unit receives the model parameters transmitted in response to the request.

2. The apparatus of claim 1.

8. The first model has a smaller amount of model parameters than a model used to output an output value according to a state value in the first plant.

2. The apparatus of claim 1.

9. The second plant is different from the first plant in at least one of its configuration or location.

2. The apparatus of claim 1.

10. receiving model parameters for a first model generated by learning so as to output an output value corresponding to a state value in a first plant; generating a data set of state and output values ​​using the first model responsive to the received model parameters; and generating, by learning, a second model that outputs an output value corresponding to a state value in a second plant by using the data set as learning data. method.

11. The method is executed by a computer, causing the computer to: a receiving unit that receives model parameters related to a first model generated by learning so as to output an output value corresponding to a state value in the first plant; a data generator for generating a data set of state values ​​and output values ​​using the first model in response to the received model parameters; The data set is used as learning data to function as a model generating unit that generates, by learning, a second model that outputs an output value corresponding to a state value in a second plant. program.

Images