JP7064075B2 - Plant learning support device and plant learning support method - Google Patents

Description

The present disclosure relates to support for learning (machine learning) in a plant that can be controlled using artificial intelligence.

Conventionally, a method of controlling a plant by using a machine learning method has been known. For example, Patent Document 1 discloses that machine learning is used to support the operation of a plant. More specifically, the operation support system generates past intervention history information in which the history of the operating state of the equipment (equipment) operating in the plant is associated with the history of the intervention operation by the operator for the equipment, and the intervention operation is performed. Rule information including the conditions for performing the above is learned and generated by a machine learning method or the like. Then, the intervention operation is determined based on the acquired device operation status information and the above rule information, and the operation instruction information is output. In addition, the intervention operation to the plant differs depending on the individual plant due to the difference in the environment of the installation location of the equipment and the individual difference, but if the operation has just started and the history is not accumulated, other operations are performed. Since the intervention operation performed in a plant of the same type may be used as a reference, it is also disclosed that the operation instruction information derived from the rule information of another plant of the same type is output as reference information.

Further, Patent Document 2 corresponds to observation data having a time related to the operating state (operating state) of the plant as an attribute from a plurality of sensors installed in the plant (equipment), and almost normal data in the normal operating state of the plant. It is disclosed that the abnormality detection of the plant is performed by using the learning data to be performed.

In Patent Document 3, the intermediate learning devices provided in each of the plurality of learning device units, which are environmental control terminals for chemical plants and facility horticulture, have the same internal state (weight) among all the learning device units. It is disclosed that each learning device unit communicates information (weight) constituting an internal state with each other at a constant frequency.

Japanese Unexamined Patent Publication No. 2017-49801 Japanese Unexamined Patent Publication No. 2014-149840 Japanese Unexamined Patent Publication No. 2017-4142

The accuracy of plant control through machine learning is expected to improve as learning is performed using various learning data. However, in Patent Documents 1 and 2, each plant learns based on the operation data of its own plant. Since it is done, it takes time to obtain sufficient accuracy. Further, in Patent Documents 1 and 2, there is a possibility that an abnormality (problem) that occurs for the first time in each plant cannot be detected at an early stage even if the results of machine learning up to that point are used. Further, as described in Patent Document 1, in order to optimize the operation of a plant through machine learning, it is necessary to perform learning for each plant.

In view of the above circumstances, at least one embodiment of the present invention aims to provide a learning support device for a plant for supporting effective learning in the plant.

(1) The learning support device of the plant according to at least one embodiment of the present invention is
It is a plant learning support device for supporting the exchange of learning data between multiple plants that can be controlled using artificial intelligence.
An operation data receiving unit configured to receive the operation data transmitted from the operation data providing plant that provides the operation data of the plant, and an operation data receiving unit.
A learning data generation unit configured to generate the learning data based on the operation data received by the operation data receiving unit, and a learning data generation unit.
A learning data storage unit configured to store the learning data,
A management database configured to manage management information including the destination of the learning data, and
A learning data transmission unit configured to transmit the learning data stored in the learning data storage unit to at least one of the plurality of plants based on the destination managed by the management database. And prepare.

According to the configuration of (1) above, the learning support device of the plant generates learning data based on the received operation data of the plant, and transmits the generated learning data to another plant. As a result, the plant that acquired the training data (learning data receiving plant) can acquire various learning data earlier, so that the accuracy of, for example, optimized operation and abnormality detection can be improved earlier. can. For example, if learning data generated based on operation data (operation data at the time of abnormality) such as when a sign of abnormality appears or an abnormality occurs in a plant is acquired, each plant does not occur in its own plant. Since it is possible to learn in advance even for a certain abnormality, even an abnormality that has not occurred can be detected at an early stage.

Therefore, it is possible to prevent the abnormality from becoming serious (serious) and improve the reliability of the plant. In addition, by providing the learning data via the learning support device of the plant, the source of the operation data on which the learning data is based is not known to the learning data receiving plant, and the concealment processing described later is performed. , It is also possible to add value such as generation of teacher data (learning data) to the learning data L. In this way, by supporting the sharing of learning data among a plurality of plants, effective learning in each plant can be supported.

(2) In some embodiments, in the configuration of (1) above,
The learning data generation unit generates the learning data by performing a concealment process for concealing the time information of the operation data.

Traditionally, plant operation data is closely related to management information, so plant owners may provide operation data to plant manufacturers for abnormality analysis, etc., but operation data There is a strong sense of resistance to providing the data as learning data. It is even more difficult to provide operational data to plants owned by others.
According to the configuration of (2) above, the time information in the operation data, that is, the time data (log time) and the time series for associating the control contents and the operation state of the plant with the time are concealed on the learning data. Can be in a state of being. As a result, it is possible to eliminate or reduce factors that make it difficult to tolerate the provision of operation data closely related to management information. Therefore, since the owner of the plant can alleviate the resistance to the provision of the operation data of the own plant, it is possible to promote the increase of the number of plants that allow the provision of the operation data.

(3) In some embodiments, in the configuration of (2) above,
The concealment process is an encryption process for encrypting the time data of the operation data.
The learning data generation unit
An encryption unit that executes the encryption process and generates a decryption key for decrypting the encrypted time data.
A decryption key return unit that transmits the decryption key to the operation data providing plant that is the source of the learning data to which the encryption process has been performed, and a decryption key return unit.
Upon receiving the decryption key from the operation data providing plant, a decoding unit that decodes the encrypted time data using the decryption key, and a decoding unit.
It has an operation data transfer unit that transfers restored operation data, which is the learning data including the time data decoded by the decoding unit, to a predetermined transfer destination.

According to the configuration of (3) above, as a concealment process, the time information of the operation data can be concealed by encrypting the time data of the operation data. Further, the decryption key for decrypting the encrypted time data is returned to the operation data providing plant that provided the operation data on which the learning data is based. As a result, only the source of the operation data manages the decryption key, and the confidentiality of the operation data can be enhanced.

Further, the operation data restored by receiving the decryption key from the operation data providing plant is configured to be transferred (transmitted) to a transfer destination such as a manufacturer. As a result, for example, when a problem (trouble) such as an abnormality or a sign thereof is detected, the operation data providing plant sends a decoding key to the learning support device of the plant, and learning that the decoding key is received. If the support device transmits the restored operation data to the manufacturer of the plant or the like, the investigation of the problem on the manufacturer side can be started promptly, and the problem can be solved promptly. Therefore, by providing an incentive for the provision of such operation data, it is possible to further promote the increase in the number of plants that allow the provision of operation data.

(4) In some embodiments, in the configuration of (2) above,
The concealment process includes a process of deleting the time data when the operation data has time data, and a process of rearranging the operation data when the operation data does not include the time data.
According to the configuration of (4) above, the time information of the operation data can be concealed while suppressing the processing load by deleting the time data or rearranging the operation data.

(5) In some embodiments, in the configurations (1) to (4) above,
The learning data generation unit generates the learning data by adding information as to whether or not the operation data includes the abnormal operation data acquired at the time of abnormality of the operation data providing plant to the operation data.
According to the configuration of (5) above, it is possible to efficiently select and learn abnormal operation data from a huge amount of operation data. As a result, each plant can more efficiently improve the abnormality detection accuracy by performing learning using the teacher data.

(6) In some embodiments, in the configurations (1) to (5) above,
The learning data receiving plant side, which is the plant that receives the learning data transmitted by the learning data transmitting unit, pays the management database the consideration to be paid for the reception of the learning data for each destination or the learning data. It also has a consideration registration unit that is configured to register each time.
At least a part of the consideration paid by the learning data receiving plant side is received by the operation data providing plant side.
According to the configuration of (6) above, the learning data receiving plant pays a consideration for receiving the learning data, and receives the learning data depending on whether or not the plant is an operation data providing plant. Consideration can be set individually for each plant or each training data. In addition, at least a part of the consideration paid by the learning data receiving plant is received by the operation data providing plant as the consideration for providing the operation data. That is, the operation data is bought and sold for a fee, and by providing an incentive for the provision of the operation data in this way, it is possible to promote the increase in the number of plants that allow the provision of the operation data.

(7) In some embodiments, in the configurations (1) to (6) above,
The learning data receiving plant side, which is the plant that receives the learning data transmitted by the learning data transmitting unit, pays the management database the consideration to be paid for the reception of the learning data for each destination or the learning data. It also has a consideration registration unit that is configured to register each time.
The consideration is set based on whether or not the learning data receiving plant is the operation data providing plant, and the case where the learning data receiving plant is the operation data providing plant is the case of the operation data. The price is set to be lower (including free of charge) than in the case of the plant that does not provide the operation data, which is the plant that does not provide the data.
According to the configuration of (7) above, by providing the operation data, the plant can acquire the learning data generated based on the operation data provided from another plant, for example, at a lower price. It becomes. In this way, by providing an incentive for the provision of operation data, it is possible to promote an increase in the number of plants that allow the provision of operation data.

(8) In some embodiments, in the configurations (6) to (7) above,
The consideration is set based on whether or not the learning data is abnormal training data generated based on the abnormal operation data including the abnormal operation data acquired at the time of abnormality of the operation data providing plant. Therefore, the consideration for the learning data is set to be higher in the case of the abnormal learning data than in the case of not the abnormal learning data.
According to the configuration of (8) above, the consideration for the learning data is set so that the learning data at the time of abnormality is higher than the case where it is not. In other words, since the learning data at the time of abnormality is more useful for improving the reliability of the plant, such as improving the accuracy of the abnormality judgment of the plant, the value of the learning data is better than the case where the learning data at the time of abnormality is not. Is also high, and this is reflected in the consideration. As a result, the consideration received by the operation data providing plant that provided the operation data that is the basis of the learning data at the time of abnormality can be made higher, so that the operation data providing plant can, for example, cost for solving the abnormality. It is possible to make more sufficient compensation. In this way, by setting the consideration according to the value of the learning data, it is possible to promote the increase of the number of plants that allow the provision of the operation data by providing the incentive for the provision of the learning data.

(9) In some embodiments, in the configurations (6) to (8) above,
Further, a consideration setting unit configured to set the consideration for the operation data providing plant that transmitted the received operation data based on the operation data received by the operation data receiving unit is further provided.
According to the configuration of (9) above, the consideration can be automatically set based on the value of the operation data.

(10) In some embodiments, in the configurations (1) to (9) above,
A member group is formed by the plurality of plants registered in the management database.
The plurality of plants that form the member group each pay a predetermined membership fee.
The membership fee paid by the operation data providing plant side is lower than the membership fee paid by the operation data non-providing plant side, which is the plant that does not provide the operation data.
According to the configuration of the above (10), the same effect as the above (7) can be obtained.

(11) The plant learning support method according to at least one embodiment of the present invention is
It is a plant learning support method for supporting the exchange of learning data between multiple plants that can be controlled using artificial intelligence.
An operation data receiving step for receiving the operation data transmitted from the operation data providing plant that provides the operation data of the plant, and an operation data receiving step.
A learning data generation step that generates the learning data based on the operation data received by the operation data receiving step, and
A learning data storage step for storing the learning data, and
A management database that manages management information including the destination of the learning data, and
A learning data transmission step of transmitting the learning data stored in the learning data storage step to at least one of the plurality of plants based on the destination managed by the management database. ..

According to the configuration of the above (11), the same effect as the above (1) can be obtained.

(12) In some embodiments, in the configuration of (11) above,
The learning data generation step generates the learning data by performing a concealment process for concealing the time information of the operation data.

According to the configuration of the above (12), the same effect as the above (2) can be obtained.

(13) In some embodiments, in the configuration of (12) above,
The concealment process is an encryption process for encrypting the time data of the operation data.
The learning data generation step is
A cryptographic step that executes the encryption process and generates a decryption key for decrypting the encrypted time data.
A decryption key return step of transmitting the decryption key to the operation data providing plant of the source of the learning data to which the encryption process has been performed, and a decryption key return step.
Upon receiving the decryption key from the operation data providing plant, a decryption step of decrypting the encrypted time data using the decryption key, and a decryption step.
It has an operation data transfer step of transferring restored operation data, which is the learning data including the time data decoded by the decoding step, to a predetermined transfer destination.

According to the configuration of the above (13), the same effect as the above (3) can be obtained.

(14) In some embodiments, in the configuration of (12) above,
The concealment process includes a process of deleting the time data when the operation data has time data, and a process of rearranging the operation data when the operation data does not include the time data.
According to the configuration of the above (14), the same effect as the above (4) can be obtained.

(15) In some embodiments, in the configurations (11) to (14) above,
The learning data generation step generates the learning data by adding information as to whether or not the operation data includes the abnormal operation data acquired at the time of abnormality of the operation data providing plant to the operation data.
According to the configuration of the above (15), the same effect as the above (5) can be obtained.

(16) In some embodiments, in the configurations (11) to (15) above,
A consideration registration step in which the learning data receiving plant side, which is the plant that receives the learning data transmitted by the learning data transmission step, registers the consideration to be paid for the reception of the learning data in the management database for each destination. To prepare further,
At least a part of the consideration paid by the learning data receiving plant side is received by the operation data providing plant side.
According to the configuration of the above (16), the same effect as the above (6) can be obtained.

(17) In some embodiments, in the configuration of (16) above,
The consideration is set based on whether or not the learning data receiving plant is the operation data providing plant, and the case where the learning data receiving plant is the operation data providing plant is the case of the operation data. It is set to be lower than the case of the plant that does not provide the operation data, which is the plant that does not provide the data.
According to the configuration of the above (17), the same effect as the above (7) can be obtained.

(18) In some embodiments, in the configurations (16) to (17) above,
The consideration is set based on whether or not the learning data is abnormal training data generated based on the abnormal operation data including the abnormal operation data acquired at the time of abnormality of the operation data providing plant. Therefore, the consideration for the learning data is set to be higher in the case of the abnormal learning data than in the case of not the abnormal learning data.
According to the configuration of the above (18), the same effect as the above (8) can be obtained.

(19) In some embodiments, in the configurations (16) to (18) above,
A consideration setting step for setting the consideration for the operation data providing plant that transmitted the received operation data based on the operation data received by the operation data reception step is further provided.
According to the configuration of the above (19), the same effect as the above (9) can be obtained.

(20) In some embodiments, in the configurations (11) to (19) above,
A member group is formed by the plurality of plants registered in the management database.
The plurality of plants that form the member group each pay a predetermined membership fee.
The membership fee paid by the operation data providing plant side is lower than the membership fee paid by the operation data non-providing plant side, which is the plant that does not provide the operation data.
According to the configuration of the above (20), the same effect as the above (10) can be obtained.

According to at least one embodiment of the present invention, a learning data providing support device for supporting effective learning in a plant is provided.

It is a figure which shows schematically the learning support system which concerns on one Embodiment of this invention. It is a figure explaining the learning support method of the plant which concerns on one Embodiment of this invention. It is a figure which shows the management database which concerns on one Embodiment of this invention. It is a figure which shows the management database which concerns on another embodiment of this invention. It is a figure which shows the structure of the learning data generation part which concerns on one Embodiment of this invention, and corresponds to the case where the encryption process is performed as a concealment process. It is a figure which shows the learning data generation step which concerns on one Embodiment of this invention, and corresponds to the case where the encryption process is performed as a concealment process. It is a figure which shows the learning support apparatus of the plant which concerns on one Embodiment of this invention, and shows the case where the learning data is traded between a plurality of plants for a fee. It is a figure which shows the learning support apparatus of the plant which concerns on one Embodiment of this invention, and shows the case where the learning data is traded for a fee within the member group of the membership fee system which consists of a plurality of plants. It is a figure which shows the learning support apparatus of the plant which concerns on one Embodiment of this invention, and shows the case where the learning data is traded free of charge in the member group of the membership fee system which consists of a plurality of plants.

Hereinafter, some embodiments of the present invention will be described with reference to the accompanying drawings. However, the dimensions, materials, shapes, relative arrangements, etc. of the components described as embodiments or shown in the drawings are not intended to limit the scope of the present invention to this, but are merely explanatory examples. not.
For example, expressions that represent relative or absolute arrangements such as "in one direction", "along a certain direction", "parallel", "orthogonal", "center", "concentric" or "coaxial" are exact. Not only does it represent such an arrangement, but it also represents a tolerance or a state of relative displacement at an angle or distance to the extent that the same function can be obtained.
For example, expressions such as "same", "equal", and "homogeneous" that indicate that things are in the same state not only represent exactly the same state, but also have tolerances or differences to the extent that the same function can be obtained. It shall also represent the existing state.
For example, the expression representing a shape such as a square shape or a cylindrical shape not only represents a shape such as a square shape or a cylindrical shape in a geometrically strict sense, but also an uneven portion or a chamfer within the range where the same effect can be obtained. It shall also represent the shape including the part and the like.
On the other hand, the expressions "equipped", "equipped", "equipped", "included", or "have" one component are not exclusive expressions excluding the existence of other components.

FIG. 1 is a diagram schematically showing a learning support system 9 according to an embodiment of the present invention. As shown in FIG. 1, the learning support system 9 includes a plurality of plants 8 and a learning support device 1 of the plant 8. Each of the plurality of plants 8 is a power plant, a garbage incinerator, etc., and various equipment necessary for achieving the purpose of the plant 8, such as a boiler, a turbine, an exhaust gas treatment device, a tank, a duct, a pipe, and a valve. have.

In this type of plant 8, for example, information such as a set value of each facility and an operating state of the plant 8 detected by using a sensor or the like is acquired and monitored periodically. Further, the information (log) acquired in this way is accumulated (stored) as the operation data D. That is, the operation data D is composed of one or more records including at least a log acquired at a certain timing. Further, each record has time data (time) indicating the acquisition timing and the like, and in the operation data D, acquisition of each log is performed from the time data and the order of the records by being sequentially stored. Time information such as time and time series can be grasped. Such operation data D is used for analysis of abnormalities occurring in the plant 8 and control for improving (optimizing) the operation of the plant 8.

Further, in the present invention, each plant 8 is provided with a plant control device (81c, 82c) which is a computer on which artificial intelligence such as an AI program is operated. This artificial intelligence executes machine learning based on the learning data L generated based on the operation data D, and the plant 8 has a plurality of controls based on the learning results (for example, control for operation optimization). Execute for the target equipment among the equipment of. The learning data L is generated, for example, based on the operation data D acquired by the operation of the own plant, or is provided from the learning support device 1 of the plant 8 as described later.

Further, in the present invention, as shown in FIG. 1, the plurality of plants 8 include at least one operation data providing plant 81 that provides (transmits) the operation data D of the plant 8. That is, the operation data providing plant 81 is a plant that is permitted (permitted) to provide the operation data D obtained as a result of the operation of the own plant to the outside. As will be described later, the provided operation data D is used to generate the training data L. Further, the plurality of plants 8 may include a plant 82 that does not provide operation data (prohibited) for which the provision of operation data D is not permitted (prohibited). The operation data non-providing plant 82 is a plant 8 that only receives (acquires) the learning data L without providing the operation data D to the outside. That is, the plant 8 constituting the learning support system 9 is classified into either an operation data providing plant 81 or an operation data non-providing plant 82.

Further, when the plant 8 that receives the training data L is referred to as a learning data receiving plant 83, both the operation data providing plant 81 and the operating data non-providing plant 82 correspond to the learning data receiving plant 83. However, the present invention is not limited to this embodiment. In some other embodiments, the training data receiving plant 83 may be configured such that only the operation data non-providing plant 82 is applicable. That is, the operation data providing plant 81 only provides the operation data D and does not receive the learning data L. In some other embodiments, only a part of the plurality of operation data providing plants 81 may be configured to correspond to the learning data receiving plant 83. In some embodiments, whether or not the learning data receiving plant 83 is applicable is determined by whether or not there is a possibility of receiving the learning data L. For example, learning is performed in the management database 5 described later. It may be determined whether or not the transmission possibility information Ig of the data L can be transmitted (see FIG. 3A described later). In some other embodiments, the management database 5 may be configured to determine whether or not the learning data management number In is registered as the transmittable learning data management information (see FIG. 3B described later).

In the embodiment shown in FIG. 1, the plurality of plants 8 of the learning support system 9 each have a power generation plant such as a power plant, and are composed of a plurality of operation data providing plants 81 and a plurality of operation data non-providing plants 82. .. Further, the plurality of plants 8 are communicably connected to the learning support device 1 of the plant 8 described below via the communication network 92 (IP network or the like). However, the present invention is not limited to this embodiment. In some other embodiments, all of the plurality of plants 8 may be the operation data providing plant 81.

Next, the learning support device 1 of the plant 8 will be described in detail with reference to FIGS. 1 to 3B. FIG. 2 is a diagram illustrating a learning support method for a plant according to an embodiment of the present invention. FIG. 3A is a diagram showing a management database 5 according to an embodiment of the present invention. Further, FIG. 3B is a diagram showing a management database 5 according to another embodiment of the present invention.

The learning support device 1 of the plant 8 (hereinafter, simply referred to as the learning support device 1) is a device for supporting the shared use of the learning data L among a plurality of plants 8 that can be controlled by using artificial intelligence. .. More specifically, the learning support device 1 generates learning data L based on the operation data D provided from the operation data providing plant 81, and mainly provides the generated learning data L to another plant 8. It is a device to support. Then, as shown in FIG. 1, the learning support device 1 includes an operation data receiving unit 2, a learning data generation unit 3, a learning data storage unit 4, a management database 5, and a learning data transmission unit 6. ..

The learning support device 1 is composed of a computer, and includes a CPU (processor) (not shown), a memory (storage device M) such as ROM and RAM, and a communication interface. Then, the CPU operates (data calculation, etc.) according to the instructions of the program (learning support program) loaded in the main storage device, thereby realizing each of the above functional units.
Each functional unit included in the learning support device 1 will be described.

The operation data receiving unit 2 is configured to receive the operation data D transmitted from the operation data providing plant 81 described above. More specifically, the operation data receiving unit 2 is connected to a communication device (not shown) included in the operation data providing plant 81 via a communication network 92, and this communication device is periodically or when an abnormality occurs. The operation data D to be transmitted at a predetermined timing such as is received. In the embodiment shown in FIG. 1, the communication device is, for example, a gateway device, and the operation data receiving unit 2 receives operation data D transmitted by the plant control device 81c via the communication device. ing.

The learning data generation unit 3 is configured to generate learning data L based on the operation data D received by the operation data reception unit 2. Specifically, in some embodiments, necessary information such as extracting features useful for control and abnormality detection (removal of noise information) from the received operation data D, or operation is performed. Classification processing may be performed such as sorting each record constituting the data D into various operation modes, or format conversion may be performed so that the format is suitable for reading by artificial intelligence. In some other embodiments, the training data L is generated by applying at least one process such as a concealment process and addition of data type information Id (described later) to the operation data D as described later. You may. In some other embodiments, the training data L may be generated from the operation data D by combining these embodiments.

More specifically, the learning data L may be generated as follows. First, the incidence of each trouble is determined in advance based on past operational data and experience. The trouble occurrence rate may be obtained from, for example, the overall occurrence rate × the occurrence frequency of each trouble. For example, if plant A operates for 2000 hours (hours) and a tube leak occurs once, plant B operates for 3000 hours and the same tube leak occurs, and plant C operates for 5000 hours and the burner burns out. Occurrence rate = 0.03% (3 times / 10,000 hours). Further, since the frequency of tube leakage = 2/3, the trouble occurrence rate of tube leakage in this case = 0.03% × 2/3 = 0.02%. On the other hand, since the frequency of burner burnout = 1/3, the trouble occurrence rate of burner burner in this case = 0.03% × 1/3 = 0.01%. The trouble occurrence rate may be updated each time a trouble occurs. Next, for example, the operation data D at the time of tube leakage and the operation data D in which no trouble has occurred are set so that the learning data L to be learned has the trouble occurrence rate calculated as described above. Synthesize.

In this way, when learning data L is generated, if only the data at the time of trouble occurrence is learned, even if the trouble originally occurs with a probability of 1/100, the occurrence probability is 1/1 for the artificial intelligence on the learning side. Since it causes trouble, it has a very large impact and may lead to erroneous learning, but by learning the trouble occurrence rate using learning data L that matches the actual situation, it is 1/100 of the original probability. Therefore, it is possible to efficiently and appropriately learn.

The learning data storage unit 4 is configured to store the learning data L generated by the learning data generation unit 3. That is, the generated learning data L is basically stored so that it can be transmitted to a plant 8 other than the source of the operation data D on which the training data L is based. In the embodiment shown in FIG. 1, the learning data storage unit 4 is formed in the storage device M. However, the present invention is not limited to this embodiment, and in some other embodiments, the learning data storage unit 4 is communicably connected to another learning support device 1, which is physically different from the learning support device 1. It may be formed in a storage device included in the device. Further, if necessary, for example, when the learning data L is found to be the learning data Lt (described later) at the time of abnormality, the training data L may be sent to the source of the operation data D which is the basis of the learning data L. good.

The management database 5 is configured to manage management information I, including information about the plant 8, as shown in FIGS. 3A-3B. More specifically, the management information I includes the plant information Ir to which the learning support device 1 transmits / receives the operation data D and the learning data L. In the plant information Ir, a communication address for communicating using the communication network 92, such as an IP address for communicating with the device on the plant 8 side that receives the operation data D, such as a plant control device (81c, 82c), is used. May be included.

In the embodiment shown in FIG. 3A, the plant information Ir includes the transmission availability information Ig indicating whether or not the learning data L can be transmitted, and the learning data L is transmitted to a plurality of plants 8 constituting the learning support system 9. By individually setting whether or not it can be done, the flexibility of managing the destination is achieved.

On the other hand, in the embodiment shown in FIG. 3B, as shown in FIG. 3B (a), the above-mentioned management information I includes the transmittable learning data management information corresponding to each plant information Ir. This transmittable learning data management information is information for indicating the learning data L that can be transmitted from the learning support device 1 due to the purchase of the plant 8 corresponding to the plant information Ir for each plant information Ir. , The learning data management number In assigned to each learning data L is used as a key for management. The learning data management number In is assigned a different number each time the learning data L is generated by the learning data generation unit 3. In this way, by individually setting whether or not each learning data L can be transmitted to the plurality of plants 8 constituting the learning support system 9, the flexibility of management of the transmission destination is achieved. As shown in FIG. 3B (a), the management information I may include plant type information It such as whether or not the operation data providing plant 81 (whether or not the operation data D is provided). Further, the plant type information It is not limited to whether or not the operation data D is provided, and may be further subdivided according to the method of providing the operation data D, the confidential processing information Ip described later, and the like.

Further, the management information I may be configured to further include information about each learning data L. More specifically, as shown in FIG. 3B (b), the data type information Id (described later), the consideration information Ic (described later), and the learning data L are the basis for the above learning data management number In. The provider plant information Is that provided the operation data D is associated and managed. The provider plant information Is may include information indicating the basic configuration and specifications of the operation data providing plant 81 that provided the operation data D that is the basis of the learning data L, and is other than the plant 8 that is the provider. It is managed to be open to the plant 8 of. As a result, it becomes possible to specify what kind of configuration the operation data D, which is the basis of the training data L, is due to the operation in the plant 8. As a result, when considering the purchase of the training data L on the plant 8 side, the training data L generated from the operation data D in the plant 8 having similar specifications can be selectively purchased, so that more efficient learning can be performed. Leads to.

In the embodiment shown in FIG. 1, the management database 5 is formed in a part of the storage device M. Further, since all the plants 8 included in the learning support system 9 are registered as plant information Ir in the management database 5, the learning support device 1 can use the learning data L for all of the plurality of plants 8. It is possible to send. However, the present invention is not limited to this embodiment, and in some other embodiments, the management database 5 is a communicably interconnected other device that is physically different from the learning support device 1. It may be formed in the storage device provided.

The learning data transmitting unit 6 transmits the learning data L stored in the learning data storage unit 4 to at least one of the plurality of plants 8 based on the destination managed by the management database 5. It is composed. In the embodiment shown in FIG. 3A, the learning data transmission unit 6 can transmit the learning data L only to the plant 8 registered in the plant information Ir of the management database 5, and is registered in the management database 5. The training data L is not transmitted to the plant 8 that does not exist. In the embodiment shown in FIG. 3B, the learning data L corresponding to the learning data management number In registered in the transmittable learning data management information is transmitted to the corresponding plant 8.

Further, in some embodiments, the learning data transmission unit 6 receives the training data L provision request Rs from the learning data receiving plant 83, and the learning support device 1 transmits the provision request Rs to the plant 8. The training data L may be transmitted (see FIG. 2 described later). In some other embodiments, the learning support device 1 transmits the learning data L when there is new learning data L to be provided, regardless of the presence or absence of the provision request Rs from the learning data receiving plant 83. You may. In some other embodiments, a combination of both of the above embodiments may be used. At this time, if the learning data transmission unit 6 provides the learning data L to the maker 85 of the plant 8, the maker 85 can independently provide additional services such as monitoring the learning data L. ..

Further, in some embodiments, the learning data receiving plant 83 receives a notification (not shown) from the learning support device 1 that the learning data L can be provided, so that the learning data L is requested to be provided Rs. May be sent. In some other embodiments, the training data receiving plant 83 may periodically transmit provision requests Rs. In this case, the learning support device 1 may transmit the learning data L to the request source when new learning data L exists for the plant 8 that has transmitted the provision request Rs. The latest learning data L or the like may be simply transmitted in response to the provision request Rs.

The learning support method of the plant executed by the learning support device 1 having the above-described configuration will be described with reference to FIG. The plant learning support method (hereinafter, appropriately referred to as a learning support method) is a method for supporting the shared use of the learning data L among a plurality of plants 8 that can be controlled using artificial intelligence, and FIG. As shown in the above, an operation data receiving step (S1) executed by the operation data receiving unit 2, a learning data generation step (S2) executed by the learning data generation unit 3, and a learning data storage executed by the learning data storage unit 4. A step (S3) and a learning data transmission step (S4) executed by the learning data transmission unit 6 are provided. The learning support method executed by the learning support device 1 will be described according to the flow of FIG.

In step S1 of FIG. 2, the operation data receiving step is executed. The operation data receiving step (S1) is a step of receiving the operation data D transmitted from the operation data providing plant 81. In the embodiment shown in FIG. 2, the operation data receiving step (S1) is executed on the premise that steps Sa to Sc performed before step S1 are performed.

Explaining step Sa to step Sc which are the premise, in step Sa, the AI program is provided (transmitted) from the learning support device 1 to each plant 8 (81, 82) via the communication network 92. Then, by installing this AI program in the plant control device (81c, 82c) in each plant 8 (81, 82), the operation control of the plant 8 by the AI program is performed in step Sb. After that, only the operation data providing plant 81 executes step Sc, so that the operation data D is transmitted to the learning support device 1 at a predetermined timing as described above. In the embodiment shown in FIG. 2, the AI program is transmitted to the operation data providing plant 81 and the operation data non-providing plant 82 at the same timing, but the timings may be different. Further, in some other embodiments, the AI program may be installed by storing the AI program in a portable recording medium and connecting the recording medium to the plant control device (81c, 82c).

In step S2, the learning data generation step is executed. The learning data generation step is a step of generating learning data L based on the operation data D received by the operation data reception step (S1). Since the learning data generation step (S2) is the same as the process executed by the learning data generation unit 3 described above, the details will be omitted.

In step S3, the learning data storage step is executed. The learning data storage step is a step of storing the learning data L generated by the learning data generation step (S2). Since the learning data storage step (S3) is the same as the process executed by the learning data storage unit 4 described above, the details will be omitted.

In step S4, the learning data transmission step is executed. In the learning data transmission step (S4), the learning data L stored in the learning data storage unit 4 is transferred to at least one of the plurality of plants 8 to the plant 8 based on the destination managed by the management database 5 described above. It is a step to send. Since the learning data transmission step (S4) is the same as the process executed by the learning data transmission unit 6 described above, the details will be omitted. In the embodiment shown in FIG. 2, in the stage prior to step S4, the learning data transmission step (S4) is executed in response to the learning data L provision request Rs (step S40) performed by the operation data non-providing plant 82. It has become like. After receiving the learning data L, the learning data receiving plant 83 relearns using the received learning data L (step S41).

According to the above configuration, the learning support device 1 of the plant 8 generates learning data L based on the received operation data D of the plant 8, and transmits the generated learning data L to another plant 8. As a result, the plant 8 (learning data receiving plant 83) that has acquired the training data L can acquire various learning data L earlier, so that the accuracy of, for example, optimized operation and abnormality detection can be improved earlier. Can be improved. For example, if the learning data L generated based on the operation data D (operation data Dt at the time of abnormality) such as when a sign of abnormality appears in the plant 8 or when an abnormality occurs is acquired, each plant 8 will be self-sufficient. Since it is possible to learn in advance even if an abnormality has not occurred in the plant, even an abnormality that has not occurred can be detected at an early stage.

Therefore, it is possible to prevent the abnormality from becoming serious (serious) and improve the reliability of the plant 8. Further, by providing the learning data L via the learning support device 1, the provider of the operation data D on which the learning data L is based is not known to the learning data receiving plant, which will be described later. It is also possible to add added value such as concealment processing and generation of teacher data to the learning data L. In this way, by supporting the sharing of the learning data L among the plurality of plants 8, effective learning in each plant 8 can be supported.

Next, some embodiments relating to the generation of the learning data L by the learning data generation unit 3 will be described with reference to FIGS. 4 to 5. FIG. 4 is a diagram showing a configuration of a learning data generation unit 3 according to an embodiment of the present invention, and shows a case where an encryption process is performed as a concealment process. Further, FIG. 5 is a diagram showing a learning data generation step (S2) according to an embodiment of the present invention, and corresponds to a case where an encryption process is performed as a concealment process.

In some embodiments, as shown in FIG. 4, the learning data generation unit 3 described above may generate the learning data L by performing a concealment process for concealing the time information of the operation data D. This time information describes the relationship between the control content and operation state of the plant 8 and its time and transition, such as the time data (log time) possessed by each record of the operation data D and the time series including the order of the records. Information to connect. The reason for this configuration is as follows. That is, conventionally, because the operation data D of the plant 8 is closely related to the management information, the owner of the plant 8 provides the operation data D to the manufacturer of the plant 8 for abnormality analysis and the like. However, there is a strong sense of resistance to providing the operation data D as the learning data L. It is even more difficult to provide operational data to plants owned by others. That is, unless the possibility that the management information can be read from the learning data L is excluded, it is unacceptable for the driving data provider to provide the driving data.

From such an analysis, the present inventors examined a method of breaking the close relationship between the driving data D and the management information, and as a result, the learning support device 1 and the learning data L to which the driving data D is provided are provided. It is possible to create a situation where the owner of the plant 8 can easily permit the provision of the operation data D by processing so that the time information cannot be grasped from the operation data D on the learning data receiving plant 83 side to be provided. Thought. Therefore, the above concealment process is applied to the operation data D to generate the learning data L. That is, in the learning data L, the time information that can be grasped from the operation data D is concealed.

In some embodiments, this concealment process may be executed for all the operation data D received by the learning support device 1. Alternatively, in some other embodiments, it may be performed only for the operation data D transmitted by the operation data providing plant 81 notifying that the concealment processing is necessary. In this case, the management database 5, as one of the management information I, is necessary for the concealment processing and the specific processing content of the concealment processing (for example, described later) in case a plurality of concealment processing is prepared. The secret processing information Ip for instructing the encryption, deletion, sorting, etc. of the time data to be performed may be configured to be managed together with the plant information Ir (see FIGS. 3A to 3B). In this way, the learning support device 1 learns from the operation data D according to the confidential processing information Ip obtained by referring to the management database 5 when the operation data D transmitted by the operation data providing plant 81 is received. It becomes possible to appropriately generate the data L. The learning data L may be generated by extracting the feature amount used for abnormality detection and control from the operation data D together with the concealment processing (see FIGS. 6 to 8 described later).

According to the above configuration, in the training data L, the time information in the operation data D that connects the control contents of the plant 8 such as the time data (log time) and the time series and the relationship between the operation state and the time is concealed. By setting the state, it is possible to eliminate or reduce factors that make it difficult to allow the provision of the operation data D, such as the relationship with the management information. Therefore, since the owner of the plant 8 can alleviate the resistance to providing the operation data D of his / her own plant, the increase of the plant 8 (operation data providing plant 81) that allows the provision of the operation data D is promoted. Can be planned.

Specifically, the embodiment relating to the processing content of the above-mentioned concealment processing will be specifically described. In some embodiments, as shown in FIGS. 4 to 5, the concealment processing encrypts the time data (described above) possessed by the operation data D. It may be an encryption process to be converted. By referring to the time data of each record constituting the operation data D, it is possible to grasp the necessary range of information such as the year, month, day, hour, minute, and second from which the log was acquired. Therefore, by encrypting the time data, the time information of the operation data D can be concealed.

At this time, in some embodiments, as shown in FIG. 4, the learning support device 1 further executes the concealment process without storing the decryption key K for decrypting the encrypted time data. The decryption key K may be transmitted (returned) to the operation data providing plant 81, which is the source of the data D. That is, only the plant 8 that provides each operation data D may manage the decoding key K. In the embodiment shown in FIG. 4, the learning data generation unit 3 described above has a time data encryption / decryption unit 3a.

The time data encryption / decryption unit 3a (learning data generation unit 3) has an encryption unit 31 that executes an encryption process and generates a decryption key K for decrypting the encrypted time data, and the encryption process. When the decryption key return unit 32 that transmits the decryption key K to the operation data providing plant 81 that is the transmission source of the learned data L and the decryption key K from the operation data providing plant 81 are received, the encryption key K is used for encryption. The decryption unit 33 that decodes the decoded time data and the restoration operation data Dr that is the learning data L including the time data decoded by the decoding unit 33 are transferred to a predetermined transfer destination such as the maker 85 of the plant 8. It has an operation data transfer unit 34. The restored operation data Dr may be completely the same as or equivalent to the operation data D when received from the operation data providing plant 81 by decoding the time data, or may be the data having the same contents as described above. The data may include the same or equivalent contents as a part of the operation data D at the time of reception.

More specifically, the encryption unit 31 inputs the learning data L including the encrypted time data to the learning data storage unit 4. Then, the learning data storage unit 4 stores the encrypted time data in a state where the correspondence between the encrypted time data and the learning data L can be understood. Further, when the decoding unit 33 receives the decoding key K, the decoding unit 33 acquires the learning data L having the time data that can be decoded by the decoding key K from the learning data storage unit 4, and decodes the encrypted time data. .. At this time, the decoding key K may be provided with the information of the learning data management number In for indicating which learning data L corresponds to, so that it may be determined which learning data L is to be decoded. .. Alternatively, the source plant information Is included in the management database 5 is referred to, and the generated timing is the latest from those that match the plant information Ir corresponding to the source plant 8 of the decoding key K (learning data management). The number In may be the largest, etc.) to be decoded. On the other hand, when the operation data providing plant 81 receives the decoding key K returned from the learning data generation unit 3, the decoding key K is stored, and when a problem such as an abnormality or a sign thereof is detected, the operation data providing plant 81 stores the decoding key K. The decryption key K is transmitted to the learning support device 1.

The learning data generation step (S2) corresponding to the process executed by the learning data generation unit 3 (time data encryption / decryption unit 3a) having the above-described configuration will be described with reference to FIG. As shown in FIG. 5, the learning data generation step (S2) includes an encryption step (S21) that executes an encryption process and generates a decryption key K for decrypting the encrypted time data, and encryption. The decryption key return step (S22) for transmitting the decryption key K to the operation data providing plant 81 of the transmission source of the processed learning data L, and the decryption key K when the decoding key K is received from the operation data providing plant 81. The decryption step (S25) for decrypting the time data encrypted using the data, and the restoration operation data Dr which is the learning data L including the time data decrypted by the decryption step (S25) are previously stored in the plant 8 manufacturer 85 or the like. It has an operation data transfer step (S26) for transferring to a predetermined transfer destination.

The learning data generation step (S2) of the present embodiment will be described according to the flow of FIG. 5. In step S21 of FIG. 5, the encryption unit 31 executes the encryption step, and the operation data D received by the operation data reception unit 2 Encrypts the time data and generates the decryption key K. In step S22, the decryption key return unit 32 executes the decryption key return step, and returns the decryption key K generated in step S21 to the source of the encrypted operation data D. In step S23, the decoding unit 33 monitors the reception of the decoding key K transmitted by the operation data providing plant 81. The operation data providing plant 81 transmits the decoding key K to the learning support device 1 when an abnormality in the own plant is detected.

Then, as a result of monitoring in step S23, if it is determined in step S24 that the decoding unit 33 has not received the decoding key K, the process returns to step S23. On the contrary, when the decoding unit 33 determines in step S24 that the decoding key K has been received, the decoding step is executed in step S25, the time data is decoded using the decoding key K, and the operation data D is restored. do. Then, in step S26, the operation data transfer unit 34 executes the operation data transfer step, and transmits the restored operation data Dr to the transfer destination (manufacturer 85).

According to the above configuration, the time information of the operation data D can be concealed by encrypting the time data of the operation data D as a concealment process. Further, the decryption key K for decrypting the encrypted time data is returned to the operation data providing plant 81 that provided the operation data D that is the basis of the learning data L. As a result, only the source of the operation data D manages the decryption key K, and the confidentiality of the operation data D can be enhanced.

Further, when the occurrence of a problem is detected in the operation data providing plant 81, the restored operation data Dr can be quickly provided to the transfer destination such as the manufacturer 85 by transmitting the decoding key K to the learning support device 1. The investigation of the problem on the manufacturer side can be started promptly, and the problem can be solved promptly. Therefore, by providing an incentive for the provision of the operation data D such as such a support service, it is possible to further promote the increase of the number of plants 8 that allow the provision of the operation data D.

In the present embodiment as shown in FIG. 4 described above, as described above, the learning data transmission unit 6 may transmit the learning data L having no encrypted time data. In this case, only the information necessary for learning can be sent to the learning data receiving plant 83. Alternatively, the learning data L including the encrypted time data may be sent. In this case, the learning data storage unit 4 can easily manage the learning data L.

However, the present invention is not limited to this embodiment. For example, in some other embodiments, the concealment process may include a process of deleting the time data when the operation data D has the time data. Further, when the operation data D does not include the time data, a process of rearranging the operation data D may be included. That is, the concealment process may include at least one of the above-mentioned time data deletion process and the process of breaking the time series by rearranging the records constituting the operation data. When the time information of the record in the operation data D is unknown due to the deletion of the time data, the learning data is further extracted from the operation data D by extracting the feature amount used for abnormality detection and control. L may be generated (see FIGS. 6 to 7 described later). By deleting or rearranging the time data in this way, it is possible to conceal the time information of the operation data D while suppressing the processing load.

Further, in the above-described embodiment, the learning support device 1 performs the concealment processing, but in some other embodiments, the operation data providing plant 81 deletes the time data possessed by the operation data D, or The time series in the operation data D, which is broken by random sorting or the like, may be transmitted to the learning support device 1 as the operation data D.

The embodiment relating to the concealment processing has been described above. In some other embodiments, the learning data generation unit 3 has information on whether or not the abnormal operation data Dt acquired at the time of abnormality of the operation data providing plant 81 is included in the operation data D (data type information Id). May be added to the operation data D to generate the training data L. Here, the “abnormal time” means a time when some problem occurs, such as when a sign of an abnormality appears in the plant 8 or when an abnormality occurs (the same applies hereinafter).

Further, in some embodiments, the data type information Id may be added to the operation data D by storing the data type information Id together with the learning data L in the learning data storage unit 4. In some other embodiments, the data type information Id may be added to the operation data D by generating the learning data L having the data type information Id. Thereby, by referring to the data type information Id, at least, it is possible to quickly determine whether or not the learning data L is the abnormal operation data Dt. In some other embodiments, the learning data generation unit 3 adds the above-mentioned data type information Id to the record constituting the operation data D, and is abnormal to the data type information Id that the record has. You may enter information indicating whether or not. As a result, it is possible to quickly determine which part of the abnormal operation data Dt is abnormal.

Alternatively, in some other embodiments, the data type information Id may be added to the operation data D by managing it in the management database 5 as shown in FIG. 3B (b). In the embodiment shown in FIG. 3B, as shown in (b), as one of the information regarding the training data L, the training data management number In assigned each time the training data L is generated by the training data generation unit 3 and the training data management number In thereof. The data type information Id corresponding to the training data L is stored in the management database 5. Since the data type information Id is one element for setting the consideration information Ic described later for each learning data L, it can be easily managed by storing it in the management database 5 together with the consideration information Ic. The data type information Id may be further subdivided according to the abnormality level (frequency of occurrence of abnormality described later and assumed damage level).

As shown in FIGS. 3A to 3B, the data type information Id may be subdivided into types according to an abnormal level (estimated damage level (described later) corresponding to the frequency of occurrence, estimated damage amount, etc.). Further, it may be determined whether or not the learning data L is the abnormal operation data Dt through comparison with the set value of the operation data D or the threshold value for each operation state. This determination may be executed by the learning support device 1. Alternatively, the operation data providing plant 81 may execute the operation, and in this case, the learning support device 1 obtains the determination result automatically or manually performed by the operation data providing plant 81, and is based on this determination result. , The learning data L is generated by adding the data type information Id to the operation data D.

According to the above configuration, it is possible to efficiently select and learn the abnormal operation data Dt from a huge amount of operation data. As a result, each plant 8 can more efficiently improve the abnormality detection accuracy by performing learning using the teacher data.

Hereinafter, an embodiment in which the learning data L is traded between the plurality of plants 8 will be described with reference to FIGS. 6 to 8. FIG. 6 is a diagram showing a learning support device 1 of a plant 8 according to an embodiment of the present invention, and shows a case where learning data L is traded between a plurality of plants 8 for a fee. FIG. 7 is a diagram showing a learning support device 1 of a plant 8 according to an embodiment of the present invention, and is a case where the learning data L is traded for a fee within a membership group of a membership fee system composed of a plurality of plants 8. show. Further, FIG. 8 is a diagram showing a learning support device 1 of a plant 8 according to an embodiment of the present invention, in which learning data L is traded free of charge within a membership group consisting of a plurality of plants 8. Show the case.

As described above, since the training data L is generated based on the operation data D obtained by actually operating the plant 8, it is premised that the plant 8 allows the provision of the operation data D, but the operation is performed. It is often difficult for the owner of plant 8 to authorize the provision of data D. Therefore, in the embodiment described below, the learning data L is targeted for commercial transactions, and the operation data providing plant 81 pays more for the acquisition (purchase) of the learning data L than the learning data non-providing plant 82. It provides an incentive to become an operation data providing plant 81, such as setting a low value.

Hereinafter, it will be described in detail. In the following description, it is assumed that the AI program executes learning using the learning data L. The AI program may be provided to the plant 8 free of charge or for a fee. The consideration is explained as a price. Further, the learning data generation unit 3 has a time data encryption / decryption unit 3a and a feature amount extraction unit 3b configured to use the feature amount for abnormality detection and control from the operation data D.

In some embodiments, as shown in FIGS. 6-7, the learning support device 1 is on the learning data receiving plant 83 side (that is, on the plant 8 side receiving the learning data L transmitted by the learning data transmitting unit 6). ) Further includes a consideration registration unit 7 configured to register the consideration to be paid for the reception of the learning data L in the management database 5 for each destination or for each learning data L. As a result, depending on whether the plant 8 is the operation data providing plant 81 or not, the consideration can be set individually for each plant 8, or the learning data L generated from the operation data D including the abnormal operation data Dt can be set. It is possible to set the consideration individually for each learning data L depending on whether or not it is.

In the embodiment shown in FIGS. 6 to 7, the consideration setting unit 72, which will be described later, sets (determines) the consideration (the consideration setting step) and notifies the consideration registration unit 7 of the set consideration, whereby the consideration registration unit 7 Is designed to register the set consideration in the management database 5. However, the present invention is not limited to this embodiment. In some other embodiments, the consideration registration unit 7 may be configured to receive the input of the consideration set by a person or the like and register the input consideration in the management database 5. Although FIGS. 6 to 7 show a case where the management database 5 is formed in storage devices that are communicably connected to each other, the learning support device 1 may include the management database 5.

For example, in the management database 5 of the embodiment shown in FIG. 3A, the above-mentioned plant information Ir is associated with the consideration information Ic regarding the consideration, and the above-mentioned consideration can be registered for each destination. In this case, the consideration registration unit 7 registers the consideration in the consideration information Ic. Specifically, the price itself required for purchase may be input to the consideration information Ic. For example, a normal price common to each learning data L or in the learning data L is set, and a price calculated by multiplying this normal price by a discount rate or the like is stored in the consideration information Ic. To. Alternatively, information such as a discount rate for the regular price may be registered. The discount rate may be, for example, 100% (that is, free of charge).

On the other hand, in the embodiment shown in FIG. 3B, as shown in FIG. 3B (b), the consideration information Ic is registered in each learning data L, and the consideration information Ic is registered for each destination or for each learning data L. , This is an example where consideration can be registered. For example, by registering the normal price of the learning data L in the consideration information Ic of FIG. 3B (b), for example, according to the data type information Id, and multiplying by the discount rate according to the plant type information It, each plant It is possible that the price of each learning data L for 8 is determined. More specifically, the consideration setting unit 72 sets a normal price according to the data type information Id or the like for the learning data L, the consideration registration unit 7 registers it in the management database 5, and the consideration notification unit 73 registers it in the management database. It may be configured to notify the price of the learning data L set for each plant 8 by multiplying the discount rate according to the plant type information It of 5.

Further, at least a part of the consideration paid by the learning data receiving plant 83 side is received by the operation data providing plant 81. That is, the operation data providing plant 81 may receive all the consideration, and the operation data providing plant 81 and the learning support system 9 are paid at a predetermined ratio between the management side (for example, the manufacturer 85 of the plant 8). You may distribute it. In this way, by buying and selling the operation data D for a fee, by providing an incentive for the provision of the operation data D (becoming the operation data providing plant 81), the number of plants 8 that allow the provision of the operation data D is increased. It can be promoted.

However, the present invention is not limited to the present embodiment. In some other embodiments, the consideration paid by the learning data receiving plant 83 may be entirely received by the management side of the learning support system 9.

Further, in setting the above consideration, in some embodiments, as shown in FIGS. 6 to 7, the above consideration is based on whether or not the learning data receiving plant 83 is the operation data providing plant 81. May be set. In this case, the case where the learning data receiving plant 83 is the operation data providing plant 81 is set to be lower than the case where the operation data non-providing plant 82 is. In the embodiment shown in FIGS. 6 to 7, the learning data receiving plant 83 (FIGS. 6 to 7) is provided by the consideration notification unit 73 configured to notify the learning data receiving plant 83 of the consideration (price) included in the learning support device 1. In 8, the normal price is notified to the operation data non-providing plant 82), but the operation data providing plant 81 is notified of the discount price to which the discount rate is applied to the normal price.

At this time, as in the embodiment shown in FIGS. 7 to 8, the transaction of the learning data L is within the member group for which the membership fee for being a member (member) of the member group of the maintenance management service of the plant is required, for example. If so, there may be a difference in membership fees. That is, in some embodiments, as shown in FIGS. 7 to 8, in the learning support system 9, a member group is formed by a plurality of plants 8 registered in the management database 5, and the member group is formed. Each of the eight plants has a mechanism for paying a predetermined membership fee. The membership fee paid by the operation data providing plant 81 side is lower than the membership fee paid by the operation data non-providing plant 82 side.

In the embodiment shown in FIGS. 7 to 8, the learning support device 1 determines the membership fee of the member belonging to the member group based on whether or not the member is an operation data providing plant, and notifies the determined membership fee. It also has a membership fee determination unit 76. Then, when the plant 8 applies for membership to join the member group, it notifies whether or not the operation data D is provided, and the membership fee determination unit 76 determines the membership fee based on this information. It is configured to notify you. Further, in the embodiment shown in FIGS. 7 to 8, the learning support device 1 further includes a payment notification receiving unit 77 for receiving the payment notification, and the payment notification receiving unit 77 receives the payment notification from the plant 8. Then, the AI program is sent to the source plant 8.

That is, in the embodiment shown in FIG. 7, a difference is provided in both the payment amount of the membership fee and the purchase amount of the learning data L depending on whether or not the plant 8 is the operation data providing plant 81. In the embodiment shown in FIG. 8, the learning data L is provided free of charge while the payment amount of the membership fee is different. Therefore, the learning support device 1 may transmit the learning data L to the learning data receiving plant 83 each time the learning data L is generated. Although the embodiment shown in FIG. 6 has been described as not forming a member group, a member group with no membership fee may be formed by a plurality of plants 8.

According to the above configuration, the plant 8 is generated by providing the operation data D, for example, receiving maintenance service at a lower membership fee or based on the operation data D provided by another plant 8. It is possible to acquire the learning data L at a lower price. In this way, by providing an incentive for the provision of the operation data D (becoming the operation data providing plant 81), it is possible to promote the increase of the number of plants 8 that allow the provision of the operation data D.

In some other embodiments, the above consideration is that the training data L is generated based on the operation data including the abnormality operation data Dt acquired at the time of the abnormality (described above) of the operation data providing plant 81. It may be set based on whether or not it is the time learning data Lt. That is, the abnormal operation data Dt is the data corresponding to the log (record) acquired at the time of the abnormality of the plant 8, and the abnormal learning data Lt is the operation including at least a part of such abnormal operation data Dt. It is the learning data L generated based on the data D.

In this case, the consideration for the learning data L is set to be higher in the case of the abnormal learning data Lt than in the case of not the abnormal learning data Lt. There are various contents of abnormalities (problems) that occur in the plant 8, and further, depending on the contents of the corresponding abnormalities of the abnormal time learning data Lt, a difference in consideration is provided between the abnormal time learning data Lts. May be. Specifically, depending on the frequency of occurrence of the abnormality, for example, the higher the frequency of occurrence, the higher the consideration may be. If the frequency of occurrence of abnormalities is high, the response to the occurrence of abnormalities will increase accordingly. It is expected that the response cost will be reduced, such as remote support (online support) without local support (offline support). Alternatively, depending on the assumed damage level at the time of occurrence of the abnormality, the higher the assumed damage level, the higher the consideration may be. For example, the assumed damage level is determined based on the amount of damage, the suspension period, and the like. Consideration may be set based on both the frequency of anomalies and the expected damage level. Further, when the consideration is distributed between the operation data providing plant 81 and the management / operation source of the learning support system 9, the higher the value of the learning data L is, the more the learning data L is distributed to the operating data providing plant 81 that provided the learning data L. You may try to increase the ratio.

According to the above configuration, the consideration for the learning data L is set so that the learning data Lt at the time of abnormality is higher than the case where it is not. That is, since the learning data Lt at the time of abnormality is more useful for improving the reliability of the plant 8 such as improving the accuracy of the abnormality determination of the plant 8, the value of the learning data L is the learning data Lt at the time of abnormality. Is higher than otherwise, and this is reflected in the consideration. As a result, the consideration received by the operation data providing plant 81 that provides the operation data D that is the basis of the learning data Lt at the time of abnormality can be made higher as a result. Therefore, the operation data providing plant 81 can solve the abnormality, for example. It will be possible to more fully compensate for the costs required for the above. In this way, by setting the consideration according to the value of the training data L, the plant that allows the provision of the operation data D by providing an incentive for providing the training data L (becoming the operation data providing plant 81). Can be promoted.

Further, in some embodiments, as shown in FIGS. 6 to 7, when the operation data receiving unit 2 receives the operation data D, the learning support device 1 transmits the received operation data D. A consideration setting unit 72 configured to set a consideration for the operation data providing plant 81 may be further provided. In the embodiment shown in FIGS. 6 to 7, the consideration setting unit 72 sets the consideration after the learning data generation unit 3 generates the learning data L based on the operation data D received by the operation data receiving unit 2. It has become. More specifically, the consideration setting unit 72 is connected to the learning data generation unit 3, and the learning data generation unit 3 sets the learning data L based on the operation data D received by the operation data reception unit 2 as the consideration setting unit 72. Send to. Then, the consideration setting unit 72 sets the consideration (price) according to the value of the learning data L as described above, for example, through the analysis of the learning data L.
According to the above configuration, the consideration can be automatically set based on the value of the operation data D.

The present invention is not limited to the above-described embodiment, and includes a modification of the above-mentioned embodiment and a combination of these embodiments as appropriate.

1 Plant learning support device 2 Operation data reception unit 3 Learning data generation unit 3a Time data encryption / decryption unit 3b Feature quantity extraction unit 31 Encryption unit 32 Decryption key return unit 33 Decoding unit 34 Operation data transfer unit 4 Learning data storage unit 5 Management Database 6 Learning data transmission unit 7 Consideration registration unit 72 Consideration setting unit 73 Price notification unit 76 Membership fee determination unit 77 Payment notification reception unit M Storage device 8 Plant 81 Operation data provision plant 81c Plant control device (operation data provision plant)
82 Plants that do not provide operation data 81c Plant controller (plants that do not provide operation data)
83 Learning data receiving plant 9 Learning support system 92 Communication network

D Operation data Dt Abnormal operation data Dr Restored operation data L Learning data L Abnormal learning data I Management information Ir Plant information Ig Transmission availability information Ip Concealment processing information Ic Consideration information It Plant type information In Learning data management number Id Data type information Is Provider Plant K Decryption Key Rs Provision Request

Claims (11)

It is a plant learning support device for supporting the exchange of learning data between multiple plants that can be controlled using a plant control device that can operate an AI program .
An operation data receiving unit configured to receive the operation data transmitted from the operation data providing plant that provides the operation data of the plant, and an operation data receiving unit.
A learning data generation unit configured to generate the learning data that the AI program can learn based on the operation data received by the operation data receiving unit.
A learning data storage unit configured to store the learning data,
A management database configured to manage management information including the destination of the learning data, and
A learning data transmission unit configured to transmit the learning data stored in the learning data storage unit to at least one of the plurality of plants based on the destination managed by the management database. A plant learning support device characterized by being equipped with. The learning support device for a plant according to claim 1, wherein the learning data generation unit generates the learning data by performing a concealment process for concealing the time information of the operation data. The concealment process is an encryption process for encrypting the time data of the operation data.
The learning data generation unit
An encryption unit that executes the encryption process and generates a decryption key for decrypting the encrypted time data.
A decryption key return unit that transmits the decryption key to the operation data providing plant that is the source of the learning data to which the encryption process has been performed, and a decryption key return unit.
Upon receiving the decryption key from the operation data providing plant, a decoding unit that decodes the encrypted time data using the decryption key, and a decoding unit.
The second aspect of claim 2 is characterized by having an operation data transfer unit that transfers restored operation data, which is the learning data including the time data decoded by the decoding unit, to a predetermined transfer destination. Plant learning support device. The concealment process includes a process of deleting the time data when the operation data has the time data.
The learning support device for a plant according to claim 2, wherein the operation data includes a process of rearranging the operation data when the operation data does not include the time data. The learning data generation unit generates the learning data by adding information as to whether or not the operation data includes the abnormal operation data acquired at the time of abnormality of the operation data providing plant to the operation data. The plant learning support device according to any one of claims 1 to 4, wherein the plant learning support device is characterized by the above-mentioned. The learning data receiving plant side, which is the plant that receives the learning data transmitted by the learning data transmitting unit, pays the management database the consideration to be paid for the reception of the learning data for each destination or the learning data. It also has a consideration registration unit that is configured to register each time.
The learning support of the plant according to any one of claims 1 to 5, wherein at least a part of the consideration paid by the learning data receiving plant side is received by the operation data providing plant side. Device. The learning data receiving plant side, which is the plant that receives the learning data transmitted by the learning data transmitting unit, pays the management database the consideration to be paid for the reception of the learning data for each destination or the learning data. It also has a consideration registration unit that is configured to register each time.
The consideration is set based on whether or not the learning data receiving plant is the operation data providing plant, and the case where the learning data receiving plant is the operation data providing plant is the case of the operation data. The learning support device for a plant according to any one of claims 1 to 6, wherein the cost is set to be lower than that of a plant that does not provide operation data, which is the plant that does not provide the data. The consideration is set based on whether or not the learning data is abnormal training data generated based on the abnormal operation data including the abnormal operation data acquired at the time of abnormality of the operation data providing plant. According to claim 6 or 7, the consideration for the training data is set to be higher in the case of the abnormal training data than in the case of not the abnormal learning data. The learning support device for the described plant. It is further provided with a consideration setting unit configured to set the consideration for the operation data providing plant that has transmitted the received operation data based on the operation data received by the operation data receiving unit. The plant learning support device according to any one of claims 6 to 8. A member group is formed by the plurality of plants registered in the management database.
The plurality of plants that form the member group each pay a predetermined membership fee.
Any of claims 1 to 9, wherein the membership fee paid by the operation data providing plant side is lower than the membership fee paid by the operation data non-providing plant side, which is the plant that does not provide the operation data. The plant learning support device according to item 1. It is a plant learning support method using a plant learning support device for supporting the exchange of learning data between a plurality of plants that can be controlled using a plant control device capable of operating an AI program .
An operation data receiving step in which the learning support device of the plant receives the operation data transmitted from the operation data providing plant that provides the operation data of the plant, and
A learning data generation step in which the learning support device of the plant generates the learning data that the AI program can learn based on the operation data received by the operation data reception step.
A learning data storage step in which the learning support device of the plant stores the learning data,
A management database in which the learning support device of the plant manages management information including a destination of the learning data, and
It includes a learning data transmission step of transmitting the learning data stored in the learning data storage step to at least one of the plurality of plants based on the destination managed by the management database. Plant learning support method.

Images