Information processing apparatus, information processing method, and computer-readable storage medium

Abstract

An information processing apparatus appropriately updates a prediction model that predicts a control content. The information processing apparatus (1) includes an evaluation section (104) that evaluates a prediction result of each unit period when a control target is controlled by a plurality of prediction models that alternately use a prediction control content for a plurality of unit periods, and a selection section (105) that selects a prediction model to be used later from among the plurality of prediction models based on the evaluation result.

Description

TECHNICAL FIELD

[0001] The present application relates to an information processing apparatus or the like for updating a prediction model constructed through mechanical learning. BACKGROUND

[0002] In the background of an increasing shortage of labor force, research and development of technology for automatically controlling equipment appropriately is currently being promoted. For example, Patent Literature 1 described below describes that a predicted amount of steam generation of a waste heat boiler after a prescribed time in a neural network model is output to a control logic, thereby performing automatic control of a waste treatment plant.

[0003] When using a prediction model constructed through mechanical learning, including a neural network model, in order to maintain or improve its prediction accuracy, it is necessary to update the prediction model through relearning or the like. For example, Patent Literature 1 described above describes that a new model is constructed during use of one model, and if an evaluation index of the new model is superior to that of the model in use, the new model is applied.

[0004] PRIOR ART DOCUMENTS

[0005] PATENT LITERATURE

[0006] Patent Literature 1: Japanese Patent Application Publication No. 2005-249349 SUMMARY

[0007] (1) Problem to be Solved

[0008] However, the above-described prior art has the following problem: although it is useful for updating a prediction model that predicts a process value such as an amount of steam generation, it is difficult to be used for updating a prediction model that predicts appropriate control content. The prediction model that predicts appropriate control content refers to a prediction model that outputs information indicating control content that should be performed in a state of operation of an apparatus itself or a facility that uses the apparatus, based on input data indicating the state of operation.

[0009] For example, in Patent Literature 1, a first prediction model that predicts appropriate control content is used to perform automatic control for a prescribed period (for example, two weeks), and a second prediction model that predicts appropriate control content is constructed using data collected during the period.

[0010] In this case, control based on a prediction result of the second prediction model is not performed during the above-described prescribed period, and control based on a prediction result of the first prediction model is performed, and a plant state changes according to the control.

[0011] The reality cannot perform the control based on the prediction result of the second prediction model in the above prescribed period, and thus it is difficult to evaluate whether the prediction result is appropriate. Also, if the appropriateness of the prediction result of the second prediction model in the above prescribed period cannot be evaluated, the prediction model cannot be appropriately updated at the end time of the prescribed period.

[0012] One embodiment of the present application is made in view of the above problems, and has an object to provide an information processing apparatus or the like capable of appropriately updating a prediction model of a prediction control content.

[0013] (II) Technical Solution

[0014] To solve the above-described technical problem, an information processing apparatus according to one embodiment of the present application includes an evaluation unit that evaluates a prediction result of each unit period when a control object is controlled by alternately using a plurality of prediction models that predict a control content that the control object should perform, in a plurality of unit periods that constitute a first prescribed period, and a selection unit that selects, from among the plurality of prediction models, a prediction model to be used in a second prescribed period after the first prescribed period, based on an evaluation result of the evaluation unit.

[0015] In addition, to solve the above-described technical problem, an information processing method according to one embodiment of the present application is executed by one or a plurality of information processing apparatuses, and includes an evaluation step of evaluating a prediction result of each unit period when a control object is controlled by alternately using a plurality of prediction models that predict a control content that the control object should perform, in a plurality of unit periods that constitute a first prescribed period, and a selection step of selecting, from among the plurality of prediction models, a prediction model to be used in a second prescribed period after the first prescribed period, based on an evaluation result of the evaluation step.

[0016] (III) Advantageous Effects

[0017] According to one embodiment of the present application, a prediction model of a prediction control content can be appropriately updated. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is a block diagram showing an example of the main part structure of an information processing apparatus according to one embodiment of the present application.

[0019] Figure 2 is a diagram showing an example of switching control performed by the above information processing apparatus and an example of update of a prediction model used by the above information processing apparatus.

[0020] Figure 3 is a flowchart showing an example of a process of updating a prediction model used by the above information processing apparatus.

[0021] Figure 4is a diagram indicating an example of correct data determination information.

[0022] Figure 5 is a flowchart indicating an example of a process of generating teacher data. DETAILED DESCRIPTION

[0023] (Structure of Apparatus)

[0024] Based on Figure 1 The structure of the information processing apparatus according to an embodiment of the present application will be described. Figure 1 is a block diagram indicating an example of the structure of the main part of the information processing apparatus 1. As shown in the figure, the information processing apparatus 1 is provided with a control section 10 that comprehensively controls each section of the information processing apparatus 1, and a storage section 12 that stores various data used by the information processing apparatus 1. In addition, the information processing apparatus 1 is provided with an input section 14 that receives an input to the information processing apparatus 1, and an output section 16 that is used for the information processing apparatus 1 to output data. Figure 1

[0025] The control section 10 includes an update management section 101, a prediction section 102, a factory control section 103, an evaluation section 104, a selection section 105, a teacher data generation section 106, a learning section 107, a manual control detection section 108, a reference information acquisition section 109, and a state determination section 110. In addition, the storage section 12 stores a prediction result DB (database) 121, a prediction model DB 122, and correct data determination information 123. The teacher data generation section 106 to the state determination section 110 and the correct data determination information 123 in the above structure will be described later.

[0026] The information processing apparatus 1 is an apparatus that realizes an automatic control system that uses a prediction model to predict a control content that should be performed on a control target, and causes the control target to operate based on the prediction result.

[0027] The prediction model described above is a model constructed in a manner that enables prediction of a control content that should be performed on a control target. In the present embodiment, an example in which the control target is various devices used in a factory will be described. That is, with the information processing apparatus 1, control is performed on various devices and the like included in a factory based on a prediction result of the prediction model described above, and automatic control of the factory is realized. Of course, the control target is arbitrary, and is not limited to this example.

[0028] In addition, the prediction model described above can be constructed using a statistical method, can be constructed using mechanical learning, or can be a model that combines both. The algorithm of the mechanical learning is arbitrary, and for example, a prediction model such as a neural network can be used.

[0029] ​The update management section 101 switches the prediction model used in the automatic control of the control object by unit period (e.g., one day). In addition, when a prescribed period (first prescribed period, e.g., 60 days) including a plurality of unit periods elapses, the update management section 101 updates the prediction model used for the automatic control. Also, in the prescribed period (second prescribed period, e.g., 60 days) after the update, the automatic control using the updated prediction model is performed. The update of the prediction model is related to the evaluation section 104 and the selection section 105.

[0030] The prediction section 102 predicts the control content that should be performed on the control object using the prediction model described above. Specifically, the prediction model described above learns a correspondence relationship between various data (e.g., process data such as the detection value of a sensor, and the like) indicating the control object and the state of the plant operation, and the control content that should be performed at the time of observing the data. Therefore, the prediction section 102 inputs the various data described above to the prediction model, and predicts the control content that should be performed on the control object based on the output value of the prediction model.

[0031] The plant control section 103 controls the operation of various devices used in the plant based on the prediction result of the prediction section 102. In addition, the plant control section 103 can directly control the devices described above, or can indirectly control by notifying the control content to the control device of the devices.

[0032] The evaluation section 104 evaluates the prediction result of each unit period when the control object is controlled using a plurality of prediction models described above alternately by unit period. The evaluation method can be any method that can compare the prediction accuracy of each prediction model of the unit period. Specific examples of the evaluation method will be described later.

[0033] The selection section 105 selects the prediction model used by the prediction section 102 later based on the evaluation result of the evaluation section 104 from a plurality of prediction models described above. That is, the prediction model used by the prediction section 102 is updated by the selection of the selection section 105.

[0034] Thus, the information processing apparatus 1 has the evaluation section 104 that evaluates the prediction result of each unit period when the control object is controlled using a plurality of prediction models that predict the control content that should be performed on the control object alternately by a plurality of unit periods constituting the first prescribed period. In addition, the information processing apparatus 1 has the selection section 105 that selects the prediction model used in the second prescribed period after the first prescribed period based on the evaluation result of the evaluation section 104 from a plurality of prediction models described above.

[0035] According to the above-described configuration, the prediction results of each unit period when the control target is controlled by alternately using a plurality of prediction models for each unit period are evaluated, and the prediction model used in the second prescribed period is selected based on the evaluation results. Therefore, the plurality of prediction models can be evaluated under substantially equal conditions, and thus the prediction model of the prediction control content can be appropriately updated.

[0036] The prediction result DB 121 is a database that stores the prediction results of the prediction section 102 and various data associated therewith. More specifically, in the prediction result DB 121, the input data input to each prediction model and the date and time of the prediction based on the input data (or the acquisition date and time of the input data) are stored in association. Also, for the input data stored in the input data that becomes the basis of the prediction control, data indicating the prediction control content is additionally associated. Also, in the prediction result DB 121, data indicating the manual control content of the control target by the operator by manual and the date and time of the manual control are stored in association. Further, the so-called "input data that becomes the basis of the prediction control" refers to data indicating that the output data for which the prediction control should be performed is output from the prediction model among the input data input to the prediction model.

[0037] The prediction model DB 122 is a database that stores the prediction models used by the prediction section 102. In the prediction model DB 122, a plurality of prediction models constructed in advance are stored, and also the prediction models constructed by the learning section 107 are stored, the details of which will be described later.

[0038] (Example of switching control and updating of the prediction model used)

[0039] Figure 2 is a diagram indicating an example of switching control and updating of the prediction model used. In this example, for 60 days, the update management section 101 switches the prediction model used by the prediction section 102 with the prediction model of the A group and the prediction model of the B group for each day. Also, the plant control section 103 performs operation control of various devices included in the plant according to the prediction results of the prediction section 102. The following describes an example in which the plant is a waste incineration plant.

[0040] Specifically, on the first day, the prediction section 102 performs prediction using the prediction model of the A group, and the plant control section 103 controls the operation of the waste incineration plant according to the prediction results. Also, on the second day, the prediction section 102 performs prediction using the prediction model of the B group, and the plant control section 103 controls the operation of the waste incineration plant according to the prediction results.

[0041] As Figure 2As shown, the prediction models in Group A include two abnormality prevention models Al, A2, and four normality maintenance models A3 to A6. In addition, the prediction models in Group B include two abnormality prevention models Bl, B2, and four normality maintenance models B3 to B6. Further, since the prediction models of Group A and the prediction models of Group B are used alternately, the prediction models Al to A6 and the prediction models Bl to B6 have the same respective prediction targets.

[0042] The abnormality prevention model is a prediction model that predicts a non-normal operation state of the plant (for example, an abnormality occurrence) and predicts a control content that should be performed to prevent the non-normal operation state. One abnormality prevention model can be prepared for each type of abnormality that is desired to be prevented. For example, the abnormality prevention models Al, Bl can be made prediction models that detect deterioration of the combustion state due to a decrease in the amount of garbage in the incinerator, that is, "garbage amount is small, abnormality," and predict a control content for preventing the abnormality. In addition, the abnormality prevention models A2, B2 can be made prediction models that detect deterioration of the combustion state due to garbage with low heat being fed into the incinerator, that is, "non-flammable garbage abnormality," and predict a control content for preventing the abnormality.

[0043] The normality maintenance model is a prediction model that predicts a control content that should be performed to maintain the operation of the garbage incineration plant as normal. The normality maintenance model can be prepared in the garbage incineration plant for each device that becomes a control target of the plant control unit 103. For example, the normality maintenance models A5, B5 can be made prediction models for control of a garbage feeding device that feeds garbage into the incinerator. In addition, for example, the normality maintenance models A6, B6 can be made prediction models for operation control of a grate that conveys garbage in the incinerator. In addition to this, for example, if the garbage incineration plant has a plurality of grates, a separate prediction model can be prepared for each grate.

[0044] Further, the conditions of "normal" and "abnormal" can be determined in advance. For example, a normal range can be set for the output values of various sensors provided at various locations in the plant, and it can be determined that a state in which the output values of all the sensors are within the normal range is "normal" and a state in which the output values of at least some of the sensors are outside the normal range is "abnormal." In addition, the type of "abnormality" can be determined in correspondence with the type of sensor whose output value is outside the normal range.

[0045] Thus, in a case where the abnormality prevention model and the normal maintenance model are included in the prediction model used for the predictive control, the evaluation section 104 evaluates the prediction results of the plurality of abnormality prevention models and the plurality of normal maintenance models, respectively. Also, the selection section 105 selects the abnormality prevention model and the normal maintenance model to be used next time on the basis of the evaluation results of the evaluation section 104. Thus, it is possible to appropriately update both the abnormality prevention model and the normal maintenance model. Also, by using these prediction models that have been appropriately updated, it is possible to perform stable automatic control of the plant.

[0046] The update management section 101 updates the prediction models used when 60 days elapse during the operation of the waste incineration plant. In Figure 2 In the example, the update management section 101 updates the prediction models used from the group A and the group B to the group C and the group D.

[0047] The group C is constituted by the prediction models selected from the group A and the group B by the selection section 105 on the basis of the evaluation results of the prediction results of the prediction models of the group A and the group B evaluated by the evaluation section 104 for the period of 60 days. In Figure 2 In the example, the evaluation section 104 evaluates the abnormality prevention models with the prevention rate of abnormality and evaluates the normal maintenance models with the maintenance rate of normal state.

[0048] For example, the evaluations of the first abnormality prevention models Al, Bl are 79% and 70%, respectively, and the abnormality prevention model Al is highly evaluated. Thus, the selection section 105 selects the abnormality prevention model Al as the first abnormality prevention model in the group C. On the other hand, the evaluations of the second abnormality prevention models A2, B2 are 75% and 78%, respectively, and the abnormality prevention model B2 is highly evaluated. Thus, the selection section 105 selects the abnormality prevention model B2 as the second abnormality prevention model in the group C. Also, the selection section 105 selects the prediction model of the group A and the group B with higher evaluation in the same manner as described above. Thus, the group C constituted by the prediction models of the group A and the group B with higher evaluation is constituted.

[0049] On the other hand, the group D is constituted by the prediction models reconstituted by the learning section 107 on the basis of the data collected during the period of 60 days in which the prediction control is performed by the prediction models of the group A and the group B. In Figure 2 In the example, the group D is constituted by the prediction models Dl to D6 reconstituted.

[0050] Also, in the above-described mechanical learning, it is possible to use only the above-described data collected during the period of 60 days or to use the data collected during other periods. For example, it is possible to perform the mechanical learning using both the data used in the constitution of the prediction models of the group A or the group B and the above-described data collected during the period of 60 days.

[0051] As described above, after the C group and the D group are constituted, the operation control of the waste incineration plant is performed using the prediction model of the C group and the prediction model of the D group alternately every day during a 60-day period. In addition, after the prediction control using the prediction models of the C group and the D group is performed for a 60-day period, new prediction models are constructed and the prediction models used are updated, as described above.

[0052] That is, the evaluation section 104 also evaluates the prediction results when the prediction model of the C group selected by the selection section 105 and the prediction model of the newly constructed D group are used alternately. Further, the selection section 105 selects the prediction model used in a later period (a third prescribed period after the second prescribed period in which the prediction models of the C group and the D group are used) based on the evaluation results of the evaluation section 104.

[0053] The prediction model selected by the selection section 105 is the one evaluated higher in the A group and the B group, and high-accuracy prediction control can be achieved in the later period. In addition, data not used in the construction of the prediction model selected by the selection section 105 is reflected in the newly constructed prediction model. Therefore, according to the above-described structure, the prediction model that takes into account newly acquired data and enables high-accuracy prediction control in the later period can be selected.

[0054] As described above, according to the information processing apparatus 1, the prediction models used can be appropriately updated periodically, and the operation control of the waste incineration plant can be stably performed. Further, the groups after the C group are groups of prediction models newly constructed at the time of updating, and therefore it is preferable to perform simulation to verify the appropriateness of the prediction before use in an actual waste incineration plant.

[0055] In addition, the prediction using each prediction model can be performed in a period in which the result of the prediction is not used for control. For example, in the example described above, the prediction using the prediction model of the B group is performed in parallel with the prediction using the prediction model of the A group in the first day control. However, the prediction using the prediction model of the B group can be performed in a period in which the result of the prediction is not used for control. Figure 2 In the example described above, a plurality of devices used in the plant are included in the control target, and the plurality of devices are controlled based on the prediction results of the prediction models corresponding to the respective devices. In this case, the evaluation section 104 can evaluate the prediction results of the prediction models corresponding to the plurality of devices respectively. In this case, the selection section 105 selects the prediction model used in a later prescribed period for each of the plurality of devices based on the evaluation results of the evaluation section 104. According to this structure, the plurality of prediction models used in the control of the plurality of devices included in the plant can be appropriately updated.

[0056] In the example described above, a plurality of devices used in the plant are included in the control target, and the plurality of devices are controlled based on the prediction results of the prediction models corresponding to the respective devices. In this case, the evaluation section 104 can evaluate the prediction results of the prediction models corresponding to the plurality of devices respectively. In this case, the selection section 105 selects the prediction model used in a later prescribed period for each of the plurality of devices based on the evaluation results of the evaluation section 104. According to this structure, the plurality of prediction models used in the control of the plurality of devices included in the plant can be appropriately updated.

[0057] Furthermore, the properties of the waste being incinerated in a waste incineration plant may change from day to day. For example, some days may contain more stable combustible waste, while others may contain more unstable combustible waste. Additionally, the combustibility of waste can vary with the seasons. Therefore, sometimes the differences in the properties of the waste being incinerated are represented as differences in the evaluation of the prediction model. In such cases, the evaluation results of the evaluation unit 104 cannot be considered to directly reflect the performance differences of the prediction model itself.

[0058] Therefore, based on the above structure, by switching the prediction model used over a relatively short period of day, it is less likely that differences in garbage characteristics will be represented as differences in the evaluation of the prediction model. Thus, appropriate updates can be made based on the performance differences of the prediction model itself.

[0059] (The process of updating the prediction model used)

[0060] based on Figure 3 This describes the process by which the information processing device 1 updates the prediction model it uses (the information processing method). Figure 3 This is a flowchart illustrating an example of updating the prediction model used.

[0061] In S1, the update management unit 101 determines whether the used forecasting model needs to be updated. Specifically, the update management unit 101 determines whether the usage period of the currently used forecasting model has reached a predetermined period. This predetermined period can be set appropriately, for example, as follows: Figure 2 The setting shown is 60 days.

[0062] Furthermore, if the specified period has elapsed, the update management unit 101 determines that an update is required (yes in S1), and then proceeds to S2. On the other hand, if the specified period has not elapsed, the update management unit 101 determines that an update is not required (no in S1), and in this case, the determination in S1 is performed again after the specified period has elapsed.

[0063] In S2 (evaluation step), the evaluation unit 104 evaluates the prediction results of the above-mentioned prediction model for the specified period. Specifically, during the specified period, the prediction results are evaluated according to multiple unit periods (in... Figure 2 In the example, multiple forecasting models were used alternately for forecasting control (one day in the example), so that the evaluation department 104 evaluated the forecast results for each unit period.

[0064] In step S3 (selection step), the selection unit 105 selects the prediction model to be used in the next prediction control from among the multiple prediction models described above, based on the evaluation results of S2. Furthermore, the selection unit 105 may utilize... Figure 2Just as in the example where the prediction models selected from groups A and B are used to construct group C, multiple types of prediction models can be selected separately.

[0065] In S4, the teacher data generation unit 106 uses the data acquired during the aforementioned period to generate teacher data for constructing a new predictive model. Furthermore, the method for generating teacher data will be explained later.

[0066] In S5, the learning department 107 constructs a new predictive model using machine learning based on the teacher data generated in S4. Furthermore, Figure 2 The prediction model for group D in the example is equivalent to the prediction model built in S5.

[0067] In S6, the update management unit 101 designates the prediction model selected in S3 and the prediction model constructed in S5 as the prediction model for prediction control by the prediction unit 102 for the next and subsequent predictions. After S6, the process returns to S1.

[0068] (Methods for generating teacher data and machine learning)

[0069] Based on Figure 1 As explained, the information processing device 1 includes: a teacher data generation unit 106, a learning unit 107, a manual control detection unit 108, a reference information acquisition unit 109, and a status determination unit 110. Furthermore, correct data confirmation information 123 is stored in the storage unit 12.

[0070] The teacher data generation unit 106 associates the input data input to the prediction model with correct data indicating the control content that should be executed, creating teacher data for constructing the prediction model. Furthermore, the "control that should be executed" refers to the control that should be executed in the state (factory operating state or the operating state of the equipment being controlled), corresponding to the state represented by the input data, in order to ensure the continuous normal operation of the factory or equipment.

[0071] The learning unit 107 uses teacher data generated by the teacher data generation unit 106 to perform machine learning, thereby constructing a predictive model that predicts the control content that should be performed on the controlled object. Furthermore, the learning unit 107 stores the constructed predictive model in the predictive model DB122.

[0072] Thus, the information processing device 1, equipped with a teacher data generation unit 106, can automatically generate teacher data for constructing the prediction model based on the input data input to the prediction model and used for prediction. Furthermore, since the information processing device 1 includes a learning unit 107, it can construct a prediction model that reflects the input data actually used in predictive control.

[0073] The manual control detection section 108 detects manual control performed by manually operating the control target. Also, in a case where the manual control detection section 108 detects manual control, the teacher data generation section 106 can associate, with respect to input data input to the prediction model before the prescribed time of the detection, additional correct data corresponding to the detected manual control, to make teacher data for constructing the prediction model.

[0074] In a case where the prediction of the prediction model is erroneous or insufficient, in a case where it is a situation where control should be performed but a prediction result where control should be performed is not output, and the like, manual control is performed. That is, the "correct" control content that the prediction model should predict can be determined from the situation where manual control is performed, the content of the manual control performed. Therefore, according to the above-described structure, teacher data of high appropriateness can be automatically generated.

[0075] The reference information acquisition section 109 acquires, as a prediction result for reference that is not used in control of the control target, a prediction result obtained by inputting input data input to the prediction model to another prediction model that predicts control content that should be performed on the control target. The teacher data generation section 106 can associate, with respect to the input data, additional correct data corresponding to the prediction result acquired by the reference information acquisition section 109, to make teacher data for constructing the prediction model.

[0076] The prediction result of the prediction model used in control of the control target can be different from the prediction result of another prediction model that is not used in control of the control target. Also, in control performed based on the prediction result of the prediction model, there can be a case where control is not performed at a timing where control should originally be performed. Also, in such a case, there can be a case where another prediction model should appropriately predict the control. Therefore, according to the above-described structure, teacher data of high appropriateness can be automatically generated while taking into account the prediction result of another prediction model.

[0077] The state determination section 110 determines a plant operation state in which the control target is provided. The determination can be performed based on output values of various sensors and the like provided in the plant. As the plant operation state, for example, normal, occurrence of an abnormality, and the like can be cited. Also, the state determination section 110 stores the determination result in association with the date and time of the determination in the prediction result DB 121.

[0078] Also, the teacher data generation section 106 associates, with respect to input data input to the prediction model, additional correct data corresponding to a plant state after control of the control target based on the prediction result of the prediction model is performed, to make teacher data for constructing the prediction model. Further, the plant operation state can be determined by referring to the prediction result DB 121.

[0079] If control measures are implemented based on the predictions of a predictive model, and the subsequent factory operation is normal, the control is considered appropriate; otherwise, it is considered inappropriate. Thus, the factory operation status after control measures are implemented based on the predictive model's results indicates the "correctness" of the control content. Therefore, based on the above structure, highly appropriate teacher data can be automatically generated.

[0080] Correct data determination information 123 is information used to determine correct data based on various pre-set conditions. The teacher data generation unit 106 uses correct data determination information 123 to determine correct data. The following is based on... Figure 4 To explain in detail the correct data confirmation information 123.

[0081] (Example of correct data confirming information)

[0082] Correct data confirms information 123, for example, it could be like this: Figure 4 The information is presented in tabular form. Figure 4 This is a diagram illustrating an example of correct data confirming information 123. More specifically, in Figure 4 The diagram shows correct data determination information 123A and correct data determination information 123B. Correct data determination information 123A is used to determine correct data in the teacher data of the anomaly prevention model; correct data determination information 123B is used to determine correct data in the teacher data of the normal maintenance model.

[0083] (Details of the correct data used by the anomaly prevention model to determine information)

[0084] Correct data determination information 123A specifies a method for determining correct data corresponding to whether predictive control was implemented based on an anomaly prevention model (with / without predictive control) and the plant's operating status at a specified time. Furthermore, the specified time is defined as the time after the moment predictive control was implemented, representing the degree to which the effect of predictive control is reflected in the plant's operating status. For example, 10 minutes or 15 minutes after predictive control was implemented could be used as the specified time.

[0085] "Predictive control" corresponds to the case where predictive control is implemented using an anomaly prevention model for the object. In the case of "predictive control," the teacher data generation unit 106 generates teacher data by associating the input data that forms the basis of the predictive control with the control content determined using the correct data determination information 123A, as correct data. Furthermore, the input data that forms the basis of the predictive control refers to the data input to the anomaly prevention model when the prediction is performed.

[0086] On the other hand, "no prediction control" corresponds to a case where the prediction control of the abnormality prevention model of the object is not performed. In the case of "no prediction control", the teacher data generation unit 106 generates, at a time point at which a prescribed time elapses from the occurrence of the abnormality, teacher data in which, with respect to the input data input to the abnormality prevention model as the object, the control content determined using the correct data determination information 123A is associated as the correct data.

[0087] In the correct data determination information 123A, four states are prescribed as the plant operation states: "abnormality occurred", "other abnormality occurred", "abnormality prevented by prediction control", and "no prediction control, normal". Further, with respect to "abnormality occurred", although the abnormality can eventually be prevented by the prediction control, states in which the abnormality needs to be prevented for a prescribed time or more, or the abnormality is prevented by additional control such as manual control, are also included.

[0088] "Abnormality occurred" is a state in which the abnormality prevention model as the prediction object has occurred. On the other hand, "other abnormality occurred" is a state in which the abnormality prevention model as the object has occurred, which is not the prediction object. For example, in a case where the prediction control is performed based on the abnormality prevention model of the garbage amount shortage, in a case where the garbage amount shortage abnormality occurs, it is "abnormality occurred", and in a case where the non-combustible garbage abnormality occurs, it is "other abnormality occurred".

[0089] In the correct data determination information 123A, the determination method of the correct data with respect to the combination of "prediction control" and "abnormality occurred" is prescribed as "increase the control amount". Therefore, the teacher data generation unit 106 generates, in a case where the correct data determination information 123A is used, teacher data in which the control amount of the correct data is increased, in a case where the prediction control using the abnormality prevention model is performed but the abnormality occurs. In this case, since the prediction control content is correct but the control amount is insufficient, it is possible that the abnormality prevention cannot be performed, and therefore, by increasing the control amount of the correct data, teacher data with high appropriateness can be generated.

[0090] Further, it is sufficient to determine in advance how to increase the control amount. For example, a prescribed amount can be added, or the control amount can be increased by multiplying a prescribed rate larger than 1 to the actual control amount. This is also the same in a case where the correct data determination information 123B described later is used.

[0091] In addition, the teacher data generation section 106 can determine how much to increase the control amount, taking into account the content of the manual control and the additional control of the other prediction model. For example, the teacher data generation section 106, in a case where manual control is performed on the same control object as the abnormality prevention model of the object, can generate correct data of a control amount obtained by adding the control amount of the abnormality prevention model of the object and the control amount of the manual control. In addition, the teacher data generation section 106, in a case where predictive control is performed on the same control object as the abnormality prevention model of the object based on the prediction result of the other prediction model, can generate the same correct data as described above, with the condition that the factory returns to the normal state in a predetermined period thereafter.

[0092] In addition, in the correct data determination information 123A, the determination method of the correct data for the combination of "with predictive control" and "occurrence of other abnormality" is specified as "decrease control amount". Therefore, the teacher data generation section 106, in a case where the correct data determination information 123A is used, generates teacher data of a control amount that decreases the correct data, for a case where other abnormality occurs after predictive control using the abnormality prevention model is performed. In this case, since the abnormality of the prediction object does not occur, the content of the predictive control can be correct. However, it is also possible that the predictive control causes other abnormality for some reason. Therefore, by decreasing the control amount of the correct data, it is possible to generate teacher data of high appropriateness that has a control amount that does not cause either the abnormality of the prediction object or other abnormality to any extent, as correct data.

[0093] Furthermore, it is sufficient to determine in advance how to decrease the control amount. For example, it is possible to decrease a specified amount, or to decrease the control amount by multiplying the actual control amount by a specified rate smaller than 1. This is the same in a case where the correct data determination information 123B described later is used.

[0094] In addition, the teacher data generation section 106, in a case where "occurrence of other abnormality" in "with predictive control", can associate, as correct data, control that does not perform predictive control, with respect to the input data that becomes the basis of this predictive control, to make teacher data.

[0095] In addition, "abnormality prevented by predictive control" is a state where the result of performing predictive control using the abnormality prevention model of the object is that the abnormality is prevented. In the correct data determination information 123A, the determination method of the correct data for the combination of "with predictive control" and "abnormality prevented by predictive control" is specified as "no adjustment of control amount".

[0096] Therefore, the teacher data generation unit 106 generates, in the case where the correct data determination information 123A is used, teacher data in which the control amount in the prediction control using the abnormality prevention model is directly used as correct data, in the case where the result of the prediction control using the abnormality prevention model is prevented from abnormality. Thus, teacher data with high appropriateness can be generated. In this case, it is considered that the prediction control is appropriate.

[0097] In addition, "no prediction control, normal" is a state in which the prediction control using the abnormality prevention model of the object is performed, and then no abnormality occurs, but it is considered that no abnormality occurs even if the prediction control is not performed. For example, the teacher data generation unit 106 can determine that "no prediction control, normal" is satisfied in the case where the plant operation state does not change before and after the prediction control, in the case where the plant operation state after the prediction control does not become abnormal but is destabilized, and the like. Further, whether or not the plant operation state is destabilized can be determined by previously defining a state in which the plant operation state is destabilized on the basis of the detection value of the sensor or the like.

[0098] In the correct data determination information 123A, the determination method of the correct data for the combination of "with prediction control" and "no prediction control, normal" is defined as "not as teacher data". Therefore, the teacher data generation unit 106 does not generate teacher data in the case where the prediction control using the abnormality prevention model is performed and no abnormality occurs, but it is considered that no abnormality occurs even if the prediction control is not performed, in the case where the correct data determination information 123A is used. Thus, it is possible to prevent teacher data with low reliability from being generated. In this case, it is considered that the prediction control is not needed to be performed.

[0099] Further, the teacher data generation unit 106 can, in this case, generate teacher data in which the control in which the prediction control is not performed is associated as correct data, with respect to the input data that becomes the basis of the prediction control.

[0100] In addition, in the correct data determination information 123A, the determination method of the correct data for the combination of "no prediction control" and "abnormality" is defined as "correct data is control that can prevent abnormality".

[0101] Therefore, the teacher data generation unit 106 generates, in the case where the correct data determination information 123A is used, teacher data in which the control amount in the prediction control using the abnormality prevention model is directly used as correct data, in the case where the result of the prediction control using the abnormality prevention model is prevented from abnormality. Thus, teacher data with high appropriateness can be generated. In this case, it is considered that the prediction control is appropriate.

[0102] ​Further, the control for preventing the abnormality can be determined in consideration of the manual control and the control based on the other prediction model, and can be set in accordance with the operation input by the user via the input section 14. For example, when the abnormal state is recovered by the manual control or the control based on the other prediction model, the teacher data generation section 106 can take the control content as the correct data.

[0103] Further, in the correct data determination information 123A, the determination method of the correct data for the combination of "no prediction control" and "occurrence of the other abnormality" is also prescribed as "take the control capable of preventing the abnormality as the correct data". Therefore, the teacher data generation section 106 generates the teacher data taking the control for preventing the other abnormality as the correct data in the case where the other abnormality has occurred without the prediction control using the abnormality prevention model, in the case where the correct data determination information 123A is used. Further, the teacher data is not the teacher data for the abnormality prevention model but the teacher data for the other abnormality prevention model for preventing the other abnormality. In this way, the teacher data generation section 106 can also generate the teacher data for the other abnormality prevention model.

[0104] (Other Examples of the Correct Data Determination Method for the Abnormality Prevention Model)

[0105] The teacher data generation section 106 can determine the correct data in accordance with the prescribed rule, and can also determine the correct data without using the correct data determination information 123A.

[0106] Further, the teacher data generation section 106 can determine the correct data based on the prediction result of the other abnormality prevention model in the case where, for example, the prediction result is different between the object abnormality prevention model and the other abnormality prevention model. Further, the other abnormality prevention model is a prediction model that is not used in the prediction control, and is a prediction model that inputs the same input data as the object abnormality prevention model and performs prediction in parallel with the object abnormality prevention model. The other abnormality prevention model can also be two or more. The prediction result of the other abnormality prevention model is acquired from the prediction result DB 121 by the reference information acquisition section 109.

[0107] For example, the teacher data generation section 106 can determine the value of the control amount in which the prediction results of the respective abnormality prevention models are added as the control amount of the correct data in the case where the control object is the same between the object abnormality prevention model and the other abnormality prevention model but the control amount is different. For example, in the case where the prediction control based on the object abnormality prevention model is control to increase the operation speed of a certain device by 2 levels, the prediction result is set such that the other abnormality prevention model outputs control to decrease the operation speed of the device by 1 level. In this case, the teacher data generation section 106 can take control to increase the operation speed of the device by 1 level as the correct data.

[0108] In addition, for example, the teacher data generating unit 106 can evaluate the appropriateness of the prediction result of the abnormality prevention model of the object and the prediction result of other abnormality prevention models, and can take the evaluation result that is most appropriate as the correct data, or can take each prediction result whose evaluation of appropriateness is equal to or higher than a threshold value as the correct data.

[0109] The evaluation method of the appropriateness of the prediction result is not particularly limited. For example, it can be evaluated as appropriate if the input data input to the abnormality prevention model is within the distribution range of the teacher data used in the learning of the abnormality prevention model, and can be evaluated as inappropriate if it is outside the range. In addition, when the abnormality prevention model outputs a value indicating the prediction accuracy as the prediction result, it can be determined that the prediction result whose value is larger is appropriate.

[0110] In addition, regarding the appropriateness of the prediction result of the abnormality prevention model of the object, it can also be determined based on the plant operation state after the prediction control based on the prediction result. That is, it can be determined that the prediction result is appropriate if the plant operation state after the prediction control based on the prediction result is stabilized, and can be determined as inappropriate if it is not stabilized. Furthermore, the evaluation of the appropriateness of the prediction result can be performed by the teacher data generating unit 106, or a block different from the teacher data generating unit 106 can be provided and become a structure in which the evaluation is performed by the block.

[0111] (Details of the correct data determination information for the normal maintenance model)

[0112] The correct data determination information 123B for the normal maintenance model specifies the determination method of the correct data corresponding to whether the prediction control is performed based on the normal maintenance model (with prediction control / without prediction control) and the additional control in the specified time. Furthermore, the specified time is the same as the correct data determination information 123A described above. In addition, the input data associated with the correct data is different in each case of "with prediction control" and "without prediction control", and is the same as the case of the correct data determination information 123A described above.

[0113] In addition, the above-mentioned additional control refers to at least one of the control performed based on the prediction result of the prediction model other than the normal maintenance model of the object, and the manual control performed by hand. In the correct data determination information 123B, the following five kinds are specified as the additional control: "the same control", "the opposite control", "other control", "no additional control (normal maintenance)", and "no additional control (cannot maintain normal)".

[0114] "the same control" is a control that is the same as the predicted control of the normal maintenance model of the object. Furthermore, the same control is a control that is the same in the control object and the direction of the control, and "the opposite control" is a control that is the same in the control object but opposite in the direction of the control. For example, a control that increases the garbage feeding speed of a garbage feeding device that feeds garbage to an incinerator by one level, and a control that increases the garbage feeding speed by two levels are the same control, and a control that decreases the garbage feeding speed is the opposite control to the above two controls.

[0115] In the correct data determination information 123B, the determination method of the correct data for the combination of "with predicted control" and "the same control" is prescribed as "increase the control amount". Therefore, the teacher data generation section 106 generates teacher data that increases the control amount of the correct data, in the case where the same control as the predicted control using the normal maintenance model is performed after the predicted control, using the correct data determination information 123B. In this case, since the predicted control content is correct but the control amount is insufficient, it is considered that the additional control is performed, and therefore teacher data with high appropriateness can be generated by increasing the control amount of the correct data.

[0116] In addition, in the correct data determination information 123B, the determination method of the correct data for the combination of "with predicted control" and "the opposite control" is prescribed as "decrease the control amount". Therefore, the teacher data generation section 106 generates teacher data that decreases the control amount of the correct data, in the case where the opposite control is performed after the predicted control using the normal maintenance model is performed, using the correct data determination information 123B. In this case, since the predicted control content is correct but the control amount is excessive, it is considered that the opposite control is performed, and therefore teacher data with high appropriateness can be generated by decreasing the control amount of the correct data.

[0117] Furthermore, in the case where the control amount of the opposite control is larger than the control amount of the predicted control, it is considered that the opposite control should be performed at the time of the predicted control. Therefore, in this case, the teacher data generation section 106 can also take the opposite control as the correct data. In addition, the teacher data generation section 106 can take the control amount in the correct data as the difference between the control amount of the predicted control and the control amount of the additional control.

[0118] In addition, "the other control" is a control that is a control for a different control object from the predicted control of the normal maintenance model of the object. For example, in the case where the predicted control of the normal maintenance model of the object is a control that increases the garbage feeding speed, a control that changes the running speed of the grate is the other control. In the correct data determination information 123B, the determination method of the correct data for the combination of "with predicted control" and "the other control" is prescribed as "decrease the control amount".

[0119] The method of determining the correct data of the combination of "other control" is defined as "adjusted according to the content of the other control". In addition, the adjustment method corresponding to the content of the other control is defined separately.

[0120] The teacher data generation section 106 generates teacher data by adjusting the correct data in the case where the other control is performed after the prediction control using the normal maintenance model is performed, in the case where the correct data determination information 123B is used. For example, in the case where the other control is control that produces the same effect as the prediction control of the normal maintenance model of the object (for example, increases the amount of steam generated from the incinerator, etc.), the teacher data generation section 106 can generate teacher data that increases the amount of control of the correct data. In addition, for example, in the case where the other control is control that produces the opposite effect to the prediction control of the normal maintenance model of the object, the teacher data generation section 106 can generate teacher data that reduces the amount of control of the correct data.

[0121] In addition, it is also possible that the prediction control of the normal maintenance model of the object cannot maintain normal or an abnormality occurs for some reason. Also, sometimes manual control or the like is performed in order to recover from such a state to a normal state. In such a case, after the prediction control of the normal maintenance model of the object, an abnormality is detected, and after that, manual control is detected.

[0122] In such a case, the control that does not perform prediction control can be associated as correct data with respect to the input data that becomes the basis of the prediction control of the normal maintenance model of the object, and made into teacher data. In addition, considering the possibility that an abnormality occurs due to the control amount being too large, the teacher data generation section 106 can also generate correct data that reduces the control amount compared to the prediction control of the normal maintenance model of the object.

[0123] In addition, both "no additional control (normal maintenance)" and "no additional control (cannot maintain normal)" in the correct data determination information 123B correspond to the case where no additional control is performed. Among these, "no additional control (normal maintenance)" corresponds to the case where the normal state is maintained although no additional control is performed, and "no additional control (cannot maintain normal)" corresponds to the case where the normal state is not maintained although no additional control is performed.

[0124] In the correct data determination information 123B, the determination method of the correct data for the combination of "with prediction control" and "without additional control (normal maintenance)" is prescribed as "not adjusting the control amount". Therefore, the teacher data generation section 106 generates, in a case where the correct data determination information 123B is used, teacher data directly using the control amount in the prediction control using the normal maintenance model for the case where the normal state is maintained by the prediction control using the normal maintenance model. Thus, teacher data with high appropriateness can be generated. Because in this case, it is appropriate to consider the prediction control content. Further, for the cases other than this case, teacher data can be generated as long as it is appropriate to judge that the prediction control using the normal maintenance model is appropriate.

[0125] Further, in the correct data determination information 123B, in either of "with prediction control" and "without prediction control", the determination method of the correct data in the case of "without additional control (cannot maintain normal)" is prescribed as "not as teacher data".

[0126] Therefore, the teacher data generation section 106 does not generate teacher data for the case where additional control is not performed and the normal state cannot be maintained. Thus, it is possible to prevent teacher data with low reliability from being generated. Because in this case, it is difficult to automatically determine what kind of control is performed to maintain the normal state. Further, the teacher data generation section 106 can also, in this case, correlate, as correct data, the control content input by the user via the input section 14 to the input data that becomes the basis of the prediction control, to make teacher data.

[0127] Further, in the correct data determination information 123B, the determination method of the correct data for the combination of "without prediction control" and

[0128] "other control" is prescribed as "applying the content of additional control". Therefore, the teacher data generation section 106 generates, in a case where the correct data determination information 123B is used, teacher data using the content of additional control as correct data for the case where additional control is performed, without performing prediction control using the normal maintenance model. Because in this case, it is considered that the normal maintenance model should predict the control performed as additional control in advance.

[0129] (Other Examples of the Correct Data Determination Method for the Normal Maintenance Model)

[0130] The teacher data generation section 106 can also not perform the additional control, but determine the correct data based on other prediction results made at the same or the same degree of timing as the normal maintenance model of the object. For example, the teacher data generation section 106 can determine the correct data based on the prediction results of other prediction models in a case where the other prediction models also make predictions in parallel with the normal maintenance model of the object. The other prediction models can be the normal maintenance model, the abnormality prevention model, or both. The prediction results of the other prediction models are acquired from the prediction result DB 121 by the reference information acquisition section 109.

[0131] In this example as well as in the above-described "other example of the correct data determination method for the abnormality prevention model", the teacher data generation section 106 can determine the correct data based on the evaluation results regarding each prediction result.

[0132] (Processing flow of generating teacher data (for abnormality prevention model))

[0133] Based on Figure 5 The processing flow of generating the teacher data for the abnormality prevention model will be described. Figure 5 is a flowchart showing one example of the processing of generating the teacher data. The processing corresponds to S4 of Figure 3 . Further, when there are a plurality of abnormality prevention models as in the Figure 2 example, the processing of S11 to S17 is performed for each abnormality prevention model. Figure 5

[0134] In S11, the teacher data generation section 106 extracts, from the prediction result DB 121, the prediction control performed according to the output of the abnormality prevention model during a prescribed period, and the input data that is the basis of the prediction control. The input data extracted here is associated with the correct data for the case of "with prediction control" in the correct data determination information 123A of Figure 4

[0135] In S12, the manual control detection section 108 extracts, from the prediction result DB 121, the manual control performed at a prescribed time after the prediction control extracted in S11 for each of the prediction controls. The extraction result of S12 is used for the determination of the correct data for the case of "abnormality occurred" in the correct data determination information 123A of Figure 4

[0136] ​​​In S13, the reference information acquisition section 109 extracts, from the prediction result DB 121, the prediction results of the other prediction models at the prescribed time after the prediction control extracted in Sll, respectively. Further, the prediction results of the other prediction models can also include the prediction results used for control and the prediction results not used for control. In addition, the other prediction models can be abnormality prevention models, normality maintenance models, or both. The extraction results of S13 are used for Figure 4 determination of the correct data in the correct data determination information 123A in the case of "abnormality occurrence" and "other abnormality occurrence".

[0137] In S14, the teacher data generation section 106 extracts, for the prediction control extracted in Sll, the determination results of the plant operation state at the prescribed time after the prediction control, respectively. Further, the plant operation state is determined by the state determination section 110, and the determination result is stored in the prediction result DB 121.

[0138] In the case of using the correct data determination information 123A of Figure 4 the state determination section 110 determines whether an abnormality occurs in the plant or the plant is in a normal state. Also, the state determination section 110 determines the kind of abnormality when it is determined that the plant is not in a normal state, and stores the determination result in the prediction result DB 121. Thus, the teacher data generation section 106 can determine, by referring to the prediction result DB 121, which of "abnormality occurrence" and "other abnormality occurrence" the plant operation state belongs to.

[0139] In addition, the teacher data generation section 106 can determine, for the state of "normal", which of "abnormality prevention by prediction control" and "normal without prediction control" the plant operation state belongs to, based on the extraction results of S12 and S13.

[0140] In S15, the teacher data generation section 106 extracts, from the prediction result DB 121, the input data satisfying a prescribed condition. The input data extracted in Sll and the input data extracted in S15 differ in that, whereas the former is based on the input data on which prediction control is performed, the latter is not based on the input data on which prediction control is performed. As for the input data extracted in S15, as long as it belongs to Figure 4 "no prediction control" and the combination of "abnormality occurrence" or "other abnormality occurrence" in the correct data determination information 123A, it becomes teacher data through the following processing.

[0141] The above-described conditions are conditions in which the teacher data generation section 106 can extract input data for which the corresponding correct data can be determined. For example, when the state in which the abnormality occurred is included in the state extracted by S14, the teacher data generation section 106 can extract input data input to the abnormality prevention model at a time point at which a prescribed time elapses from when the abnormality occurred.

[0142] In S16, the teacher data generation section 106 determines correct data corresponding to the extraction results and the detection results of S12 to S14, using the correct data determination information 123A, for the input data extracted in S11 and the input data extracted in S15, respectively. The method of determining the correct data is as described above.

[0143] In S17, the teacher data generation section 106 generates teacher data by associating the correct data determined in S16 with the input data extracted in S11 and the input data extracted in S15, respectively. Thereby, Figure 5 the processing of S11 to S17 ends, and thereafter, the abnormality prevention model (S5) is constructed using the generated teacher data. Figure 3

[0144] The processing flow of generating teacher data for the normal maintenance model will be described below. The processing flow of generating teacher data for the normal maintenance model is the same as that described above, and will be described below based on

[0145] Figure 5 Figure 2 When there are a plurality of normal maintenance models as in the example, the processing of S11 to S17 is performed for each normal maintenance model. Figure 5

[0146] In S11, the teacher data generation section 106 extracts, from the prediction result DB 121, prediction control performed in accordance with the output of the normal maintenance model during a prescribed period, and input data that becomes the basis of the prediction control. The input data extracted here is associated with correct data of the case in which "there is prediction control" in the correct data determination information 123B of Figure 4

[0147] In S12, the manual control detection section 108 extracts, from the prediction result DB 121, manual control performed a prescribed time after the prediction control extracted in S11, for each prediction control. In addition, in S13, the reference information acquisition section 109 extracts, from the prediction result DB 121, prediction results of other prediction models a prescribed time after the prediction control extracted in S11, for each prediction control.

[0148] The teacher data generation section 106 determines correct data corresponding to the extraction results of S12 and S13, using the correct data determination information 123B, for the input data extracted in S11. Figure 4 ​​​​​the correct data determination information 123B, the teacher data generation section 106 determines, based on the extraction results of S14, which of "no additional control (normal maintenance)" or "no additional control (unable to maintain normal)" belongs to in the case of "no additional control".

[0149] In S14, the teacher data generation section 106 extracts, for each of the prediction controls extracted in S11, a detection result of the plant operation state at a prescribed time after the prediction control. In the case of using the correct data determination information 123B, the teacher data generation section 106 determines, based on the extraction results of S14, which of "no additional control (normal maintenance)" or "no additional control (unable to maintain normal)" belongs to in the case of "no additional control". Figure 4

[0150] In S15, the teacher data generation section 106 extracts input data satisfying a prescribed condition from the prediction result DB 121. The input data extracted in S11 and the input data extracted in S15 differ in that, whereas the former is based on the input data on which the prediction control is performed, the latter is not based on the input data on which the prediction control is performed. As for the input data extracted in S15, as long as it belongs to the combination of "no prediction control" and "other control" in the correct data determination information 123B, it becomes the teacher data through the following processing. Figure 4

[0151] The input data extracted in S15 can be, for example, input data input to the normal maintenance model at a time point that is a prescribed time back from when an abnormality occurred, when the abnormality-occurred state is included in the operation state checked through S14.

[0152] In S16, the teacher data generation section 106 determines, using the correct data determination information 123B, the correct data corresponding to the extraction results and the detection results of S12 to S14, for each of the input data extracted in S11 and the input data extracted in S15. The method of determining the correct data is as described above. Further, the processing of S17 is the same as the processing at the time of generating the teacher data for the abnormality prevention model described above, and thus the description thereof is not repeated.

[0153] (Based on Software Implementation Example)

[0154] The control module of the information processing apparatus 1 (particularly, each section included in the control section 10) can be realized by a logic circuit (hardware) formed in an integrated circuit (IC chip) or the like, or by software.

[0155] ​​In the latter case, the information processing apparatus 1 has a computer that executes a command of a program (information processing program) that realizes each function. The computer has, for example, one or more processors, and has a storage medium that is readable by the computer and stores the program. In the computer, the processor reads the program from the storage medium and executes it, thereby realizing the object of the present application. As the processor, for example, a CPU (Central Processing Unit) can be used. As the storage medium, a "non-transitory tangible medium" such as a magnetic tape, an optical disk, a card, a semiconductor memory, a programmable logic circuit, or the like can be used in addition to a ROM (Read Only Memory) or the like. In addition, a RAM (Random Access Memory) or the like that expands the program can be provided. In addition, the program can be provided to the computer via any transmission medium (a communication network, a radio wave, or the like) that can transmit the program. Furthermore, one embodiment of the present application can also be realized in the form of a data signal embedded in a carrier wave by electronically transmitting the program.

[0156] (Modified Example)

[0157] The execution subject of each process described in the above-described embodiments can be appropriately changed. For example, in the flowchart of Figure 3 S1, another information processing apparatus different from the information processing apparatus 1 can be caused to execute the process of S1, and still another information processing apparatus can be caused to execute the process of S2, and the information processing apparatus 1 can be caused to execute the process of S3. Similarly, another information processing apparatus can be caused to execute the process of S4 and the process of S5, respectively. Figure 5 The flowchart of Figure 3 Figure 5 Each information processing method described in

[0158] The present application is not limited to the above-described embodiments, and various changes can be made within the scope of the claims, and embodiments obtained by appropriately combining the technical solutions disclosed in different embodiments, respectively, are also included in the technical scope of the present application.

[0159] Explanation of Reference Numerals

[0160] 1 - information processing apparatus; 104 - evaluation section; 105 - selection section; 106 - teacher data generation section; 108 - manual control detection section; 109 - reference information acquisition section.​

Claims

1. An information processing apparatus, characterized by, Possessing: an evaluation section that evaluates a prediction result for each unit period when a control object is controlled by alternately using a plurality of prediction models that predict a control content that should be performed on the control object, in a plurality of unit periods that constitute a first prescribed period; and a selection section that selects, from among the plurality of prediction models, a prediction model to be used in a second prescribed period after the first prescribed period, based on an evaluation result of the evaluation section, the evaluation section evaluates a prediction result for each unit period when the control object is controlled by alternately using the prediction model selected by the selection section, and a prediction model constructed using input data input to the prediction model at the time of prediction in the first prescribed period, the selection section selects, based on an evaluation result of the evaluation section, a prediction model to be used in a third prescribed period after the second prescribed period.

2. The information processing apparatus according to claim 1, wherein a plurality of devices used in one factory are included in the control object, the plurality of devices are controlled based on a prediction result of a prediction model corresponding to each device, the evaluation section evaluates a prediction result of a prediction model corresponding to each of the plurality of devices, the selection section selects, based on an evaluation result of the evaluation section, a prediction model to be used in the second prescribed period for each of the plurality of devices.

3. The information processing apparatus according to claim 1, wherein a teacher data generation section that associates correct data indicating a control content that should be performed with input data input to the prediction model, to make teacher data for constructing the prediction model, is possessed.

4. The information processing apparatus according to claim 3, wherein a manual control detection section that detects manual control performed on the control object by hand is possessed, the teacher data generation section associates correct data corresponding to the detected manual control with input data input to the prediction model before the detected prescribed time, to make teacher data for constructing the prediction model, when the manual control detection section detects manual control.

5. The information processing apparatus according to claim 3 or 4, wherein a reference information acquisition section that acquires, as a prediction result for reference, a prediction result obtained by inputting input data input to the prediction model to another prediction model that predicts a control content that should be performed on the control object is possessed, the teacher data generation section associates correct data corresponding to the prediction result acquired by the reference information acquisition section with the input data, to make teacher data for constructing the prediction model.

6. The information processing apparatus according to claim 3 or 4, wherein The teacher data generation section associates correct data corresponding to an operation state of a plant having the control target after control of the control target based on a prediction result of the prediction model is performed with input data input to the prediction model, and creates teacher data for constructing the prediction model.

7. An information processing apparatus, characterized by comprising: Possessing: an evaluation section that evaluates a prediction result of each unit period when the control target is controlled by alternately using a plurality of prediction models that predict a control content that should be performed on the control target, for each unit period constituting a first prescribed period; and a selection section that selects, from among the plurality of prediction models, a prediction model to be used in a second prescribed period after the first prescribed period, based on an evaluation result of the evaluation section, the prediction models include: an abnormality prevention model that predicts a case where a plant having the control target becomes an abnormal operation state, and predicts a control content that should be performed on the control target in order to prevent the plant from becoming an abnormal operation state; and a normal maintenance model that predicts a control content that should be performed on the control target in order to maintain the plant in a normal operation state, the evaluation section evaluates a prediction result of each of the plurality of abnormality prevention models and the plurality of normal maintenance models, the selection section selects an abnormality prevention model and a normal maintenance model to be used in the second prescribed period, based on the evaluation result of the evaluation section.

8. An information processing method executed by one or a plurality of information processing apparatuses, characterized by, including: an evaluation step of evaluating a prediction result of each unit period when the control target is controlled by alternately using a plurality of prediction models that predict a control content that should be performed on the control target, for each unit period constituting a first prescribed period; and a selection step of selecting, from among the plurality of prediction models, a prediction model to be used in a second prescribed period after the first prescribed period, based on an evaluation result of the evaluation step, in the evaluation step, a prediction result of each unit period when the control target is controlled by alternately using a prediction model selected in the selection step and a prediction model constructed using input data input to the prediction model at the time of prediction in the first prescribed period, for a plurality of unit periods constituting the second prescribed period is evaluated, in the selection step, a prediction model to be used in a third prescribed period after the second prescribed period is selected, based on the evaluation result of the evaluation step.

9. A computer-readable storage medium storing an information processing program for causing a computer to function as the information processing apparatus according to claim 1, i.e., for causing the computer to function as the evaluation section and the selection section.

Images