Information processing device, information processing method, and information processing program

Abstract

In the prediction of parameters, both reliability and accuracy are taken into consideration. An information processing device (1) comprises: a first prediction unit (12) which calculates a first prediction value using a mathematical model; a second prediction unit (13) which calculates a second prediction value using a learned model; and a prediction value determination unit (15) which determines a prediction value of a parameter based on the first prediction value and the second prediction value.

Description

Technical Field

[0001] The present invention relates to an information processing device and the like that predicts a parameter value related to a process in a factory. Background Art

[0002] Various technologies for appropriately controlling various factories have been studied and developed. For example, Patent Document 1 below discloses a technology for predictive operation control of various process data in a learned model composed of a neural network in operation control of a treatment plant equipment that performs various treatments, such as a waste treatment plant and a water treatment plant.

[0003] Prior art literature

[0004] Patent Literature

[0005] Patent Document 1: Japanese Patent Publication No. 2005-242524 Summary of the invention

[0006] 1. Technical issues to be resolved

[0007] According to the technology of the above-mentioned patent document 1, a learned model composed of a neural network can be used to make a high-precision prediction of a process exhibiting a nonlinear state. However, the prediction accuracy of the learned model depends on what kind of learning has been performed. Therefore, when input data that belongs to a range that has not been learned at all or has been insufficiently learned is input to the learned model, the prediction accuracy will be reduced.

[0008] In particular, the predicted value based on the learned model may deviate from reality. Furthermore, if control is performed based on such a value that deviates from reality, it may also cause serious problems in plant operation. This leads to the problem of reliability of predictions using only the learned model.

[0009] One aspect of the present invention has been accomplished in view of the above-mentioned problems, and an object of the present invention is to realize an information processing device or the like that can achieve both reliability and accuracy in parameter prediction of a factory.

[0010] (II) Technical solution

[0011] In order to solve the above-mentioned problem, an information processing device of one embodiment of the present invention comprises: a first prediction unit, which uses a mathematical model that models a specified process of a factory to calculate a predicted value of a specified parameter related to the process, i.e., a first predicted value; a second prediction unit, which uses a learned model that models the above-mentioned process through machine learning to calculate a predicted value of the above-mentioned parameter, i.e., a second predicted value; and a predicted value determination unit, which determines the predicted value of the above-mentioned parameter based on at least any one of the above-mentioned first predicted value and the above-mentioned second predicted value.

[0012] In order to solve the above-mentioned problem, an information processing method of one embodiment of the present invention is an information processing method of an information processing device, comprising: a first prediction step, using a mathematical model that models a specified process of a factory to calculate a predicted value of a specified parameter related to the process, i.e., a first predicted value; a second prediction step, using a learned model that models the above-mentioned process through machine learning to calculate a predicted value of the above-mentioned parameter, i.e., a second predicted value; and a predicted value determination step, determining the predicted value of the above-mentioned parameter based on at least any one of the above-mentioned first predicted value and the above-mentioned second predicted value.

[0013] (III) Beneficial effects

[0014] According to one embodiment of the present invention, it is possible to achieve both reliability and accuracy in parameter prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a block diagram showing an example of the configuration of main parts of an information processing device according to one embodiment of the present invention.

[0016] Figure 2 This is a diagram showing an example of a superheater that superheats steam in three stages and a prediction model corresponding to the superheater.

[0017] Figure 3 This is a diagram for explaining determination of a prediction result by a prediction value determination unit.

[0018] Figure 4 It is a diagram showing the prediction result of the steam outlet temperature by the above-mentioned information processing device and the actual measured value of the steam outlet temperature.

[0019] Figure 5 This is a flowchart showing an example of processing executed by the above-mentioned information processing device.

[0020] Figure 6 This is a diagram for explaining each unit related to prediction in the information processing device according to the second embodiment of the present invention.

[0021] Figure 7 This is a diagram showing an example of time-series changes in the steam outlet temperature and the pressure in the high-pressure steam storage unit during the normal operation period and the sootblowing operation period.

[0022] Figure 8 It is a diagram for explaining the control details of the FB control unit and the MPC unit.

[0023] Fig. 9 This is a block diagram showing an example of the configuration of main parts of an information processing device according to a fourth embodiment of the present invention. DETAILED DESCRIPTION

[0024] [Implementation method 1]

[0025] <Structure of information processing device>

[0026] based on Figures 1 to 5 An information processing device according to this embodiment will be described. Figure 1 1 is a block diagram showing an example of the configuration of a main part of the information processing device 1 according to the present embodiment. The information processing device 1 is a device that calculates a predicted value of a predetermined parameter related to a predetermined process in a factory.

[0027] In the present embodiment, the factory is an example of a waste incineration facility that incinerates waste and generates electricity using the heat generated during the incineration. In the present embodiment, the predetermined process is an example of a process that superheats steam generated in a boiler to a predetermined temperature using a superheater (hereinafter referred to as a steam superheating process). In the present embodiment, the predetermined parameter is an example of a steam temperature near the outlet of the superheater (hereinafter referred to as a steam outlet temperature).

[0028] like Figure 1 As shown, the information processing device 1 includes: a control unit 10 that performs overall control of each unit of the information processing device 1, and a storage unit 20 that stores various data used by the information processing device 1. In addition, the information processing device 1 includes: an input unit 30 that receives input operations to the information processing device 1; and a communication unit 40 that is used for the information processing device 1 to communicate with other devices.

[0029] In addition, the control unit 10 includes an input data generating unit 11, a first prediction unit 12, a second prediction unit 13, an operation mode determining unit 14, a predicted value determining unit 15, and a plant control unit 16. Furthermore, the plant control unit 16 includes a FB control unit 161 and an MPC unit 162. In Embodiment 2, the operation mode determining unit 14 is described, and in Embodiment 3, the FB control unit 161 and the MPC unit 162 are described. That is, the operation mode determining unit 14, the FB control unit 161, and the MPC unit 162 are preferably provided components, but are not essential components.

[0030] The input data generating unit 11 generates input data used for prediction by the first prediction unit 12 and input data used for prediction by the second prediction unit 13. In addition, it is also possible to configure that different input data generating units respectively generate input data used for prediction by the first prediction unit 12 and input data used for prediction by the second prediction unit 13. The input data is basically measured data measured by sensors installed in the factory. The measured data can be input via the input unit 30 or received via the communication unit 40.

[0031] However, the input data generating unit 11 may use a mathematical model instead of the measured data to calculate at least a portion of the input data for at least one of the mathematical model used by the first prediction unit 12 and the learned model used by the second prediction unit 13. In this way, data that is difficult to measure can be included in the input data, further improving the accuracy of the predicted value.

[0032] For example, if the flow rate of steam flowing in the superheater is known, the steam temperature near the outlet of the superheater can be predicted more accurately than when the flow rate is unknown, but there are cases where it is difficult to actually measure the flow rate. In this case, a mathematical model that uses other parameters related to the steam flow rate (parameters for which actual measurement data can be obtained) to calculate the steam flow rate can be generated. Thus, the input data generation unit 11 can calculate the steam flow rate using the mathematical model and include the calculated flow rate in the input data.

[0033] The first prediction unit 12 uses a mathematical model to calculate a first prediction value, which is a prediction value of a predetermined parameter related to a predetermined process of the plant. The mathematical model is generated by modeling the above process. Specifically, the first prediction unit 12 calculates a prediction value of the steam outlet temperature in the steam superheating process. Since the first prediction unit 12 uses a mathematical model, the possibility of calculating a prediction value that deviates from reality is low, and a prediction value with high reliability can be stably calculated.

[0034] The second prediction unit 13 calculates a second prediction value, which is a prediction value of the parameter, using a learned model obtained by modeling the process through machine learning. Since the second prediction unit 13 uses the learned model, it is possible to calculate a highly accurate prediction value for a process exhibiting a nonlinear state.

[0035] Specifically, the above-mentioned learned model is generated by machine learning of a neural network (hereinafter referred to as NN) in order to be able to calculate the predicted value of the steam outlet temperature in the steam superheating process. Such a learned model can be generated by machine learning, and the machine learning uses the following teacher data, which is data that corresponds input data related to the steam outlet temperature, such as the steam temperature near the steam inlet of the superheater, and the correct data representing the steam outlet temperature at this time. For the data that can be measured, such as the steam outlet temperature and the steam inlet temperature, which are included in the teacher data, measured data measured in the waste incineration facility can be used. In addition, the data generated by the input data generation unit 11 and calculated by a prescribed mathematical model can also be included in the input data.

[0036] In addition, the algorithm used in the learned model is not particularly limited, and for example, a convolutional neural network, a recurrent neural network, or a combination of the two can be applied. In addition, other algorithms other than NN that can perform machine learning can also be used, such as LSTM (Long Short-Term Memory: Long Short-Term Memory Neural Network).

[0037] The predicted value determination unit 15 determines the predicted value of the parameter based on at least one of the first predicted value calculated by the first prediction unit 12 and the second predicted value calculated by the second prediction unit 13. Figure 3 A method for determining the predicted value of a parameter is described.

[0038] The plant control unit 16 controls the plant based on the predicted value determined by the predicted value determination unit 15. The plant control unit 16 controls the plant based on the predicted value with high reliability and high accuracy determined by the predicted value determination unit 15, so that appropriate control can be performed. In addition, the plant control unit 16 performs control based on the predicted value, so it is also possible to perform control (for example, feedforward control) at an earlier timing than feedback control based on the actual measurement value. In addition, the plant control unit 16 can also perform feedback control in addition to the above-mentioned control based on the predicted value, and the details will be described in Embodiment 3.

[0039] <Superheater Configuration Example and Prediction Model>

[0040] based on Figure 2 A configuration example of a superheater and a prediction model corresponding to the superheater will be described. Figure 2 : is a diagram showing an example of a superheater B that superheats steam in three stages and a prediction model corresponding to the superheater B. Figure 2 In the figure, a boiler A for generating steam using heat generated by burning garbage is also shown.

[0041] The steam generated by the boiler A is introduced into the superheater B and superheated to a predetermined temperature. The superheated steam discharged from the superheater B is temporarily stored in a high-pressure steam storage unit (accumulator) not shown, and then supplied to a steam turbine generator not shown. In order for the steam turbine generator to generate electricity stably, the temperature of the steam supplied to the steam turbine generator needs to be maintained at a predetermined temperature.

[0042] The illustrated superheater B includes a primary superheater 11B, a primary regulating valve 12B, a secondary superheater 21B, a secondary regulating valve 22B, and a tertiary superheater 31B. The steam introduced into the interior of the superheater B is superheated in the primary superheater 11B to the tertiary superheater 31B in sequence and discharged to the outside of the superheater B. The primary regulating valve 12B is a valve for regulating the supply amount of a coolant (e.g., water) used to cool the steam superheated by the primary superheater 11B. Similarly, the secondary regulating valve 22B is a valve for regulating the supply amount of a coolant used to cool the steam superheated by the secondary superheater 21B. The plant control unit 16 appropriately adjusts the opening of these valves based on the predicted value of the steam outlet temperature determined by the predicted value determination unit 15. As a result, the steam discharged to the outside of the superheater B can be maintained at a specified temperature.

[0043] like Figure 2 In the case of a steam superheating process in which steam is superheated in multiple stages, it is preferable to combine prediction models corresponding to each stage to predict the steam outlet temperature from the viewpoint of improving prediction accuracy. Figure 2 As shown, the steam outlet temperature can be predicted by combining process models A1 to A3.

[0044] The process model A1 is a model that models the steam superheating process using the primary superheating section 11B. More specifically, the process model A1 is a model for predicting the steam temperature near the outlet of the primary superheating section 11B based on input data related to the superheating process such as the steam temperature near the inlet of the superheater B. In addition, the steam superheating process using the primary superheating section 11B is referred to as the primary superheating process hereinafter. Similarly, the superheating processes using the secondary superheating section 21B and the tertiary superheating section 31B are referred to as the secondary superheating process and the tertiary superheating process, respectively.

[0045] The process model A2 is a model that models the secondary superheating process. More specifically, the process model A2 is a model for predicting the steam temperature near the outlet of the secondary superheating section 21B based on input data related to the secondary superheating process, such as the steam temperature near the outlet of the primary superheating section 11B and the opening of the primary control valve 12B.

[0046] The process model A3 is a model that models the tertiary superheating process. More specifically, the process model A3 is a model for predicting the steam temperature near the outlet of the superheater B based on input data related to the tertiary superheating process, such as the steam temperature near the outlet of the secondary superheating section 21B and the opening of the secondary control valve 22B.

[0047] Here, the steam temperature near the outlet of the primary superheater 11B predicted by the process model A1 is used as input data of the process model A2. In addition, the steam temperature near the outlet of the secondary superheater 21B predicted by the process model A2 is used as input data of the process model A3. In this way, by using the prediction results of the process model A1 for the prediction of the process model A2, and using the prediction results of the process model A2 for the prediction of the process model A3, the entire steam superheating process can be predicted (prediction of the steam temperature near the outlet of the superheater B).

[0048] As described above, there is a case where the process to be predicted for parameters (steam superheating process using superheater B) includes a plurality of sub-processes (single to triple superheating processes). In such a case, the first prediction unit 12 and the second prediction unit 13 may perform prediction by combining models corresponding to the sub-processes (process models A1 to A3).

[0049] In this case, the first prediction unit 12 performs prediction using the process models A1 to A3 as mathematical models. This makes it possible to perform a highly accurate prediction compared to a case where a steam superheating process is represented by one mathematical model.

[0050] Similarly, the second prediction unit 13 performs prediction using the process models A1 to A3 which are learned models generated by machine learning. This makes it possible to perform a more accurate prediction than when a steam superheating process is represented by one learned model.

[0051] <Determination of parameter prediction values>

[0052] Since various factors are related to various processes in a factory, including the steam superheating process, it is preferable that various factors are also included in the teacher data used for machine learning. However, there are cases where the measured data measured in the factory are missing or not recorded in a certain numerical range. Therefore, when a learned model reflecting various factors is generated by using a plurality of measured data as teacher data, the accuracy of the output data relative to the input data in a certain numerical range will decrease.

[0053] Therefore, if the measured data used by the second prediction unit 13 for calculating the second predicted value is within the range of the teacher data used in the machine learning of the learned model, the second predicted value is determined as the predicted value of the parameter by the prediction value determination unit 15. On the other hand, if the measured data used by the second prediction unit 13 for calculating the second predicted value is outside the range of the teacher data used in the machine learning of the learned model, the first predicted value is determined as the predicted value of the parameter by the prediction value determination unit 15.

[0054] Thus, when the measured data input to the learned model is within the range of the teacher data for machine learning and the calculated second predicted value can be expected to have high reliability and high accuracy, the second predicted value can be determined as the predicted value of the parameter. Furthermore, when the measured data input to the learned model is outside the range of the teacher data for machine learning and the calculated second predicted value is difficult to be expected to have high reliability and high accuracy, the first predicted value can be determined as the predicted value of the parameter. Therefore, according to the above-mentioned structure, both reliability and accuracy can be taken into account in the prediction of the parameter.

[0055] based on Figure 3 The method of determining the predicted value of the parameter is described. Figure 3 This is a diagram for explaining how the prediction value determination unit 15 determines the prediction result. Figure 3 (a) shows an example of a criterion for determining whether to use NN (whether to use the second prediction value based on the learned model) based on a plurality of factors included in the teacher data. Factors 1 and 2 shown in the figure are factors included in the input data for the learned model. Examples of these factors include, for example, the steam temperature near the inlet of the superheater B, the temperature of the exhaust gas generated in the incinerator used as the heat source of the superheater B, and the like.

[0056] like Figure 3 As shown in (a), the prediction value determination unit 15 preferably determines whether to use NN based on the combination of the value of factor 1 and the value of factor 2. More specifically, if the value of factor 1 in the input data input to the learned model is included in the range that has been learned and the value of factor 2 in the input data is also included in the range that has been learned, the prediction value determination unit 15 determines to use NN. In this case, the prediction value determination unit 15 determines the second prediction value as the prediction value of the steam outlet temperature. On the other hand, if the value of factor 1 in the input data input to the learned model is not included in the range that has been learned and the value of factor 2 in the input data is also not included in the range that has been learned, the prediction value determination unit 15 determines not to use NN. In this case, the prediction value determination unit 15 determines the first prediction value as the prediction value of the steam outlet temperature.

[0057] Furthermore, when more than three factors are included in the teacher data, factors that are relatively affected by the missing measured data can be selected from those factors and used as a criterion for determining whether to use the NN-based predicted value by the predicted value determination unit 15. When selecting factors that are relatively affected, methods such as principal component analysis, clustering, or sparseness modeling can be used.

[0058] In addition, the prediction value determination unit 15 may determine the prediction value based on whether the input data belongs to a numerical range that can be machine-learned using relatively more teacher data. Figure 3 (b) and (c) are explained. Figure 3 (b) shows the corresponding relationship between the steam temperature near the inlet of the superheater B (horizontal axis) and the number of teacher data of the steam temperature (vertical axis). Figure 3 (c) shows the correspondence between the steam temperature near the inlet of the superheater B included in the input data (horizontal axis) and the contribution of the second predicted value in the predicted values ​​determined by the predicted value determination unit 15 (vertical axis).

[0059] When generating teacher data based on the actual measured data of the factory, it is not possible to obtain an equal number of teacher data for all steam temperatures. As shown in the figure, the teacher data is usually biased toward the central value or average value of the steam temperature, and becomes less near the maximum and minimum values ​​of the steam temperature.

[0060] Therefore, if the steam temperature included in the input data is within the range of steam temperatures in which the number of teacher data is relatively large, it is preferred that the predicted value determination unit 15 increases the contribution of the second predicted value. For example, when the contribution when the second predicted value is not used (only the first predicted value is used) is set to 0, and when the contribution when only the second predicted value is used is set to 1, it is preferred that the contribution near the central value or average value of the steam temperature is 1 or a value close to 1.

[0061] On the other hand, if the steam temperature included in the input data is within the steam temperature range for which the amount of teacher data is relatively small, the predicted value determination unit 15 preferably reduces the contribution of the second predicted value. Furthermore, if the steam temperature included in the input data is within the steam temperature range for which teacher data does not exist, the predicted value determination unit 15 preferably sets the contribution of the second predicted value to zero.

[0062] exist Figure 3 In the example (c), the contribution is made to be the largest near the central value or the average value of the steam temperature, and the contribution is gradually reduced as the steam temperature moves away from the central value or the average value, and the contribution is sharply reduced near the end of the steam temperature range. In this way, by setting the contribution corresponding to the input data, it is possible to achieve both reliability and accuracy in the prediction of the parameters.

[0063] In addition, when determining the predicted value of the parameter based on the contribution, the predicted value determination unit 15 determines the predicted value of the parameter by multiplying the second predicted value by the contribution. For example, if the contribution of the second predicted value is 0.7, the predicted value determination unit 15 can determine the sum of the value of the second predicted value multiplied by 0.7 and the value of the first predicted value multiplied by 0.3 as the predicted value of the parameter. This operation can also be said to be an operation of correcting the first predicted value with the second predicted value.

[0064] <Verification of prediction accuracy>

[0065] Figure 4 is a graph showing the prediction result of the steam outlet temperature by the information processing device 1 and the actual value of the steam temperature. As shown in the figure, it can be seen that the prediction result of the steam outlet temperature by the information processing device 1 (solid line) and the actual value of the steam outlet temperature (dashed line) match well, and the prediction accuracy of the information processing device 1 is high. Figure 4 There are two locations where the temperature drops significantly, and the predicted results deviate from the measured values ​​at these locations.

[0066] The temperature reduction at the two locations is caused by the soot blowing operation for removing coal and the like attached to the heat transfer surface of the superheater B. More specifically, during the soot blowing operation, the coal and the like are removed by spraying steam onto the heat transfer surface. Therefore, the temperature of the heat transfer surface drops significantly, thereby also reducing the steam outlet temperature. In this way, since steam spraying is performed during the soot blowing operation, which is not performed during the normal operation, the prediction result based on the process model assuming the normal operation will deviate from the actual measured value. In the second embodiment, a strategy for preventing such a deviation is described.

[0067] <Processing Flow>

[0068] Next, based on Figure 5 The flow of processing executed by the information processing device 1 of this embodiment will be described. Figure 5 This is a flowchart showing an example of processing (information processing method) executed by the information processing device 1 of this embodiment.

[0069] In S1, the input data generator 11 acquires various measured data required to generate input data. The input data generator 11 generates input data used by the first prediction unit 12 and the second prediction unit 13 based on the acquired measured data.

[0070] In S2 (first prediction step), the first prediction unit 12 inputs the input data generated by the input data generation unit 11 to the mathematical model and calculates the predicted value of the steam outlet temperature, i.e., the first predicted value. In addition, in S3 (second prediction step), the second prediction unit 13 inputs the input data generated by the input data generation unit 11 to the learned model and calculates the predicted value of the steam outlet temperature, i.e., the second predicted value. In addition, S2 and S3 may be executed in parallel, or S3 may be executed first.

[0071] In S4 (prediction value determination step), the prediction value determination unit 15 determines the prediction value of the steam outlet temperature based on at least one of the first prediction value calculated in S2 and the second prediction value calculated in S3 .

[0072] In S5, the plant control unit 16 controls the plant based on the predicted value determined by the predicted value determination unit 15. For example, when the predicted value of the steam outlet temperature determined in S4 is higher than the specified temperature, the plant control unit 16 performs control to increase the opening of at least one of the primary control valve 12B and the secondary control valve 22B. On the contrary, when the predicted value of the determined steam outlet temperature is lower than the specified temperature, the plant control unit 16 performs control to reduce the opening of at least one of the primary control valve 12B and the secondary control valve 22B. In this way, it is possible to avoid the steam outlet temperature being higher or lower than the specified temperature, or to minimize the time for maintaining the state.

[0073] As described above, the information processing device 1 comprises: a first prediction unit 12, which uses a mathematical model that models the steam superheating process to calculate a predicted value of the steam outlet temperature, i.e., a first predicted value; a second prediction unit 13, which uses a learned model that models the steam superheating process by machine learning to calculate a predicted value of the steam outlet temperature, i.e., a second predicted value; and a predicted value determination unit 15, which determines the predicted value of the steam outlet temperature based on at least any one of the above-mentioned first predicted value and the above-mentioned second predicted value.

[0074] The prediction based on the learned model has a high prediction accuracy for processes that present nonlinear states. On the other hand, the prediction based on the mathematical model is a linear prediction, and the possibility of calculating a prediction value that deviates from reality is low. Therefore, according to the above structure, a prediction value that takes into account both reliability and accuracy can be calculated.

[0075] [Implementation method 2]

[0076] Another embodiment of the present invention is described below. In addition, for the convenience of description, the same reference numerals are used for components having the same functions as those described in the above-mentioned embodiment 1, and their description is not repeated. The same is true for embodiment 3 and subsequent embodiments.

[0077] In the present embodiment, a case where the operation mode is determined by the operation mode determination unit 14 and a case where the steam outlet temperature is predicted based on the determination result will be described. Figure 6 This is a diagram for explaining each unit related to prediction in the information processing device 1 of this embodiment. The second prediction unit 13 of this embodiment includes a normal operation prediction unit 131 , an SB operation prediction unit 132 , and a second prediction value determination unit 133 .

[0078] The prediction unit 131 in normal operation calculates the predicted value of the steam outlet temperature using a learned model that is machine-learned using teacher data based on actual measured data measured during normal operation (operation in which soot blowing is not performed but steam is superheated by superheater B, hereinafter also referred to as normal operation mode). In addition, the prediction unit 132 in SB operation calculates the predicted value of the steam outlet temperature using a learned model that is machine-learned using teacher data based on actual measured data measured during soot blowing operation (hereinafter also referred to as SB operation mode). The prediction based on the prediction unit 131 in normal operation and the prediction based on the prediction unit 132 in SB operation are both nonlinear predictions using the learned model.

[0079] Furthermore, when the operation mode determination unit 14 determines that the operation mode is in the normal operation mode, the second prediction value determination unit 133 determines the prediction value calculated by the normal operation prediction unit 131 as the second prediction value. On the other hand, when the operation mode determination unit 14 determines that the operation mode is in the SB operation mode, the second prediction value determination unit 133 determines the prediction value calculated by the SB operation prediction unit 132 as the second prediction value. That is, as a whole, the second prediction unit 13 calculates the second prediction value using the learned model corresponding to the operation mode determined by the operation mode determination unit 14.

[0080] Thus, the information processing device 1 of this embodiment includes an operation mode determination unit 14 for determining the operation mode of the plant, and the second prediction unit 13 calculates the second prediction value using the learned model corresponding to the operation mode determined by the operation mode determination unit 14. Thus, by considering the difference in operation modes, a highly accurate prediction value can be calculated. For example, even in Figure 4 Even during the soot blowing operation of the example, a predicted value close to the actual measured value can be calculated.

[0081] In addition, if a mathematical model corresponding to the operation mode is prepared in advance, the first prediction unit 12 can also be configured to calculate the first prediction value using the mathematical model corresponding to the operation mode determined by the operation mode determination unit 14, as described above. The prediction based on the first prediction unit 12 is a linear prediction. After the first prediction value is calculated using the linear prediction and the second prediction value is calculated using the nonlinear prediction, the prediction value determination unit 15 determines the prediction value based on at least one of the first prediction value and the second prediction value, as in the information processing device 1 of the first embodiment.

[0082] <Determination of operation mode>

[0083] based on Figure 7 A case where the operation mode is determined by the operation mode determination unit 14 will be described. Figure 7This is a diagram showing an example of the time series change of the steam outlet temperature and the pressure in the high-pressure steam storage section (hereinafter referred to as pressure) during normal operation and soot blowing operation. In addition, the upper graph of the figure shows the time series change of the pressure, and the lower graph of the figure shows the time series change of the steam outlet temperature.

[0084] exist Figure 7 In the example of , during the soot blowing operation, three steam injections are performed to the superheater B, and the steam outlet temperature and pressure drop sharply at each steam injection. Thus, during the soot blowing operation, a unique time series change is observed in the steam outlet temperature and pressure. Therefore, the operation mode determination unit 14 can determine whether the operation mode is the normal operation mode or the SB operation mode based on the time series change in the steam outlet temperature and pressure.

[0085] For example, the operation mode determination unit 14 may determine that the pressure is less than a threshold value (at Figure 7 In the example, for example, when the pressure is 3.4 [MPa], it is determined that the operation mode is switched from the normal operation mode to the SB operation mode. Moreover, the operation mode determination unit 14 can determine that the operation mode is switched to the normal operation mode when the pressure is above the threshold and the duration of the SB operation mode is above the specified lower limit time (the minimum time required in a series of soot blowing operations).

[0086] The method of determining the operation mode is not limited to the above example. For example, the operation mode determination unit 14 may cluster the time series measured data of the steam outlet temperature and pressure into measured data of the SB operation mode and measured data of the normal operation mode using a variational Bayesian method, thereby determining the current operation mode.

[0087] [Implementation method 3]

[0088] In the present embodiment, details of control performed by the plant control unit 16 , particularly details of control performed by the FB control unit 161 and the MPC unit 162 included in the plant control unit 16 will be described. Figure 8 It is a diagram for explaining the details of the control performed by the FB control unit 161 and the MPC unit 162 .

[0089] The FB control unit 161 performs feedback control (hereinafter referred to as FB control) based on various measured data actually measured in the factory. Figure 8 The control performed by the FB control unit 161 is a feedback control executed every time new actual measurement data is obtained.

[0090] Specifically, when the measured data of the steam outlet temperature is obtained, the FB control unit 161 calculates the difference between the measured data and the target value of the steam outlet temperature (primary operation). Moreover, the FB control unit 161 determines the opening of the secondary control valve 22B corresponding to the difference calculated in the primary operation, and controls the secondary control valve 22B in a manner to achieve the determined opening. This control can be called, for example, PID (Proportional-Integral-Differential) control. In addition, the FB control unit 161 determines the opening of the primary control valve 12B in the same manner as described above, and controls the primary control valve 12B in a manner to achieve the determined opening.

[0091] Here, based on Figure 2 As described, the steam superheating process performed by the superheater B includes the primary to tertiary superheating processes. Furthermore, as a process related to the primary superheating process, it is a process of regulating the steam temperature via the primary regulating valve 12B, and as a process related to the secondary superheating process, it is a process of regulating the steam temperature via the secondary regulating valve 22B. In the following, the primary superheating process and the process of regulating the steam temperature via the primary regulating valve 12B are collectively referred to as the first sub-process, and the secondary superheating process and the process of regulating the steam temperature via the secondary regulating valve 22B are collectively referred to as the second sub-process.

[0092] The MPC unit 162 performs state predictive control (hereinafter referred to as MPC: Model Predictive Control) based on the predicted value determined by the predicted value determination unit 15. In this MPC, the MPC unit 162 performs control for the first sub-process so that the predicted value determined by the predicted value determination unit 15 approaches a predetermined target value. In addition, the MPC unit 162 performs control for the second sub-process so that the predicted value determined by the predicted value determination unit 15 approaches a predetermined target value. Thus, the control amount for each process is appropriately allocated, so that stable process control can be achieved.

[0093] exist Figure 8 In FIG. 1 , the control executed by the MPC unit 162 of the present embodiment is indicated by a single-dot chain line. The MPC unit 162 determines the opening degrees of the primary control valve 12B and the secondary control valve 22B based on the predicted value of the steam outlet temperature determined by the predicted value determination unit 15 .

[0094] When determining the opening of each control valve, the MPC unit 162 preferably determines the opening of each control valve in a manner that avoids a large deviation in the temperature adjustment range of each control valve. For example, the MPC unit 162 may distribute the difference between the predicted value of the steam outlet temperature and the target value of the steam outlet temperature, that is, the temperature range to be adjusted, equally to the primary control valve 12B and the secondary control valve 22B.

[0095] For example, the difference between the predicted value and the target value of the steam outlet temperature is set to -10°C (the predicted value is 10°C lower than the target value). In this case, the MPC unit 162 reduces the opening of the primary control valve 12B so that the steam temperature near the outlet of the secondary superheater 21B rises by 5°C, and reduces the opening of the secondary control valve 22B so that the steam temperature near the outlet of the superheater B rises by 5°C.

[0096] In addition, the temperature of the exhaust gas used as the heat source of the superheater B is a factor that has a great influence on the superheating of the superheater B. This temperature varies according to the combustion state in the incinerator and is not a constant value in many cases. Therefore, the MPC unit 162 can determine the superheating characteristics of each superheating unit based on the actual measurement data of the current exhaust gas temperature (for example, the temperature near the outlet of the chimney that discharges the exhaust gas to the outside of the incinerator) using a predetermined prediction model, and determine the control value based on the determination result.

[0097] In addition, the method for determining the opening degree based on the predicted value is not particularly limited, and the opening degree may be determined by methods such as Kalman filtering, NN, linear multiple regression, or logistic regression analysis. For example, when the Kalman filter is used, the opening degree of the primary control valve 12B that can stabilize the steam outlet temperature can be predicted based on process changes such as the temperature of the exhaust gas, and feedforward control can be performed.

[0098] In addition, as described above, the FB control unit 161 also determines the openings of the primary control valve 12B and the secondary control valve 22B. Therefore, the MPC unit 162 also considers the openings determined by the FB control unit 161 and finally determines what openings are to be set for each control valve. For example, the MPC unit 162 may take the sum of the openings determined by itself and the openings determined by the FB control unit 161 as the final openings.

[0099] [Implementation method 4]

[0100] based on Fig. 9 Another embodiment of the present invention will be described. Fig. 9 This is a block diagram showing an example of the configuration of the main parts of the information processing device 1 of this embodiment. The information processing device 1 of this embodiment is different from the information processing device 1 of Embodiment 1 (see Figure 1 ) is that the first prediction unit 12, the second prediction unit 13, and the prediction value determination unit 15 are respectively changed into the first control value calculation unit 21, the second control value calculation unit 22, and the control value determination unit 23.

[0101] The first control value calculation unit 21 uses a mathematical model to calculate a control value for controlling a predetermined process of the factory. Specifically, the first control value calculation unit 21 uses a mathematical model to calculate a predicted value (first predicted value) of the steam outlet temperature in the same manner as the first prediction unit 12 of the first embodiment. Moreover, the first control value calculation unit 21 calculates control values ​​for the primary control valve 12B and the secondary control valve 22B based on the calculated first predicted value. Since the control object is the control valve, the control value is the opening of the control valve. Hereinafter, the control value calculated by the first control value calculation unit 21 is referred to as the first control value.

[0102] The second control value calculation unit 22 calculates the control value using the learned model that models the process by machine learning. Specifically, the second control value calculation unit 22 calculates the predicted value (second predicted value) of the steam outlet temperature using the same learned model as the second prediction unit 13 of the first embodiment. Moreover, the second control value calculation unit 22 calculates the control values ​​(opening degrees) for the primary control valve 12B and the secondary control valve 22B based on the calculated second predicted value. Hereinafter, the control value calculated by the second control value calculation unit 22 is referred to as the second control value.

[0103] The control value determination unit 23 determines the control value used for controlling the plant based on at least one of the first control value and the second control value. The control value determination unit 23 determines the control value in the same manner as the predicted value determination unit 15 determines the predicted value.

[0104] Furthermore, the plant control unit 16 of the present embodiment controls the plant based on at least one of the control values ​​determined by the control value determination unit 23, namely, the first control value and the second control value. As described above, the prediction based on the learned model has high accuracy for the prediction of a process exhibiting a nonlinear state, while the prediction based on the mathematical model is a linear prediction, and the possibility of calculating a predicted value that deviates from reality is low. Therefore, according to the structure of the present embodiment, a control value that takes both reliability and accuracy into consideration can be determined, and reliable control can be performed stably.

[0105] In addition, the plant control unit 16 of the present embodiment controls the plant based on the operation mode determined by the operation mode determination unit 14. This allows appropriate control to be performed at an earlier timing than the FB control in consideration of the difference in the operation mode.

[0106] Specifically, the second control value calculation unit 22 is connected to the second prediction unit 13 described in the second embodiment (see Figure 6) Similarly, the second prediction value is calculated based on the learned model for the normal operation mode and the learned model for the SB operation mode. Furthermore, the second control value calculation unit 22 calculates the second control value for the normal operation mode and the second control value for the SB operation mode based on the calculated second prediction values, and outputs the second control values ​​to the control value determination unit 23.

[0107] Furthermore, if the operation mode determined by the operation mode determination unit 14 is the SB operation mode, the control value determination unit 23 determines the control value for plant control based on at least one of the second control value for the SB operation mode and the first control value output from the first control value calculation unit 21. Thus, plant control based on the operation mode determined by the operation mode determination unit 14 is performed.

[0108] In addition, similarly to the second control value calculation unit 22 , the first control value calculation unit 21 can calculate the first control value for the normal operation mode using the mathematical model for the normal operation mode and calculate the first control value for the SB operation mode using the mathematical model for the SB operation mode.

[0109] [Variation]

[0110] In the above-mentioned embodiments, an example of switching the model used in the normal operation mode and the SB operation mode is described, but it is also possible to configure such that the same switching of the used model is performed in any operation mode that is characteristic of the target plant. In addition, in the parameter prediction of the plant, for example, a model that takes into account the control instructions executed in the plant (in the example of the above-mentioned embodiment, the control instructions of the opening of the regulating valve) and a model that does not take into account the control instructions can be used. In this case, the model that takes into account the control instructions can be used for prediction during the period when the control instructions are executed, and the model that does not take into account the control instructions can be used for prediction during the period when the control instructions are not executed.

[0111] It is also possible to make an external device (such as a server) of the information processing device 1 have a part of the functions of the information processing device 1 described in the above embodiments, and constitute an information processing system including the device and the information processing device 1. If it is such an information processing system, it is possible to achieve the same effect as the information processing device 1. For example, the operation processing using NN usually has a high load, so by making the external device have such an operation processing function, the load of the information processing device 1 can be reduced. In addition, for example, it is also possible to configure so that instead of providing the plant control unit 16, the external plant control device of the information processing device 1 performs plant control.

[0112] In addition, in the above-mentioned embodiments, examples of predicting the steam outlet temperature in a waste incineration facility and controlling the steam outlet temperature to approach a target value are described. However, these are only examples, and the object of prediction or control of the present invention, i.e., the factory, is not limited to a waste incineration facility, the object of prediction is not limited to the steam temperature, and the control is not limited to controlling the steam temperature to approach a target value.

[0113] For example, in a thermal power generation facility, the same process as in a waste incineration facility is performed by changing the combustion material from waste to a predetermined fuel (oil, coal). Therefore, similarly to the above-mentioned embodiments, it is possible to calculate the predicted values ​​of parameters for various processes in a thermal power generation facility. In addition, the process in the superheater B can be replaced by a process in a heat exchanger other than the superheater.

[0114] 〔Software-based implementation example〕

[0115] The control module of the information processing device 1 (particularly, each unit included in the control unit 10 ) may be realized by a logic circuit (hardware) formed on an integrated circuit (IC chip) or the like, or may be realized by software.

[0116] In the latter case, the information processing device 1 has a computer that executes commands of software, i.e., a program, that realizes each function. The computer has, for example, one or more processors and a computer-readable storage medium that stores the program. Moreover, in the computer, the processor reads the program from the storage medium and executes it, thereby achieving the purpose of the present invention. As the processor, for example, a CPU (Central Processing Unit) can be used. As the storage medium, a "non-temporary tangible medium" can be used, for example, in addition to ROM (Read Only Memory), a magnetic tape, an optical disk, a card, a semiconductor memory, a programmable logic circuit, etc. can also be used. In addition, a RAM (Random Access Memory) that expands the program can also be provided to the computer via any transmission medium (communication network, radio waves, etc.) that can transmit the program. In addition, one embodiment of the present invention can also be implemented in the form of a data signal embedded in a carrier wave that is embodied by electronically transmitting the program.

[0117] [Appendix]

[0118] An information processing device according to one embodiment of the present invention comprises: a first prediction unit, which uses a mathematical model that models a specified process of a factory to calculate a predicted value of a specified parameter related to the process, namely a first predicted value; a second prediction unit, which uses a learned model that models the above process through machine learning to calculate a predicted value of the above parameter, namely a second predicted value; and a predicted value determination unit, which determines the predicted value of the above parameter based on at least any one of the above first predicted value and the above second predicted value.

[0119] In the above-mentioned information processing device, it may be that, with respect to the above-mentioned prediction value determination unit, if the input data input by the above-mentioned second prediction unit to the above-mentioned learned model is within the range of the teacher data used in the mechanical learning of the above-mentioned learned model, the above-mentioned second prediction value is determined as the prediction value of the above-mentioned parameter, and if the input data input by the above-mentioned second prediction unit to the above-mentioned learned model is outside the range of the teacher data used in the mechanical learning of the above-mentioned learned model, the above-mentioned first prediction value is determined as the prediction value of the above-mentioned parameter.

[0120] In addition, it can be configured that the above-mentioned information processing device has an operation mode determination unit, which determines the operation mode of the above-mentioned factory, the above-mentioned first prediction unit uses a mathematical model corresponding to the operation mode determined by the above-mentioned operation mode determination unit to calculate the above-mentioned first prediction value, and the above-mentioned second prediction unit uses a learned model corresponding to the operation mode determined by the above-mentioned operation mode determination unit to calculate the above-mentioned second prediction value.

[0121] Furthermore, the information processing device may include an input data generating unit that calculates at least a part of input data for at least any one of the mathematical model and the learned model using another mathematical model.

[0122] Furthermore, the information processing device may include a plant control unit configured to control the plant based on the predicted value determined by the predicted value determination unit.

[0123] In addition, the above-mentioned prescribed process may include a first sub-process and a second sub-process executed after the first sub-process, and as far as the above-mentioned factory control unit is concerned, with respect to the above-mentioned first sub-process, control is performed so that the above-mentioned predicted value determined by the above-mentioned predicted value determination unit is close to a prescribed target value, and with respect to the above-mentioned second sub-process, control is performed so that the above-mentioned predicted value determined by the above-mentioned predicted value determination unit is close to a prescribed target value.

[0124] In addition, an information processing device according to another embodiment of the present invention comprises: a first control value calculation unit, which uses a mathematical model to calculate a first control value used in controlling a specified process of a factory; a second control value calculation unit, which uses a learned model that models the above process through machine learning to calculate a second control value used in controlling the above process; a control value determination unit, which determines the control value used in controlling the above factory based on at least any one of the above first control value and the above second control value; and a factory control unit, which controls the above factory based on the above control value.

[0125] The information processing device may include an operation mode determination unit that determines an operation mode of the plant, and the plant control unit may control the plant based on the operation mode determined by the operation mode determination unit.

[0126] In addition, an information processing method of one embodiment of the present invention is an information processing method of an information processing device, comprising: a first prediction step, using a mathematical model that models a specified process of a factory to calculate a predicted value of a specified parameter related to the process, i.e., a first predicted value; a second prediction step, using a learned model that models the above process through machine learning to calculate a predicted value of the above parameter, i.e., a second predicted value; and a predicted value determination step, determining the predicted value of the above parameter based on at least any one of the above first predicted value and the above second predicted value.

[0127] Furthermore, the present invention also includes an information processing program that causes a computer to function as the above-mentioned information processing device and causes the computer to function as the above-mentioned first prediction unit, the above-mentioned second prediction unit, and the above-mentioned prediction value determination unit.

[0128] The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope of the claims. Embodiments obtained by appropriately combining technical solutions disclosed in different embodiments are also included in the technical scope of the present invention.

[0129] Description of Reference Numerals

[0130] 1-information processing device; 11-input data generating unit; 12-first prediction unit; 13-second prediction unit; 14-operation mode determining unit; 15-prediction value determining unit; 16-plant control unit; 21-first control value calculating unit; 22-second control value calculating unit; 23-control value determining unit.

Claims

1. An information processing device, characterized in that: have: Input data generation unit; a first prediction unit that calculates a first prediction value, which is a prediction value of a predetermined parameter related to the process, using a mathematical model that models a predetermined process of the factory; a second prediction unit that calculates a second prediction value that is a prediction value of the parameter using a learned model that models the process through machine learning; and a predicted value determination unit that determines a predicted value of the parameter based on at least one of the first predicted value and the second predicted value, The input data generating unit calculates at least a portion of the input data for at least one of the mathematical model and the learned model using another mathematical model. Regarding the above-mentioned predicted value determination unit, If the input data inputted by the second prediction unit to the learned model is within the range of the teacher data used in the machine learning of the learned model, the second prediction value is determined as the prediction value of the parameter. The second prediction unit determines the first prediction value as the prediction value of the parameter if the input data input to the learned model is outside the range of teacher data used for machine learning of the learned model.

2. The information processing device according to claim 1, characterized in that The information processing device includes an operation mode determination unit that determines an operation mode of the plant. The first prediction unit calculates the first prediction value using a mathematical model corresponding to the operation mode determined by the operation mode determination unit. The second prediction unit calculates the second prediction value using a learned model corresponding to the operation mode determined by the operation mode determination unit.

3. The information processing device according to claim 1 or 2, characterized in that: The information processing device includes a plant control unit configured to control the plant based on the predicted value determined by the predicted value determination unit.

4. The information processing device according to claim 3, characterized in that: The above-mentioned prescribed process includes a first sub-process and a second sub-process performed after the first sub-process. As for the above-mentioned factory control unit, with respect to the above-mentioned first sub-process, control is performed to make the above-mentioned predicted value determined by the above-mentioned predicted value determination unit approach the prescribed target value, and with respect to the above-mentioned second sub-process, control is performed to make the above-mentioned predicted value determined by the above-mentioned predicted value determination unit approach the prescribed target value.

5. An information processing device, characterized in that: have: Input data generation unit; a first control value calculation unit that calculates a first control value used in controlling a predetermined process of the factory using a mathematical model; a second control value calculation unit that calculates a second control value used in controlling the process using a learned model obtained by modeling the process through machine learning; A control value determination unit that determines a control value used in controlling the plant based on at least one of the first control value and the second control value; as well as a plant control unit that controls the plant based on the control value, The input data generating unit calculates at least a portion of the input data for at least one of the mathematical model and the learned model using another mathematical model. Regarding the control value determination unit, If the input data inputted by the second control value calculation unit to the learned model is within the range of the teacher data used in the machine learning of the learned model, the second control value is determined as the control value used in the control of the plant. If the input data inputted into the learned model by the second control value calculation unit is out of the range of teacher data used for machine learning of the learned model, the first control value is determined as the control value used for controlling the plant.

6. The information processing device according to claim 5, characterized in that: The information processing device includes an operation mode determination unit that determines an operation mode of the plant. The plant control unit controls the plant based on the operation mode determined by the operation mode determination unit.

7. An information processing method, which is an information processing method of an information processing device, characterized in that: Include: Input data generation step; A first prediction step is to calculate a predicted value, i.e., a first predicted value, of a predetermined parameter related to the process using a mathematical model that models a predetermined process of the factory; A second prediction step of calculating a predicted value of the parameter, that is, a second predicted value, using a learned model that models the process by machine learning; and A prediction value determination step is to determine the prediction value of the parameter based on at least one of the first prediction value and the second prediction value. In the input data generating step, another mathematical model is used to calculate at least a portion of the input data for at least one of the mathematical model and the learned model. In the above prediction value determination step, If the input data input to the learned model in the second prediction step is within the range of the teacher data used in the machine learning of the learned model, the second prediction value is determined as the prediction value of the parameter, If the input data input to the learned model in the second prediction step is outside the range of teacher data used in machine learning of the learned model, the first prediction value is determined as the prediction value of the parameter.

8. A computer-readable storage medium storing an information processing program, wherein the information processing program enables a computer to function as the information processing device described in claim 1, and enables the computer to function as the above-mentioned input data generation unit, the above-mentioned first prediction unit, the above-mentioned second prediction unit, and the above-mentioned prediction value determination unit.

Images