JP7304401B1 - Support device, support method and support program - Google Patents

Abstract

Kind Code: A1 It is possible to easily switch to an optimal machine learning model that is suitable for each mode of plant operation. A storage unit (14) stores a correspondence model (14a) that is a correspondence between a mode of an object and a model. The acquisition unit 15a acquires data. The determination unit 15c determines the mode of the object using the acquired data. The identifying unit 15d identifies a model corresponding to the determined object mode in the corresponding model 14a. [Selection drawing] Fig. 2

Description

The present invention relates to a support device, a support method, and a support program.

Conventionally, plant operation has various modes other than normal operation mode, such as raw material interchange mode, brand change mode, rain mode, and fine weather mode. In each mode, machine learning models are used for driving assistance and quality prediction.

Japanese Patent No. 6535130

However, it is not easy to use the optimal machine learning model for each mode. For example, if the same machine learning model is used for operation support, quality prediction, etc. without considering the operation mode of the plant, the estimation accuracy will be degraded. In addition, if you manually switch to a machine learning model that matches the mode, there is a risk of forgetting to switch, a risk of switching to an inappropriate machine learning model, an increase in the time and effort required for model switching, and an increase in the time required for switching by the person in charge of switching. There are issues such as differences. In addition, there is a problem that it takes time to judge the validity of model switching.

The present invention has been made in view of the above, and an object of the present invention is to facilitate switching to an optimum machine learning model that is suitable for each mode of plant operation or the like.

In order to solve the above-described problems and achieve the object, an assisting apparatus according to the present invention provides a storage unit that stores correspondences between modes of objects and models; an acquisition unit that acquires data; and a determining unit that determines a mode of the object using the data that has been obtained, and a specifying unit that specifies a model corresponding to the determined mode of the object in the association. do.

According to the present invention, it is possible to easily switch to the optimum machine learning model for each mode of plant operation or the like.

FIG. 1 is a diagram for explaining an outline of a support device. FIG. 2 is a schematic diagram illustrating the schematic configuration of the support device. FIG. 3 is a diagram illustrating the data configuration of rules. FIG. 4 is a diagram for explaining the processing of the generating unit; FIG. 5 is a diagram for explaining the processing of the clustering unit. FIG. 6 is a diagram for explaining the processing of the presentation unit; FIG. 7 is a diagram for explaining the processing of the presentation unit; FIG. 8 is a diagram for explaining the processing of the presentation unit; FIG. 9 is a diagram for explaining the processing of the presentation unit; FIG. 10 is a flow chart illustrating an assistance processing procedure. FIG. 11 is a flowchart illustrating a support processing procedure. FIG. 12 is a diagram exemplifying a computer that executes a support program.

An embodiment of the present invention will be described in detail below with reference to the drawings. It should be noted that the present invention is not limited by this embodiment. Moreover, in the description of the drawings, the same parts are denoted by the same reference numerals.

[Overview of support device]
FIG. 1 is a diagram for explaining an outline of a support device. The operation of the plant includes various modes other than the normal operation mode, such as a raw material interchange mode, a brand change mode, a rain mode, and a fine weather mode. As illustrated in FIG. 1, in a plant, a machine learning model learned using operation data of an actual plant is applied for operation support and quality prediction. In this machine learning model, when the plant operation mode changes, unless the algorithm, preprocessing, parameters related to learning, or learning data used for learning are switched appropriately, the operation will continue based on the wrong operation method. there is a risk that

Therefore, the support device of the present embodiment determines, for example, the operation mode of the actual plant data as the mode of the object, identifies an appropriate model, automatically switches the model, or allows the operator to manually switch the model. support the switching of In addition, the support device presents materials for judging the appropriateness of model switching.

[Configuration of support device]
FIG. 2 is a schematic diagram illustrating the schematic configuration of the support device. As illustrated in FIG. 2 , the support device 10 is implemented by a general-purpose computer such as a personal computer, and includes an input unit 11 , an output unit 12 , a communication control unit 13 , a storage unit 14 and a control unit 15 .

The input unit 11 is implemented using an input device such as a keyboard and a mouse, and inputs various instruction information such as processing start to the control unit 15 in response to input operations by the operator. The output unit 12 is implemented by a display device such as a liquid crystal display, a printing device such as a printer, or the like.

The communication control unit 13 is implemented by a NIC (Network Interface Card) or the like, and controls communication between an external device and the control unit 15 via a network. For example, the communication control unit 13 controls communication between the control unit 15 and a sensor that outputs data to be processed in support processing described later, a management device that manages data, and the like.

The storage unit 14 is implemented by a semiconductor memory device such as a RAM (Random Access Memory) or a flash memory, or a storage device such as a hard disk or an optical disk. In the storage unit 14, a processing program for operating the support device 10, data used during execution of the processing program, and the like are stored in advance, or are temporarily stored each time processing is performed.

In this embodiment, the storage unit 14 stores correspondence models 14a, rules 14b, and the like. Further, the storage unit 14 stores a classifier 14c generated in support processing described later, parameters of a classification algorithm, and the like. Note that the storage unit 14 may be configured to communicate with the control unit 15 via the communication control unit 13 .

The correspondence model 14a indicates the correspondence between the mode of the object and the model. In this embodiment, the mode of the object is, for example, the operating mode of the plant. Examples of operation modes include a normal operation mode, a raw material interchange mode, a brand change mode, a rain mode, a fine weather mode, and the like. The correspondence model 14a is, for example, information in which each operation mode, information for identifying a model applied in each operation mode, and parameters of the model are associated with each other.

The rule 14b is preset information indicating conditions for judging the mode of the object . Here, FIG. 3 is a diagram illustrating the data configuration of rules. As illustrated in FIG. 3, in the rule 14b, models and application conditions are associated with each other. The application condition is a condition for judging the operation mode. FIG. 3 shows, for example, that the model with model ID=1 is applied to the operation mode under the condition of "product A and furnace temperature <400 degrees".

Returning to the description of FIG. The control unit 15 is implemented using a CPU (Central Processing Unit) or the like, and executes a processing program stored in a memory. Thereby, the control unit 15 functions as an acquisition unit 15a, a generation unit 15b, a determination unit 15c, a specification unit 15d, a classification unit 15e, a presentation unit 15f, and a learning unit 15g, as illustrated in FIG. Each or part of these functional units may be implemented in different hardware. For example, the determination unit 15c and the classification unit 15e may be implemented in separate devices. Also, the control unit 15 may include other functional units.

The acquisition unit 15a acquires data. For example, the acquisition unit 15a communicates time-series data such as sensor values of a plant targeted for support processing, which will be described later, via the input unit 11, from a sensor, or from a management device that manages sensor values. Acquired via the control unit 13 .

The generation unit 15b uses past data to generate a classifier 14c that classifies objects by mode through learning. Here, FIG. 4 is a diagram for explaining the processing of the generation unit. As shown in FIG. 4, the generation unit 15b generates the classifier 14c for identifying the driving mode by learning, using the learning data to which the tag of the driving mode is attached in advance. For example, the generation unit 15b generates the classifier 14c using a machine learning algorithm such as SVM/Random Rorest/deep learning. In the example shown in FIG. 4, a tag indicating the driving mode is added to the past data to form learning data.

Returning to the description of FIG. The determination unit 15c determines the mode of the object using the acquired data. For example, the determination unit 15c determines the driving mode using the rule 14b. Specifically, the determination unit 15c determines which application condition of the rule 14b illustrated in FIG. 3 is met by the acquired actual data.

Alternatively, the determination unit 15c determines the mode of the object using the generated classifier 14c. In this case, instead of using the rule 14b, a classifier 14c trained on past data is used to determine the mode of the object . For example, the determination unit 15c determines the driving mode indicated by the tag by estimating the tag of the actual data using the learned classifier 14c. This eliminates the need to manually set the complicated rule 14b for determining the driving mode, making it possible to easily determine the driving mode.

The specifying unit 15d specifies a model corresponding to the determined mode of the object in the corresponding model 14a. For example, when the determination unit 15c determines that the operation mode matches the application condition “product A and furnace temperature <400 degrees” of the rule 14b illustrated in FIG. do.

Note that the support device 10 may have a clustering unit 15e. The clustering unit 15e clusters data into clusters. Here, FIG. 5 is a diagram for explaining the processing of the clustering unit. For example, the clustering unit 15e performs clustering using a machine learning algorithm such as k-means/k-NN, as shown in FIG. Note that in the example shown in FIG. 5, the data is clustered by the two-dimensional feature amounts of output and temperature. However, the dimension of clustering is not limited to two dimensions, and may be three or more dimensions.

In this case, the correspondence model 14a indicates the correspondence between clusters of clustered past data and the model instead of the correspondence between the driving mode and the model described above. That is, the clustering unit 15e learns a clustering algorithm using past data. Also, the correspondence model 14a indicates correspondence between each cluster and a model applied to the past data of each cluster.

Then, the clustering unit 15e identifies to which cluster the actual plant data belongs using the learned algorithm. Further, the specifying unit 15d specifies a model corresponding to the cluster to which the acquired data belongs in the corresponding model 14a.

In this case, it is possible to save the trouble of attaching a tag indicating the mode of the object such as the driving mode to the learning data. In addition, it is possible to classify only the features inherent in the data without relying on human experience, and to specify the optimum model for each class.

Returning to the description of FIG. The presentation unit 15f presents the identified model as a candidate for switching from the current model being applied. In addition, the presenting unit 15f presents information that serves as a basis for switching to assist the operator in model switching. Here, FIGS. 6 to 9 are diagrams for explaining the processing of the presentation unit.

Specifically, the presentation unit 15f presents an inference result obtained by applying the specified model to past data for a predetermined period. For example, the presentation unit 15f applies the model specified by the specifying unit 15d to data for 5 hours retroactively, as indicated by a dashed line in a region a surrounded by a thick frame in FIG. Present the results.

6, the presenting unit 15f also displays the inference result output by the current model being applied and the inference result output by the current model being applied, in addition to the inference result by the specified model, that is, the switching candidate model. Data are presented. This allows the operator to easily compare the current model and the switching candidate model.

Note that a model selection dialog is displayed in FIG. All or part of a plurality of models set in the corresponding model 14a are displayed in this model selection dialog. For example, when the determination unit 15c determines the driving mode of the actual data using the classifier 14c, or when the clustering unit 15e specifies to which cluster the actual data belongs, the presentation unit 15f Multiple models may be displayed in order of likelihood of matching. When the operator selects one of the displayed models, the presentation unit 15f presents the inference result of the selected model. This allows the operator to easily compare multiple switching candidates with the current model.

In addition, the presentation unit 15f presents a change from the inference result of the current model during application of the inference result of the specified model. For example, the presentation unit 15f performs factor calculation for each parameter of each model, as shown in FIG. Here, the factor calculation result (A, T, p2, t) is the degree of contribution of the parameter p2 to the current time T from the past time t when the model A is applied. The presentation unit 15f finds the difference between the factor calculation results at time T between the switching candidate model and the current model, and extracts a portion exceeding the threshold k.

For example, in the part enclosed by a thick frame in FIG. exceeds the threshold k.

Then, the presentation unit 15f highlights the extracted portion as illustrated in FIG. Although the method of highlighting is not particularly limited, for example, it may be blinked, displayed with a thick line, or changed in color. Inference results from a plurality of switching candidate models may be displayed. In that case, when the operator selects any inference result graph, the factors for the selected model are displayed. This allows the operator to confirm empirically whether switching between the current model and the plurality of switching candidates is appropriate.

In addition, the presentation unit 15f presents an alert when the specified model deviates from a predetermined applicable range. For example, the presentation unit 15f calculates the guarantee range of the model before applying the model or periodically. Guaranteed bounds are defined for inference results or explanatory variables. For example, using specific tags used as explanatory variables, the mean and standard deviation σ of specific tags in the learning data are calculated and defined as in the following equation (1). It should be noted that it may be extended to multiple variables by targeting multiple tags.

(Average tag value during learning - 2σ) < Tag value during inference < (Average tag value during learning + 2σ) (1)

Then, as shown in FIG. 9, when the inference result deviates from a predetermined guaranteed range, the presenting unit 15f displays an alert such as "outside the guaranteed range of the model." In that case, the presentation unit 15f may display a message that recommends switching models. Also, the presentation unit 15f may output a warning sound instead of displaying the message or in addition to displaying the message. In addition, the presentation unit 15f may display a plurality of models within the guarantee range in descending order of probability of matching the actual data.

The learning unit 15g learns the model using the acquired data when the inference result of the specified model is within a predetermined range. That is, when the inference result by the identified model is within a predetermined guaranteed range, the learning unit 15g adds the newly acquired data to the past data to search for past data close to the current conditions. Re-learning is performed by using the data as learning data. This improves the accuracy of the default model.

[Support processing procedure]
Next, an example of support processing by the support device 10 according to the present embodiment will be described with reference to FIGS. 10 and 11. FIG. 10 and 11 are flowcharts illustrating the support processing procedure. First, the flowchart of FIG. 10 is started, for example, at the timing when there is an input instructing the start of the support process.

First, the acquisition unit 15a acquires data (step S1). For example, the acquisition unit 15a acquires time-series data such as sensor values targeted for support processing.

Next, the determination unit 15c determines the driving mode using the acquired data (step S2). For example, the determination unit 15c determines the driving mode using the rule 14b. Alternatively, the determination unit 15c determines the driving mode using the generated classifier 14c.

Then, the identifying unit 15d identifies a model corresponding to the determined driving mode among the corresponding models 14a (step S3).

Alternatively, the clustering unit 15e identifies which cluster the data belongs to using a learned algorithm, and the identifying unit 15d identifies a model corresponding to the cluster to which the acquired data belongs in the correspondence model 14a (steps S2 to S3). This completes the series of support processes.

Next, the flowchart of FIG. 11 is started, for example, at the timing when an instruction to start processing is received. First, a specified model is input (step S11). For example, the process of step S11 may be started following the process of step S3 shown in FIG.

Next, the acquisition unit 15a acquires the current data (step S12). Then, if the specified model is more suitable than the current model (step S13, Yes), the presenting unit 15f presents information that serves as a basis for model switching (steps S14 and S15).

For example, the presentation unit 15f performs factor calculation for each parameter of the current model and the switching candidate model, and highlights a portion where the difference between the factor calculation results exceeds the threshold (step S14).

In addition, the presenting unit 15f presents an inference result outputted by retroactively applying the switching candidate model to data for a predetermined period of time (step S15).

Then, the presentation unit 15f confirms with the operator whether or not the model should be switched (step S16). When receiving a model switching instruction from the operator (step S17), the support device 10 switches the applicable model from the current model to the switching candidate model (step S18).

Note that the support device 10 may automatically switch to the identified model when the identified model is more suitable than the current model (step S13, Yes). In that case, the support device 10 may omit the processing of steps S14 to S18 and proceed to step S19.

If switching from the current model is unnecessary (step S13, No), the process proceeds to step S19.

In the process of step S19, an inference result based on the applied model is output (step S19). This completes the series of support processes.

[effect]
As described above, in the support device 10 of the above-described embodiment, the storage unit 14 stores the correspondence model 14a, which is the correspondence between the mode of the object and the model. The acquisition unit 15a acquires data. The determination unit 15c determines the mode of the object using the acquired data. The specifying unit 15d specifies a model corresponding to the determined mode of the object in the corresponding model 14a.

For example, the storage unit 14 stores a rule 14b indicating conditions for determining the mode of the object , and the determination unit 15c uses the rule 14b to determine the mode of the object . This makes it possible to easily switch to the optimum machine learning model for each mode of plant operation.

Alternatively, the generation unit 15b uses past data to generate a classifier 14c that classifies the object by mode by learning, and the determination unit 15c uses the generated classifier 14c to classify the object mode. This eliminates the need to manually set the complicated rule 14b for determining the mode of the object such as the driving mode, and makes it possible to easily determine the mode of the object .

Alternatively, the acquisition unit 15a acquires data. The clustering unit 15e clusters data into clusters. The storage unit 14 stores a correspondence model 14a that is a correspondence between clusters of past data clustered and models. The identifying unit 15d identifies a model corresponding to the cluster to which the acquired data belongs in the corresponding model 14a.

As a result, it is possible to save the trouble of attaching a tag indicating the mode of the object to the learning data. In addition, it is possible to classify only the features inherent in the data without relying on human experience, and to specify the optimum model for each class.

In addition, the presentation unit 15f presents the identified model as a candidate for switching from the currently applied model. This allows the operator to select whether or not switching is possible.

In addition, the presentation unit 15f presents an inference result obtained by applying the specified model to past data for a predetermined period. This makes it possible to easily compare the current model and the switching candidate model.

In addition, the presenting unit 15f presents changes in the inference result of the specified model from the inference result of the model being applied. This enables the operator to confirm whether or not the switching between the switching candidate and the current model is experientially appropriate.

In addition, the presentation unit 15f presents an alert when the specified model deviates from a predetermined applicable range. This makes it possible to prevent missing the timing of model switching.

Further, when the inference result of the identified model is within a predetermined range, the learning unit 15g learns the model using the acquired data. This makes it possible to improve the accuracy of the default model.

[System configuration, etc.]
Each component of each device illustrated is functionally conceptual, and does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution and integration of each device is not limited to the one shown in the figure, and all or part of them can be functionally or physically distributed and integrated in arbitrary units according to various loads and usage conditions. Can be integrated and configured. Furthermore, all or any part of each processing function performed by each device is realized by a CPU or GPU and a program analyzed and executed by the CPU or GPU, or as hardware by wired logic can be realized.

Further, among the processes described in this embodiment, all or part of the processes described as being automatically performed can be performed manually, or the processes described as being performed manually can be performed manually. All or part of this can also be done automatically by known methods. In addition, information including processing procedures, control procedures, specific names, and various data and parameters shown in the above documents and drawings can be arbitrarily changed unless otherwise specified.

[program]
It is also possible to create a program in which the processing executed by the support device described in the above embodiment is written in a computer-executable language. For example, it is possible to create a program in which the processing executed by the support device 10 according to the embodiment is described in a computer-executable language. In this case, the same effects as those of the above embodiments can be obtained by having the computer execute the program. Further, such a program may be recorded in a computer-readable recording medium, and the program recorded in this recording medium may be read by a computer and executed to realize processing similar to that of the above embodiments.

FIG. 12 is a diagram illustrating an example of a computer that executes a support program; Computer 1000 includes, for example, memory 1010 , CPU 1020 , hard disk drive interface 1030 , disk drive interface 1040 , serial port interface 1050 , video adapter 1060 and network interface 1070 . These units are connected by a bus 1080 .

The memory 1010 includes a ROM (Read Only Memory) 1011 and a RAM 1012 . The ROM 1011 stores a boot program such as BIOS (Basic Input Output System). Hard disk drive interface 1030 is connected to hard disk drive 1031 . Disk drive interface 1040 is connected to disk drive 1041 . A removable storage medium such as a magnetic disk or an optical disk is inserted into the disk drive 1041, for example. A mouse 1051 and a keyboard 1052 are connected to the serial port interface 1050, for example. For example, a display 1061 is connected to the video adapter 1060 .

Here, the hard disk drive 1031 stores an OS 1091, application programs 1092, program modules 1093 and program data 1094, for example. Each piece of information described in the above embodiment is stored in the hard disk drive 1031 or the memory 1010, for example.

Also, the support program is stored in hard disk drive 1031 as, for example, program module 1093 in which instructions to be executed by computer 1000 are described. Specifically, the hard disk drive 1031 stores a program module 1093 that describes each process executed by the support device 10 described in the above embodiment.

Data used for information processing by the support program is stored as program data 1094 in the hard disk drive 1031, for example. Then, the CPU 1020 reads out the program module 1093 and the program data 1094 stored in the hard disk drive 1031 to the RAM 1012 as necessary, and executes each procedure described above.

Note that the program module 1093 and program data 1094 related to the support program are not limited to being stored in the hard disk drive 1031. For example, they are stored in a removable storage medium and read by the CPU 1020 via the disk drive 1041 or the like. may be Alternatively, the program module 1093 and program data 1094 related to the support program are stored in another computer connected via a network such as a LAN (Local Area Network) or WAN (Wide Area Network), and are transmitted via the network interface 1070. It may be read by CPU 1020 .

Although the embodiments to which the invention made by the present inventor is applied have been described above, the present invention is not limited by the descriptions and drawings forming a part of the disclosure of the present invention according to the embodiments. That is, other embodiments, examples, operation techniques, etc. made by those skilled in the art based on this embodiment are all included in the scope of the present invention.

10 support device 11 input unit 12 output unit 13 communication control unit 14 storage unit 14a correspondence model 14b rule 14c classifier 15 control unit 15a acquisition unit 15b generation unit 15c determination unit 15d identification unit 15e clustering unit 15f presentation unit 15g learning unit

Claims (9)

a storage unit that stores the correspondence between the mode of the object and the model;
an acquisition unit that acquires data;
a determination unit that determines the mode of the object using the acquired data;
an identifying unit that identifies a model corresponding to the determined mode of the object in the association;
a presentation unit that presents the identified model as a candidate for switching from the model being applied;
has
For each parameter of the identified model and the model being applied, the presentation unit selects a location where a difference in factor calculation results representing the degree of contribution from the past time to the current time of the parameter exceeds a predetermined threshold. is extracted, and the extracted part is highlighted . The storage unit stores rules indicating conditions for determining the mode of the object,
The support device according to claim 1, wherein the determination unit determines the mode of the object using the rule. further comprising a generation unit that uses past data to generate, through learning, a classifier that classifies the object by mode;
The support device according to claim 1, wherein the determination unit determines the mode of the object using the generated classifier. an acquisition unit that acquires data;
a clustering unit that clusters data into clusters;
a storage unit that stores associations between clusters of clustered past data and models;
an identifying unit that identifies a model corresponding to a cluster to which the acquired data belongs in the association;
a presentation unit that presents the identified model as a candidate for switching from the model being applied;
has
For each parameter of the identified model and the model being applied, the presentation unit selects a location where a difference in factor calculation results representing the degree of contribution from the past time to the current time of the parameter exceeds a predetermined threshold. is extracted, and the extracted part is highlighted . 5. The support device according to claim 1, wherein the presenting unit presents an inference result obtained by applying the identified model to past data for a predetermined period. 5. The support device according to claim 1, wherein the presenting unit presents an alert when the specified model deviates from a predetermined applicable range. 5. The method according to claim 1, further comprising a learning unit that performs learning of the model by adding newly acquired data when past inference results by the specified model are within a predetermined range. A support device as described in . A support method executed by a support device,
The support device has a storage unit that stores a correspondence between the mode of the object and the model,
an acquisition step of acquiring data;
a determination unit that determines the mode of the object using the acquired data;
an identifying step of identifying a model corresponding to the determined mode of the object in the matching;
a presentation step of presenting the identified model as a candidate for switching from the model being applied;
including
In the presenting step, for each parameter of the identified model and the model being applied, a point where a difference in a factor calculation result representing the degree of contribution of the parameter from the past time to the current time exceeds a predetermined threshold and highlighting the extracted portion . referring to a storage unit that stores the correspondence between the mode of the object and the model;
an acquisition step for acquiring data;
a determination step of determining the mode of the object using the acquired data;
an identifying step of identifying a model corresponding to the determined mode of the object in the matching;
a presenting step of presenting the identified model as a candidate for switching from the model being applied;
on the computer, and
In the presenting step, for each parameter of the identified model and the model being applied, the point where the difference in the factor calculation result representing the contribution of the parameter from the past time to the current time exceeds a predetermined threshold and highlighting the extracted part .

Images