JP2015179454A - Prediction system, monitoring system, operation support system, gas turbine installation, and prediction method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a prediction system and the like capable of accurately predicting a predetermined physical quantity related to installation.SOLUTION: A prediction system comprises: a database storing process data; and a computing unit calculating a predetermined physical quantity as a prediction value on the basis of the process data, and the computing unit constructs a plurality of prediction models for calculating prediction values using the process data contained in a data set as input variables and calculates the prediction values from the constructed prediction models, respectively (S21); diagnoses whether process data D1 used as the input variable is abnormal for each of the constructed prediction models (S22); makes setting such that the prediction model including the process data D1 diagnosed as being abnormal as the input variable is not used (S24); and selects one of the prediction values predicted by the prediction models in response to preset priorities of the prediction models (S25).

Description

  The present invention relates to a prediction system, a monitoring system, an operation support system, a gas turbine facility, and a prediction method for predicting a predetermined physical quantity related to equipment.

  Conventionally, an anomaly detection system that detects an anomaly using observation data and learning data is known (see, for example, Patent Document 1). This abnormality detection system detects an abnormality in observation data based on the degree of separation between the observation data and the learning data.

International Publication No. 2010/095314

  However, the abnormality detection system described in Patent Document 1 cannot detect quantitatively because the presence or absence of abnormality is binarized and detected from observation sensor data using a plurality of discriminators. At this time, in order to quantitatively predict a predetermined physical quantity related to the equipment, a prediction model is created using various process data that can be acquired from the equipment as input variables, and the predetermined physical quantity is predicted using the created prediction model. Can be calculated quantitatively as In this case, when there is an abnormality in the process data, if the prediction model is constructed using the abnormal process data as an input variable, the prediction accuracy of the prediction model may be reduced.

  Therefore, an object of the present invention is to provide a prediction system, a monitoring system, an operation support system, a gas turbine facility, and a prediction method that can accurately predict a predetermined physical quantity related to the facility.

  The prediction system of the present invention uses a storage unit that stores various process data that can be acquired in a facility, and a predetermined physical quantity that is predicted for the facility based on the process data stored in the storage unit as a predicted value. An arithmetic unit for calculating, and the arithmetic unit includes a plurality of process data selected in advance among various process data stored in the storage unit as a single data set for a plurality of sets. Obtaining, using the process data included in the obtained data set of each set as an input variable, constructing a prediction model for calculating the prediction value, respectively, calculating the prediction value from each of the constructed prediction models Diagnosing the presence or absence of abnormality in the process data used as the input variable for each of the constructed prediction models The prediction model including the process data diagnosed as having an abnormality as the input variable is set not to be used, and the prediction predicted by the prediction model according to a plurality of preset priorities of the prediction models It is characterized by selecting a value.

  The prediction method of the present invention is a prediction method for deriving a predetermined physical quantity related to the facility as a predicted value based on various process data that can be acquired in the facility, and is selected in advance from the various process data. A plurality of the process data obtained as a set of data sets, a data set acquisition step of acquiring a plurality of sets, and the process data included in the acquired data sets of each set as input variables, For each of the prediction model construction step for constructing each prediction model to be calculated, the prediction value calculation step for calculating the prediction value from each of the plurality of constructed prediction models, and each of the plurality of constructed prediction models An input variable abnormality diagnosis step for diagnosing the presence or absence of abnormality in the process data used as a variable, and a diagnosis that there is an abnormality The model non-use setting step of setting that the prediction model including the process data as the input variable is not used, and the prediction of the prediction model according to the preset priorities of the prediction models And a predicted value selection step of selecting a predicted value.

  According to this configuration, a prediction model including process data diagnosed as having an abnormality as an input variable can be set to an unused state. For this reason, when selecting a predicted value calculated from a plurality of prediction models, it is possible to select a highly accurate predicted value calculated by a prediction model using appropriate process data having no abnormality as an input variable. The predetermined physical quantity related to the facility is, for example, temperature, pressure, flow rate, vibration, etc. in a predetermined part of the facility (compressor, combustor, turbine, internal fluid, etc. in the case of a gas turbine facility), in particular. It is not limited.

  Another prediction system of the present invention predicts a predetermined physical quantity predicted for the facility based on a storage unit that stores various process data that can be acquired in the facility, and the process data stored in the storage unit. A calculation unit that calculates a value, and the calculation unit acquires a plurality of the process data selected in advance as a set of data among the various process data stored in the storage unit, Diagnosing the presence or absence of abnormality of the process data included in the acquired data set, omitting the process data diagnosed as abnormal, and using the remaining process data included in the data set as an input variable, the predicted value Is constructed, and the prediction value is calculated from the constructed prediction model.

  Further, another prediction method of the present invention is a prediction method for deriving a predetermined physical quantity related to the facility as a predicted value based on various process data that can be acquired in the facility, and among the various process data, A data set acquisition step of acquiring a plurality of process data selected in advance as a set of data sets, an input variable abnormality diagnosis step of diagnosing whether or not the process data included in the acquired data set is abnormal, From the constructed prediction model, the prediction model construction step of constructing a prediction model for calculating the prediction value using the remaining process data included in the data set as input variables, excluding the process data diagnosed as being present A predicted value calculating step of calculating the predicted value.

  According to this configuration, it is possible to construct a prediction model by omitting process data diagnosed as having an abnormality and using the remaining process data as input variables. For this reason, an accurate predicted value can be calculated by a prediction model using appropriate process data without abnormality as an input variable.

  In this case, the storage unit stores alternative information that associates the predetermined process data with alternative process data that is an alternative to the predetermined process data, and the arithmetic unit diagnoses that there is an abnormality. When there is process data, based on the substitute information, the process data diagnosed as having an abnormality is substituted with the substitute process data, and the process model including the substituted process data is substituted as an input variable. It is preferable to construct.

  According to this configuration, alternative process data can be used instead of the process data diagnosed as having an abnormality. Then, the process data diagnosed as having an abnormality can be omitted, and a prediction model can be constructed using the alternative process data and the remaining process data as input variables. For this reason, an accurate predicted value can be calculated by a prediction model using appropriate process data without abnormality as an input variable.

  In this case, when diagnosing the presence or absence of abnormality in the process data, the calculation unit uses the plurality of process data in different combinations among the plurality of process data included in the data set as the input variable. A plurality of the prediction models for calculating values are constructed according to the number of combinations, and errors in the prediction accuracy of the plurality of prediction models are respectively calculated, and among the calculated plurality of errors, the set error equal to or greater than a preset setting error It is preferable to identify the process data having an abnormality from among the plurality of process data included in the prediction model that is an error.

  According to this configuration, since the calculation unit can identify the process data having an abnormality from the process data included in the prediction model having a large error, it is possible to accurately diagnose whether the process data is abnormal. it can.

  A monitoring system according to the present invention includes the above prediction system and a monitoring device that monitors the facility using the prediction value calculated by the prediction system.

  According to this configuration, the monitoring device can monitor the facility using the accurate predicted value calculated by the prediction system. For this reason, the monitoring apparatus can issue a warning or the like to the operator before the actually measured value becomes the threshold value when the predicted value becomes the abnormal threshold value. For this reason, the monitoring device can perform monitoring while predicting the behavior of the facility.

  The driving support system of the present invention includes the above-described prediction system and a driving support device that supports driving of the facility using the predicted value calculated by the prediction system.

  According to this configuration, the driving support device can support the operation of the facility using the accurate predicted value calculated by the prediction system. For this reason, the driving assistance device can assist the facility to operate efficiently based on the predicted value.

  A gas turbine facility according to the present invention includes at least one of the prediction system, the monitoring system, and the operation support system, and a gas turbine as the facility.

  According to this configuration, it is possible to predict a predetermined physical quantity related to the gas turbine, to monitor while predicting the behavior of the gas turbine, or to assist the gas turbine in an efficient operation.

FIG. 1 is a schematic configuration diagram of a gas turbine facility according to a first embodiment. FIG. 2 is a flowchart regarding the construction of the prediction model of the first embodiment. FIG. 3 is a flowchart regarding abnormality diagnosis of the first embodiment. FIG. 4 is a flowchart regarding construction of a prediction model and abnormality diagnosis according to the second embodiment. FIG. 5 is a flowchart regarding abnormality diagnosis of the third embodiment.

  Embodiments according to the present invention will be described below in detail with reference to the drawings. Note that the present invention is not limited to the embodiments. In addition, constituent elements in the following embodiments include those that can be easily replaced by those skilled in the art or those that are substantially the same. Furthermore, the constituent elements described below can be combined as appropriate, and when there are a plurality of embodiments, the embodiments can be combined.

  FIG. 1 is a schematic configuration diagram of a gas turbine facility according to a first embodiment. FIG. 2 is a flowchart regarding the construction of the prediction model of the first embodiment. FIG. 3 is a flowchart regarding abnormality diagnosis of the first embodiment.

  As illustrated in FIG. 1, the prediction system 20 according to the first embodiment is incorporated in a gas turbine facility 1 such as a gas turbine power plant, for example, and is a system that predicts a predetermined physical quantity related to the gas turbine facility 1. Yes. First, the gas turbine facility 1 will be described prior to the description of the prediction system 20.

  The gas turbine equipment 1 includes a gas turbine (equipment) 10, a prediction system 20, a monitoring system 30, and an operation support system 40. At this time, the monitoring system 30 is configured to include the prediction system 20 and the monitoring device 31, and the driving support system 40 is configured to include the prediction system 20 and the driving support device 41. ing.

  The gas turbine 10 includes a compressor, a combustor, and a turbine (not shown), and the compressor and the turbine are coupled to each other by a rotating shaft so as to be integrally rotatable. A compressor compresses the taken-in air and supplies the compressed air compressed toward the combustor. The combustor mixes and burns the compressed air supplied from the compressor and fuel. The turbine is rotated by the combustion gas supplied from the combustor.

  The gas turbine 10 is provided with various sensors for detecting the operation status of the gas turbine 10. Examples of the various sensors include a compressor, a combustor, a pressure sensor 11 that detects a pressure in the turbine, and a temperature sensor 12 that detects the temperature. Data detected by these various sensors 11 and 12 is input to the database 22 as process data D1. Further, the gas turbine 10 is provided with a valve 13 for adjusting the fuel flow rate, a guide vane 14 for adjusting air taken into the compressor, and the like. Data relating to the opening degree of the valve 13 and the angle of the guide vanes 14 are input to the database 22 as process data D1. Here, in addition to the data acquired by the various sensors 11 and 12, the data related to the valves 13 and the guide vanes 14, data input by the operator when operating the gas turbine 10 is input to the database 22 as process data D <b> 1. The That is, the process data D1 input to the database 22 is not particularly limited as long as it is data related to the gas turbine 10.

  Next, the prediction system 20 will be described. The prediction system 20 includes a prediction device 21 and the database (storage unit) 22 that stores the process data D1. The prediction device 21 predicts a predetermined physical quantity related to the gas turbine 10 based on the process data D1 stored in the database 22. Examples of the predetermined physical quantity include temperature, pressure, flow rate, vibration, and the like in a compressor, a combustor, a turbine, and an internal fluid of the gas turbine 10. The predetermined physical quantity is not particularly limited as long as it is a physical quantity related to the gas turbine 10.

  The prediction device 21 is connected to the database 22 and includes a calculation unit 25 that calculates a predetermined physical quantity to be predicted as a predicted value. The calculation unit 25 includes an integrated circuit, and executes various calculations by executing a predetermined program. Specifically, the arithmetic unit 25 calculates a predetermined physical quantity to be predicted for the gas turbine 10 as a predicted value based on the process data D1 stored in the database 22 by executing a prediction program. The calculation of the predicted value by the prediction device 21 will be described later.

  The database 22 stores various process data D1 input from the gas turbine 10. In the database 22, process data D1 is accumulated at a predetermined sampling period. The database 22 is connected to the prediction device 21 and outputs predetermined process data D1 to the prediction device 21 in response to a request from the prediction device 21.

  The monitoring device 31 is connected to the prediction system 20 and monitors the operation status of the gas turbine 10 using the predicted value predicted by the prediction system 20. The monitoring device 31 includes a notification device such as an alarm device, a lamp, and a display device that notifies the operator of predetermined information. When the predicted value predicted by the prediction system 20 exceeds a preset threshold value, the monitoring device 31 performs abnormality notification by the notification device to the operator. As described above, the monitoring device 31 predicts an abnormality of the gas turbine 10 before the actually measured value becomes the abnormality threshold value.

  The driving support device 41 is connected to the prediction system 20 and uses the predicted value predicted by the prediction system 20 to support the operation of the gas turbine 10. For example, the operation support device 41 provides support for the operation of the gas turbine 10 to the operator so that the predicted value predicted by the prediction system 20 becomes a predetermined physical quantity that allows the gas turbine 10 to operate efficiently. Do.

  Next, prediction value calculation control by the calculation unit 25 of the prediction device 21 will be described. The calculation unit 25 of the prediction device 21 performs data extraction control, prediction model construction control, data diagnosis control, prediction value calculation control, and prediction value selection control when calculating the prediction value.

  The calculation unit 25 executes data extraction control, and extracts (acquires) only a plurality of sets from the various process data D1 stored in the database 22 with the plurality of process data D1 as one set. To do. The plurality of extracted data sets have different numbers or types of process data D1 included. Further, the data set is appropriately prepared according to the predicted physical quantity. Note that the process data D1 included in the data set is acquired in advance. Here, when the calculation unit 25 automatically selects the process data D1 included in the data set, for example, the calculation unit 25 may select the process data D1 having a high correlation coefficient with respect to the physical quantity calculated as the predicted value. The process data D1 may be selected at random, or the process data D1 previously selected by the operator may be selected. In addition, the process data D1 may be selected by using a statistical method such as a variable reduction method or a stepwise method, or the process data D1 may be selected in a round robin manner.

  Moreover, the calculating part 25 builds a prediction model based on the extracted data set by performing prediction model construction control. As the prediction model, for example, there is a multivariate analysis model in which a plurality of process data D1 is an input variable and a predetermined process data D1 is an output variable. Note that the prediction model is not limited to the multivariate analysis model, and other analysis models or non-linear neural networks may be applied. A plurality of process data D1 is used as an input variable, and predetermined process data D1 is used as an output variable. Any prediction model may be used as long as it is a prediction model. A prediction model constructed as a multivariate analysis model is a prediction model capable of calculating a quantitative prediction value. In the prediction model construction control, since one prediction model is constructed for one data set, a plurality of prediction models are constructed for a plurality of data sets extracted in the data extraction control.

  Furthermore, the calculation part 25 diagnoses the presence or absence of abnormality of the process data D1 used as an input variable for each of a plurality of constructed prediction models by executing data diagnosis control. For example, when the upper limit value and the lower limit value are set in advance and the process data D1 is within the set upper limit value and lower limit value, it is determined that there is no abnormality in the process data D1. If it is not within the set upper limit value and lower limit value, it is determined that there is an abnormality in the process data D1. The diagnosis of abnormality in the process data D1 is not limited to the above configuration.

  Moreover, the calculating part 25 calculates a predicted value using the constructed | assembled several prediction model by performing predicted value calculation control, respectively.

  Further, the calculation unit 25 executes the predicted value selection control, and selects a predicted value with a higher priority from the calculated predicted values based on the model priority. Here, the model priority is a priority for selecting a predicted value predicted by the prediction model, and is stored in the database 22. The model priority may be set in advance by the user, or may be set based on an error in prediction accuracy of the prediction model. Note that the model priority is not limited to the above configuration, and may be set based on AIC (Akaike information criterion) or BIC (Base information criterion).

  Next, a prediction method for predicting a predetermined physical quantity related to the gas turbine 10 by the prediction system 20 will be described with reference to FIGS. 2 and 3. First, with reference to FIG. 2, the control operation | movement which builds a prediction model in the prediction system 20 is demonstrated. When various process data D1 acquired in the gas turbine 10 is input to the database 22 (step S11), the input process data D1 is stored in the database 22 (step S12). When constructing a prediction model, the calculation unit 25 of the prediction device 21 first extracts data sets from the database 22 by executing data extraction control (step S13: data set acquisition step). Thereafter, the calculation unit 25 executes (predicts model construction) control to construct (create) a plurality of prediction models from the extracted plurality of data sets (step S14: prediction model construction step). And the calculating part 25 complete | finishes the control action which builds a prediction model.

  Next, with reference to FIG. 3, a control operation for diagnosing the presence or absence of abnormality in the process data D1 by the calculation unit 25 of the prediction device 21 and calculating a prediction value based on the abnormality diagnosis result will be described. When calculating a prediction value from a plurality of prediction models, the calculation unit 25 of the prediction device 21 calculates a prediction value from each of the plurality of constructed prediction models by executing prediction value calculation control (step S21: prediction value). Calculation step). Thereafter, the calculation unit 25 executes data diagnosis control to diagnose the presence or absence of abnormality in the process data D1 used as an input variable for each of the plurality of constructed prediction models (step S22: input variable). Abnormal diagnosis process). Subsequently, the calculation unit 25 determines whether or not there is an abnormality in the process data D1 based on the result of executing the data diagnosis control (step S23). When the calculation unit 25 determines that the process data D1 is abnormal (step S23: Yes), the calculation unit 25 sets that the prediction model including the process data D1 diagnosed as abnormal as an input variable is not used (step S24: model not used). Setting process). And the calculating part 25 selects a predicted value with a high priority from the calculated several predicted values based on a model priority by performing predicted value selection control. At this time, the calculation unit 25 omits the prediction value predicted by the prediction model that has been set not to be used, and then selects a prediction value with a high priority from the calculated plurality of prediction values (step S25). : Predicted value selection process). On the other hand, if the calculating part 25 determines in step S23 that there is no abnormality in the process data D1 (step S23: No), step S25 is executed without executing step S24. And the calculating part 25 derives | leads-out the predicted value selected in step S25 as a final predicted value, and complete | finishes the control operation which calculates a predicted value.

  As described above, according to the first embodiment, a prediction model including the process data D1 diagnosed as having an abnormality as an input variable can be set to an unused state. For this reason, when selecting a predicted value calculated from a plurality of prediction models, it is possible to select a highly accurate predicted value calculated by a prediction model using appropriate process data D1 having no abnormality as an input variable.

  Further, according to the first embodiment, in the monitoring system 30, the monitoring device 31 can monitor the gas turbine 10 using a highly accurate predicted value calculated by the prediction system 20. For this reason, the monitoring device 31 can monitor the behavior of the gas turbine 10 while predicting it.

  Further, according to the first embodiment, in the driving support system 40, the driving support device 41 can support the operation of the gas turbine 10 using the accurate predicted value calculated by the prediction system 20. For this reason, the driving assistance device 41 can assist the gas turbine 10 to perform an efficient driving based on the predicted value.

  Next, the prediction system 20 and the prediction method according to the second embodiment will be described with reference to FIG. FIG. 4 is a flowchart regarding construction of a prediction model and abnormality diagnosis according to the second embodiment. In the second embodiment, parts that are different from the first embodiment will be described in order to avoid redundant descriptions, and the same reference numerals are given to the parts that have the same configuration as the first embodiment. The prediction system 20 according to the first embodiment diagnoses whether or not the process data D1 is abnormal after the prediction model is constructed. However, the prediction system 20 according to the second embodiment predicts after the diagnosis of the presence or absence of the abnormality of the process data D1. A model is being built. Hereinafter, the prediction system 20 according to the second embodiment will be described.

  In the prediction system 20 of the second embodiment, the process data D1 is accumulated in the database 22 at a predetermined sampling period, as in the first embodiment. The database 22 stores alternative information. The substitute information is information for using, as substitute process data, highly correlated process data D1 that can be substituted for the predetermined process data D1, and is information that associates the predetermined process data D1 with the substitute process data.

  Next, prediction value calculation control by the calculation unit 25 of the prediction device 21 will be described. The calculation unit 25 of the prediction device 21 performs data extraction control, data diagnosis control, prediction model construction control, prediction value calculation control, and prediction value selection control when calculating the prediction value. The data extraction control, the prediction model construction control, the prediction value calculation control, and the prediction value selection control are the same as those in the first embodiment, and thus the description thereof is omitted.

  The calculation unit 25 diagnoses whether or not there is an abnormality in the included process data D1 for each of a plurality of data sets acquired in the data extraction control by executing data diagnosis control. Note that the diagnosis of abnormality in the process data D1 is the same as that in the first embodiment, and a description thereof will be omitted.

  Next, with reference to FIG. 4, a control operation for constructing a prediction model in the prediction system 20 will be described. Note that steps S31 to S33 are the same as steps S11 to S13 of the first embodiment, and a description thereof will be omitted. After step S33, the calculation unit 25 performs data diagnosis control to diagnose whether there is an abnormality in the included process data D1 for each of the plurality of data sets acquired in the data extraction control (step S33). S34: Input variable abnormality diagnosis step). Subsequently, the computing unit 25 determines whether or not the process data D1 is abnormal based on the result of executing the data diagnosis control (step S35). If the calculation unit 25 determines that the process data D1 is abnormal (step S35: Yes), it sets that the process data D1 diagnosed as abnormal is not used (step S36). Thereafter, the calculation unit 25 sets, based on the substitution information, substitution process data that can be substituted for the non-use set process data D1 to be used instead of the non-use set process data D1 (step S37). And the calculating part 25 builds (creates) a some prediction model based on the extracted several data sets by performing prediction model construction control (step S38: prediction model construction process). At this time, the arithmetic unit 25 omits the process data D1 set not to be used, and the process data D1 included in the data set and the substitute process data to be used instead of the process data D1 set not to be used as input variables. A prediction model having the predetermined process data D1 as an output variable is constructed. If the calculation unit 25 determines in step S35 that the process data D1 is normal (step S35: No), the calculation unit 25 proceeds to step S38 without executing steps S36 and S37.

  After that, the calculation unit 25 calculates the predicted value from each of the constructed prediction models by executing the predicted value calculation control, similarly to the first embodiment (predicted value calculating step). And the calculating part 25 selects a predicted value with a high priority from the calculated several predicted values based on a model priority by performing predicted value selection control. Thereafter, the calculation unit 25 derives the selected predicted value as a final predicted value, and ends the control operation for calculating the predicted value.

  As described above, according to the second embodiment, it is possible to construct a prediction model by omitting the process data D1 diagnosed as having an abnormality and using the remaining process data D1 as input variables. For this reason, it is possible to calculate a highly accurate predicted value using a prediction model that uses appropriate process data D1 having no abnormality as an input variable.

  Further, according to the second embodiment, alternative process data can be used instead of the process data D1 diagnosed as having an abnormality. Then, the process data D1 diagnosed as having an abnormality can be omitted, and a prediction model can be constructed using the alternative process data and the remaining process data D1 as input variables. For this reason, it is possible to calculate a highly accurate predicted value using a prediction model that uses appropriate process data D1 having no abnormality as an input variable.

  In the second embodiment, the process model D1 that has been set not to be used is replaced with alternative process data, and a prediction model is constructed. However, the present invention is not limited to this configuration, and the process data D1 that has not been set to be used is omitted. A prediction model may be constructed using the remaining process data D1.

  In the second embodiment, a plurality of types of data sets are acquired in the data extraction control. However, the present invention is not limited to this configuration, and one data set may be acquired. In this case, since there is one prediction model constructed from one data set and there is one prediction value predicted from one prediction model, selection of prediction values based on model priority should be executed. It is not necessary.

  In the second embodiment, in the data extraction control, the process data D1 included in the data set may be automatically selected and acquired by the calculation unit 25. For example, the correlation is performed with respect to the physical quantity calculated as the predicted value. The process data D1 having a high number may be selected, the process data D1 may be selected at random, or the process data D1 previously selected by the operator may be selected. In addition, the process data D1 may be selected by using a statistical method such as a variable reduction method or a stepwise method, or the process data D1 may be selected in a round robin manner.

  Next, the prediction system 20 and the prediction method according to the third embodiment will be described with reference to FIG. FIG. 5 is a flowchart regarding abnormality diagnosis of the third embodiment. In Example 3, parts different from Example 1 and Example 2 will be described in order to avoid redundant description, and parts having the same configurations as those of Example 1 and Example 2 will be denoted by the same reference numerals. The prediction system 20 according to the third embodiment performs a diagnosis of whether there is an abnormality in the process data D1 by a method different from the first and second embodiments. Hereinafter, the prediction system 20 according to the third embodiment will be described.

  The calculation unit 25 performs data diagnosis control, thereby diagnosing whether there is an abnormality in the process data D1 used as an input variable for each of the plurality of constructed prediction models by the control operation shown in FIG. As illustrated in FIG. 5, the arithmetic unit 25 selects and extracts a plurality of process data D1 having a different combination from a plurality of process data D1 included in a predetermined data set. Then, the calculation unit 25 constructs a plurality of prediction models having the plurality of process data D1 in different combinations as input variables and the predetermined process data D1 as an output variable according to the number of combinations (step S41).

  Thereafter, the calculation unit 25 calculates an error in prediction accuracy of each of the constructed prediction models (step S42). Here, the error in the prediction accuracy of the prediction model can be acquired by comparing the predicted value related to the predetermined physical quantity predicted by the prediction model with the process data D1 as the measured true value related to the predetermined physical quantity. It has become.

  Subsequently, when calculating the errors of the plurality of prediction models, the calculation unit 25 determines whether there is a prediction model that has an error greater than or equal to a preset setting error among the calculated plurality of errors (step S43). . When there is a prediction model that is greater than or equal to the setting error (step S43: Yes), the computing unit 25 selects process data D1 that has an abnormality from among the plurality of process data D1 that are included in the prediction model that is greater than or equal to the setting error. Specify (step S44). Specifically, there are process data D1a, D1b, and D1c as the plurality of process data D1 included in the prediction model that causes an error equal to or greater than the setting error. On the other hand, there are process data D1a, D1b, and D1d as the plurality of process data D1 included in the prediction model that has an error smaller than the setting error. At this time, the calculation unit 25 compares the process data D1 included in the plurality of prediction models, and the process data D1a, D1b, and D1c included in the prediction model that has an error equal to or larger than the setting error is less than the setting error. The process data D1c not included in the prediction model that causes a small error is specified.

  And the calculating part 25 will determine with the process data D1c being abnormal, if the process data D1c which is abnormal is specified (step S45). On the other hand, in Step S43, when there is no prediction model which becomes more than a setting error (Step S43: No), operation part 25 determines that there is no abnormality in process data D1 (Step S46). And the calculating part 25 complete | finishes the control action regarding data diagnosis control.

  As described above, according to the third embodiment, the calculation unit 25 can identify the process data D1 having an abnormality from the process data D1 included in the prediction model having a large error. The presence or absence of abnormality can be diagnosed with high accuracy.

  In addition, in Example 1- Example 3, although the prediction system 20 estimated the predetermined physical quantity regarding the gas turbine 10, it is not limited to the gas turbine 10, Other facilities may be sufficient.

DESCRIPTION OF SYMBOLS 1 Gas turbine equipment 10 Gas turbine 11 Pressure sensor 12 Temperature sensor 13 Valve 20 Prediction system 21 Prediction device 22 Database 25 Operation part 30 Monitoring system 31 Monitoring device 40 Driving support system 41 Driving support device D1 Process data

Claims (9)

A storage unit for storing various process data that can be acquired in the facility;
A calculation unit that calculates, as a predicted value, a predetermined physical quantity predicted for the facility based on the process data stored in the storage unit, and
The computing unit is
Among the various process data stored in the storage unit, a plurality of process data selected in advance are obtained as a set of data, and a plurality of sets are acquired,
Using the process data included in the acquired data set of each set as an input variable, each constructs a prediction model for calculating the prediction value,
Each of the predicted values is calculated from a plurality of the predicted models constructed,
For each of the plurality of prediction models constructed, diagnose the presence or absence of abnormality of the process data used as the input variable,
Set not to use the prediction model including the process data diagnosed as abnormal as the input variable,
A prediction system, wherein the prediction value predicted by the prediction model is selected according to priorities of a plurality of prediction models set in advance. A storage unit for storing various process data that can be acquired in the facility;
A calculation unit that calculates, as a predicted value, a predetermined physical quantity predicted for the facility based on the process data stored in the storage unit, and
The computing unit is
Among the various process data stored in the storage unit, a plurality of the process data selected in advance is acquired as one data set,
Diagnosing the presence or absence of abnormality of the process data included in the acquired data set,
Constructing a prediction model that calculates the predicted value, with the remaining process data included in the data set as input variables, omitting the process data diagnosed as having an abnormality,
A prediction system, wherein the prediction value is calculated from the constructed prediction model. The storage unit stores alternative information that associates the predetermined process data with alternative process data that is an alternative to the predetermined process data,
The computing unit is
When there is the process data diagnosed as having an abnormality, the process data diagnosed as having an abnormality is substituted with the substitute process data based on the substitute information, and the process data including the substituted process data is substituted as an input variable. The prediction system according to claim 2, wherein the prediction model is constructed. The computing unit is
When diagnosing the presence / absence of abnormality in the process data, the prediction model for calculating the prediction value using a plurality of process data in different combinations among the plurality of process data included in the data set as input variables. Build several according to the number of combinations,
Calculating an error in the prediction accuracy of each of the plurality of prediction models,
The process data having an abnormality is specified from among the plurality of process data included in the prediction model that is the error that is equal to or larger than a preset setting error among the plurality of calculated errors. Item 4. The prediction system according to any one of Items 1 to 3. The prediction system according to any one of claims 1 to 4,
A monitoring system comprising: a monitoring device that monitors the facility using the predicted value calculated by the prediction system. The prediction system according to any one of claims 1 to 4,
A driving support system comprising: a driving support device that supports driving of the facility using the predicted value calculated by the prediction system. At least one of the prediction system according to any one of claims 1 to 4, the monitoring system according to claim 5, and the driving support system according to claim 6,
A gas turbine facility comprising: a gas turbine as the facility. A prediction method for deriving a predetermined physical quantity related to the facility as a predicted value based on various process data that can be acquired in the facility,
A data set acquisition step of acquiring a plurality of sets of process data selected in advance as a set of data among various types of process data; and
A prediction model construction step of constructing a prediction model for calculating the prediction value using the process data included in the acquired data set of each set as an input variable,
A predicted value calculation step of calculating the predicted value from the plurality of the predicted models constructed;
An input variable abnormality diagnosis step for diagnosing the presence or absence of abnormality of the process data used as the input variable for each of the plurality of constructed prediction models,
A model non-use setting step for setting that the prediction model including the process data diagnosed as having an abnormality as the input variable is not used,
A prediction value selecting step of selecting the prediction value predicted by the prediction model according to the priorities of the plurality of prediction models set in advance. A prediction method for deriving a predetermined physical quantity related to the facility as a predicted value based on various process data that can be acquired in the facility,
A data set acquisition step of acquiring a plurality of process data selected in advance as a set of data among various types of process data;
An input variable abnormality diagnosis step of diagnosing the presence or absence of abnormality of the process data included in the acquired data set;
Prediction model construction step of constructing a prediction model for calculating the prediction value using the remaining process data included in the data set as an input variable, omitting the process data diagnosed as having an abnormality,
A prediction value calculating step of calculating the prediction value from the constructed prediction model.

Images