Plant system, detection system, detection method, and detection program

Abstract

Monitor the plant's operating status. [Solution] The plant system includes a generation unit that generates a first data pattern by operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained prediction model that predicts the process values ​​of the plant using the calculation results of the tuned physical model, using a dataset of the plant in a predetermined state, modifying a portion of the data, and operating the model using a dataset that includes the modified data; and an output unit that compares a second data pattern generated by operating the tuned physical model and the trained prediction model using measured values ​​taken in the plant's processing steps with the first data pattern and outputs the comparison result.

Description

Technical Field

[0001] The present disclosure relates to a plant system, a detection system, a detection method, and a detection program in a plant.

Background Art

[0002] In recent plant control systems, the introduction of AI (Artificial Intelligence) in various forms has been proposed. As an example, in a plant control system for an incinerator, a prediction technique has been proposed in which image data (measurement values) taken inside the furnace is input into AI to predict the process values of the incinerator (Patent Document 1). According to this prediction technique, by predicting the process values based on the image data (measurement values), it becomes possible to predict in advance the operating status of the incinerator, specifically, the incineration state of the incinerator (normal state to abnormal state), the combustion efficiency of the incinerator, etc. As a result, since it becomes possible to take measures such as manual operation as needed, the possibility of achieving a stable and efficient operating status in the incinerator is increasing.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, since the image data (measurement values) only fragmentarily represents the behavior of the incineration process in the incinerator, it is difficult to accurately predict the process values of the incinerator even if AI is used.

[0005] In response to this, the applicant is working on developing a prediction technology that generates multiple physical models to reproduce the behavior of the incineration process, calculates unobservable state variables in the incineration process, and uses observable measurements and these unobservable state variables to predict the process values ​​of the incinerator. This prediction technology makes it possible to predict the process values ​​of the incinerator with high accuracy.

[0006] Furthermore, in plant control as described above, it is desirable not only to predict process values ​​with high accuracy, but also to detect signs of changes in the plant's state, such as what measurement values ​​indicate that the plant is in a normal or abnormal state. In particular, being able to detect signs of abnormality while monitoring the plant's operating status is considered useful in achieving stable and efficient operation of the plant.

[0007] This disclosure is intended to monitor the operating status of the plant, and in particular to detect signs of anomalies during monitoring. [Means for solving the problem]

[0008] The plant system relating to the first aspect of this disclosure is A generation unit generates a first data pattern by operating a pre-tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the pre-tuned physical model, using a dataset of the plant in a predetermined state, by modifying a portion of the data and operating the model using a dataset that includes the modified data. The system includes an output unit that compares a second data pattern, generated by operating the adjusted physical model and the trained prediction model using the measured values ​​taken in the processing step of the plant, with the first data pattern and outputs the comparison result.

[0009] The detection system relating to the second aspect of this disclosure is: A generation unit generates data patterns during state changes in the plant by operating a pre-tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the pre-tuned physical model, using a dataset that includes some data indicating signs of state changes. The system includes a detection unit that detects signs of a state change by comparing data patterns generated by operating the adjusted physical model and the trained prediction model using measurements taken in the processing steps of the plant with data patterns at the time of the state change.

[0010] A third aspect of this disclosure is a detection system as described in the second aspect, The generation unit stores the cause of the state change corresponding to a dataset that includes some data indicating the precursor of the state change, in association with the data pattern at the time of the state change. When the detection unit detects an indication of a change in state, it outputs the cause of the change in state along with the detection result.

[0011] The plant system relating to the fourth aspect of this disclosure is: A generation unit generates a first data pattern when the process values ​​of the plant are predicted in a predetermined state by operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the process values ​​of the plant using the calculation results of the tuned physical model. The system includes an output unit that compares a second data pattern, generated by operating the adjusted physical model and the trained prediction model using the measured values ​​taken in the processing step of the plant, with the first data pattern and outputs the comparison result.

[0012] The detection system relating to the fifth aspect of this disclosure is A generation unit generates a data pattern for the state change of the plant when a process value with data indicating a state change is predicted, by operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained prediction model that predicts the process values ​​of the plant using the calculation results of the tuned physical model. The system includes a detection unit that detects signs of a state change by comparing data patterns generated by operating the adjusted physical model and the trained prediction model using measurements taken in the processing steps of the plant with data patterns at the time of the state change.

[0013] A sixth aspect of this disclosure is a detection system as described in the fifth aspect, The generation unit generates a data pattern during a state change by searching for state variables and measured values ​​that enable the trained prediction model to operate so that a process value with data indicating the state change is predicted.

[0014] A seventh aspect of this disclosure is a detection system as described in the sixth aspect, The generation unit stores the cause of the state change identified by searching for the state variable and measured value, in association with the data pattern at the time of the state change. When the detection unit detects an indication of a change in state, it outputs the cause of the change in state along with the detection result.

[0015] An eighth aspect of this disclosure is a detection system as described in the sixth aspect, If the cause of the state change cannot be identified by searching for the state variables and measured values, the generation unit searches for measured values ​​that enable the adjusted physical model to operate in such a way that the searched state variables can be calculated.

[0016] A ninth aspect of this disclosure is a detection system as described in the second or fifth aspect, The aforementioned adjusted physical model is The parameters of the physical model are adjusted using a Kalman filter so that the calculated value calculated by the physical model by inputting the measured value measured in the processing step of the plant approaches the measured value measured in the processing step of the plant.

[0017] The tenth aspect of the present disclosure is the detection system according to the second or fifth aspect, The learned prediction model is Constructed by performing learning of the prediction model using the unobservable state variables in the processing step of the plant and the measured values measured in the processing step of the plant, which are calculated by operating the adjusted physical model using the measured values measured in the processing step of the plant.

[0018] The eleventh aspect of the present disclosure is the detection system according to the second aspect, The processing step of the plant is an incineration processing step of an incinerator, The adjusted physical model is an adjusted fire grate combustion model, or an adjusted gas phase combustion model, or an adjusted boiler heat absorption and power generation model.

[0019] The twelfth aspect of the present disclosure is the detection system according to the eleventh aspect, The data pattern at the time of the state change is The measured value input to the adjusted fire grate combustion model, or the measured value after addition to which data indicating a sign of a state change is added to the measured value, The measured value input to the adjusted gas phase combustion model, or the measured value after addition to which data indicating a sign of a state change is added to the measured value, The measured value input to the adjusted boiler heat absorption and power generation model, or the measured value after addition to which data indicating a sign of a state change is added to the measured value, The measured value input to the learned prediction model, or the measured value after addition to which data indicating a sign of a state change is added to the measured value, An unobservable state variable calculated by the aforementioned adjusted grate combustion model, or a state variable to which data indicating a precursor to a state change has been added, An unobservable state variable calculated by the aforementioned adjusted gas-phase combustion model, or a state variable to which data indicating a precursor to a state change has been added, An unobservable state variable calculated by the adjusted boiler heat acquisition and power generation model, or a state variable to which data indicating a precursor to a state change has been added, This includes process values ​​predicted by the aforementioned trained predictive model.

[0020] A thirteenth aspect of this disclosure is a detection system as described in the eleventh aspect, The aforementioned adjusted grate combustion model is This is a physical model that reproduces the behavior of the incineration process of the aforementioned incinerator when the waste introduced is completely combusted by sequentially going through the processes of drying, thermal decomposition, and combustion while moving on a movable grate, and is an adjusted physical model in which the parameters have been adjusted based on the measured values ​​taken in the incineration process.

[0021] A fourteenth aspect of this disclosure is a detection system as described in the eleventh aspect, The aforementioned modified gas-phase combustion model is This is a physical model that reproduces the behavior of the combustible gas generated by combustion in the incineration process of the aforementioned incinerator when it is combusted by supplied air, and is an adjusted physical model in which the parameters have been adjusted based on measurements taken in the incineration process.

[0022] A fifteenth aspect of this disclosure is a detection system as described in the eleventh aspect, The aforementioned adjusted boiler heat recovery / power generation model is, This is a physical model that reproduces the behavior of the incineration process of the aforementioned incinerator, specifically when steam generated by the absorption of heat from exhaust gas in a waste heat boiler rotates a steam turbine and generates electricity, and is an adjusted physical model in which the parameters have been adjusted based on measurements taken during the incineration process.

[0023] A sixteenth aspect of this disclosure is a detection system as described in the eleventh aspect, The aforementioned trained predictive model is This is a trained predictive model that predicts process values ​​of the incinerator, constructed by training a predictive model using the unobservable state variables in the incineration process, calculated by operating the adjusted grate combustion model, the adjusted gas phase combustion model, and the adjusted boiler heat recovery and power generation model using the measured values ​​taken in the incineration process of the incinerator, and the measured values ​​taken in the incineration process.

[0024] The detection method relating to the 17th aspect of this disclosure is: A first data pattern is generated by operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the tuned physical model, using a dataset of the plant in a predetermined state, by modifying a portion of the data and operating the model using a dataset that includes the modified data. Each computer performs the following steps: using the measurements taken in the processing step of the plant, it operates the adjusted physical model and the trained prediction model to generate a second data pattern, which is then compared with the first data pattern and the comparison result is output.

[0025] The detection method relating to the 18th aspect of this disclosure is: A process to generate data patterns during state changes in the plant by operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the tuned physical model, using a dataset that includes data that indicates signs of state changes; Each computer performs a step of detecting signs of a state change by comparing the data patterns generated by operating the adjusted physical model and the trained prediction model using the measured values ​​taken in the processing steps of the plant with the data patterns at the time of the state change.

[0026] The detection method relating to the 19th aspect of this disclosure is: A step of generating a first data pattern when the process values ​​of the plant in a predetermined state are predicted by operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the process values ​​of the plant using the calculation results of the tuned physical model, Each computer performs the following steps: using the measurements taken in the processing step of the plant, it operates the adjusted physical model and the trained prediction model to generate a second data pattern, which is then compared with the first data pattern and the comparison result is output.

[0027] The detection method relating to the 20th aspect of this disclosure is: A process of generating a data pattern for a change in the state of the plant when a process value with data indicating a change in state is predicted, by operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the process values ​​of the plant using the calculation results of the tuned physical model, Each computer performs a step of detecting signs of a state change by comparing the data patterns generated by operating the adjusted physical model and the trained prediction model using the measured values ​​taken in the processing steps of the plant with the data patterns at the time of the state change.

[0028] The detection program relating to the 21st aspect of this disclosure is: When operating a pre-tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the pre-tuned physical model, using a dataset of the plant in a predetermined state, a portion of the data is modified, and the model is operated using a dataset that includes the modified data, thereby enabling the computer of the detection device that stores the first data pattern to perform the operation. Using the measurements taken in the processing step of the plant, the second data pattern generated by operating the adjusted physical model and the trained prediction model is compared with the first data pattern, and a comparison result is output.

[0029] The detection program relating to the 22nd aspect of this disclosure is: By operating a pre-tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the pre-tuned physical model, with a dataset that includes some data indicating signs of state changes, the computer of the detection device that stores data patterns during state changes in the plant can be configured to: Using the measurements taken in the processing steps of the plant, the system performs a step to detect signs of a state change by comparing the data patterns generated by operating the adjusted physical model and the trained prediction model with the data patterns at the time of the state change.

[0030] The detection program relating to the 23rd aspect of this disclosure is: By operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the tuned physical model, the computer of the detection device that stores the first data pattern when the process values ​​for a predetermined state of the plant are predicted, Using the measurements taken in the processing step of the plant, the second data pattern generated by operating the adjusted physical model and the trained prediction model is compared with the first data pattern, and a comparison result is output.

[0031] The detection program relating to the 24th aspect of this disclosure is: By operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the tuned physical model, when process values ​​with data indicating state changes are predicted, the computer of the detection device that stores the data pattern at the time of the plant's state change is configured to: Using the measurements taken in the processing steps of the plant, the system performs a step to detect signs of a state change by comparing the data patterns generated by operating the adjusted physical model and the trained prediction model with the data patterns at the time of the state change. [Effects of the Invention]

[0032] According to this disclosure, it will be possible to detect signs of changes in the state of a plant, particularly signs of abnormalities. [Brief explanation of the drawing]

[0033] [Figure 1] This figure shows an example of the system configuration during the learning phase of the target plant. [Figure 2] This figure shows an example of the hardware configuration of a learning device. [Figure 3A] This is a diagram showing an example of an incinerator configuration. [Figure 3B] This figure shows an example of multiple physical models that reproduce the behavior of the incineration process in an incinerator. [Figure 4] This figure shows an example of the functional configuration of the learning device during the parameter tuning phase. [Figure 5] This is the first diagram illustrating the overview of the parameter adjustment process. [Figure 6] This is the second diagram, illustrating the overview of the parameter adjustment process. [Figure 7] This is the third diagram, illustrating the overview of the parameter adjustment process. [Figure 8] This figure shows an example of the functional configuration of the learning device during the predictive model construction phase. [Figure 9] This figure shows a specific example of the learning process performed by a learning device during the predictive model construction phase. [Figure 10] This is a flowchart showing the processing flow during the learning phase. [Figure 11] Figure 1 shows an example of the system configuration during the prediction phase of the target plant. [Figure 12] This figure shows an example of the hardware configuration of a pattern generation device and an anomaly detection device. [Figure 13] The first figure shows an example of the functional configuration of the pattern generation device during the pattern generation phase. [Figure 14A] The first figure shows an example of the functional configuration of the gray box model section. [Figure 14B] The second figure shows an example of the functional configuration of the gray box model section. [Figure 15] Figure 1 shows a specific example of data patterns during abnormal conditions. [Figure 16] The second figure shows a specific example of data patterns during abnormal conditions. [Figure 17] This is the first flowchart illustrating the pattern generation process using a pattern generation device. [Figure 18] This figure shows an example of the functional configuration of an anomaly detection device during the anomaly detection phase. [Figure 19] This flowchart shows the flow of the anomaly detection process by the anomaly detection device. [Figure 20] The second figure shows an example of the system configuration during the prediction phase of the target plant. [Figure 21] The second figure shows an example of the functional configuration of the pattern generation device during the pattern generation phase. [Figure 22] This diagram illustrates the general outline of the pattern generation process performed by the pattern generation device. [Figure 23] The third figure shows a specific example of data patterns during abnormal conditions. [Figure 24] This is the second flowchart illustrating the pattern generation process using a pattern generation device. [Modes for carrying out the invention]

[0034] Each embodiment will be described below with reference to the attached drawings. In this specification and the drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant descriptions will be omitted.

[0035] [First Embodiment] <System configuration during the learning phase of the target plant> First, we will describe the system configuration during the learning phase of the target plant. Figure 1 shows an example of the system configuration during the learning phase of the target plant.

[0036] As shown in Figure 1, the target plant 10 comprises the target equipment 100 and the plant control system 110. In the first embodiment, the target plant 10 is, for example, a general waste treatment plant intended for waste disposal.

[0037] The target equipment 100 is, for example, an incinerator, which transmits measured values ​​taken during operation to the plant control system 110 and performs incineration based on control values ​​transmitted from the plant control system 110.

[0038] The plant control system 110 includes a monitoring and control unit 111. The monitoring and control unit 111 monitors the target equipment 100 in operation based on measurements transmitted from the target equipment 100. The monitoring and control unit 111 also calculates control values ​​to operate the target equipment 100 so that the process values ​​used to control the processing process (in the case of the target equipment 100 being an incinerator, the incineration process) from the measurements transmitted from the target equipment 100 match the target values. Furthermore, the monitoring and control unit 111 transmits the calculated control values ​​to the target equipment 100.

[0039] The learning device 120 has a physical model that reproduces the behavior of the processing steps of the target equipment 100.

[0040] The learning device 120 acquires measurement values ​​measured in the processing steps of the target equipment 100 from the plant control system 110, and generates an adjusted physical model by adjusting the parameters of the physical model using the acquired measurement values. The phase in the learning phase in which the learning device 120 adjusts the parameters of the physical model and generates an adjusted physical model is called the "parameter adjustment phase."

[0041] Furthermore, the learning device 120 has a predictive model that predicts process values ​​used to control the processing steps of the target equipment 100. The learning device 120 uses unobservable state variables calculated by the adjusted physical model and measured values ​​taken in the processing steps of the target equipment 100 to train the predictive model and construct a trained predictive model. The phase in the learning process in which the learning device 120 trains the predictive model to construct a trained predictive model is referred to as the "predictive model construction phase."

[0042] In other words, in this embodiment, the learning phase includes a parameter tuning phase and a predictive model building phase.

[0043] If the target equipment 100 is an incinerator, the behavior of the incineration process can be reproduced by running simulations using multiple physical models (e.g., a grate combustion model, a gas-phase combustion model, a boiler heat absorption / power generation model, etc. Details will be described later).

[0044] Thus, the learning device 120 is • It has multiple physical models to reproduce the behavior of the processing steps of the target equipment 100. • Multiple physical models are adjusted using measurements taken during the processing steps of the target equipment 100.

[0045] As a result, the learning device 120 can generate multiple adjusted physical models that accurately reproduce the behavior of the processing steps of the target equipment 100.

[0046] Furthermore, the learning device 120 is • Multiple pre-tuned physical models each run simulations to reproduce the behavior of the processing steps of the target equipment 100, thereby calculating highly accurate unobservable state variables. In addition to the measured values ​​taken during the processing steps of the target equipment 100, a predictive model is trained using highly accurate, unobservable state variables.

[0047] As a result, the learning device 120 can construct a trained predictive model that can predict process values ​​with higher accuracy compared to a trained predictive model that was trained based only on measurements taken in the processing steps of the target equipment 100.

[0048] <Hardware configuration of the learning device> Next, the hardware configuration of the learning device 120 will be described. Figure 2 shows an example of the hardware configuration of the learning device.

[0049] As shown in Figure 2, the learning device 120 includes a processor 201, memory 202, auxiliary storage device 203, connection device 204, communication device 205, and drive device 206. The processor 201, memory 202, auxiliary storage device 203, connection device 204, communication device 205, and drive device 206 of the learning device 120 are interconnected via a bus 207.

[0050] The processor 201 has various computing devices such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). The processor 201 reads various programs (for example, learning programs, etc.) into memory 202 and executes them.

[0051] Memory 202 has main memory devices such as ROM (Read Only Memory) and RAM (Random Access Memory). The processor 201 and memory 202 form a so-called computer, and the computer realizes various functions by having the processor 201 execute various programs read from memory 202.

[0052] The auxiliary storage device 203 stores various programs and various information used when those programs are executed by the processor 201.

[0053] The connection device 204 is a device for connecting the learning device 120 to an example of an external device, such as the operating device 211 and the display device 212.

[0054] The communication device 205 is a device for communicating with various devices (for example, the plant control system 110) via a network.

[0055] The drive device 206 is a device for setting the recording medium 213. The recording medium 213 here includes media that record information optically, electrically, or magnetically, such as CD-ROMs, flexible disks, and magneto-optical disks. The recording medium 213 may also include semiconductor memory that records information electrically, such as ROMs and flash memory.

[0056] The various programs to be installed on the auxiliary storage device 203 are installed, for example, when the distributed recording medium 213 is set in the drive device 206 and the various programs recorded on the recording medium 213 are read. Alternatively, the various programs to be installed on the auxiliary storage device 203 may be installed by downloading them from the network via the communication device 205.

[0057] <Outline configuration of the incinerator> Next, we will describe the general configuration of an incinerator, which is an example of the target equipment 100. Figure 3A is a diagram showing an example of the configuration of an incinerator, and it shows an example of the configuration of an incinerator called a stoker-type incinerator.

[0058] As shown in Figure 3A, in a stoker-type incinerator, the waste that is fed in moves on a movable grate. Air is supplied to the movable grate, and the waste moving on the grate undergoes complete combustion by sequentially going through the processes of drying, thermal decomposition, and combustion.

[0059] The incinerated ash generated by combustion is sent to an ash pit (not shown). The exhaust gas generated by combustion (including exhaust gas generated from the combustion of combustible gases) is absorbed in a waste heat boiler and then sent to a filtration-type dust collector (not shown). The steam generated by the heat absorption in the waste heat boiler is sent to a steam turbine and used for power generation.

[0060] In this embodiment, the following three physical models are used to reproduce the behavior of the incineration process of the stoker-type incinerator shown in Figure 3A. • Grate combustion model: A physical model that reproduces the behavior when waste introduced in the region indicated by reference numeral 310 undergoes complete combustion by sequentially going through the processes of drying, thermal decomposition, and combustion as it moves along a movable grate. • Gas-phase combustion model: A physical model that reproduces the behavior when combustible gas generated by combustion is combusted by supplied air in the region indicated by reference numeral 320. • Boiler heat recovery / power generation model: A physical model that reproduces the behavior when a steam turbine rotates and generates electricity using steam generated by heat absorption in a waste heat boiler from exhaust gas in the region shown by reference numeral 330.

[0061] Each physical model is represented by a predetermined relational expression from which desired information can be calculated by performing a simulation. Specific examples include the relational expression disclosed in Japanese Patent Application No. 2023-182295, or the relational expression disclosed in Japanese Patent Application No. 2023-182691.

[0062] A grate combustion model is composed of one, two, or all of the following models: a water evaporation model, a volatile matter release model, and a fixed carbon combustion model, and is constructed according to the type of information to be calculated. This type of information includes, for example, the amount of water evaporation (amount of water vapor), the amount of combustible gas generated, and the amount of fixed carbon combustion. However, the models that constitute a grate combustion model are not limited to these.

[0063] The same applies to gas-phase combustion models and boiler heat recovery / power generation models, which are represented by predetermined relational equations that allow for the calculation of desired information by running simulations. These predetermined relational equations include, but are not limited to, those used to calculate information such as the amount of combustible gas burned, the heat recovery efficiency, and the amount of heat removed from the furnace body.

[0064] <Overview of the physical model> Next, we will describe the outlines of the three physical models mentioned above. Figure 3B shows an example of multiple physical models that reproduce the behavior of the incineration process in an incinerator.

[0065] As shown in Figure 3B, the grate combustion model 310M includes: Examples of measurements taken during the incineration process of an incinerator include the composition of the waste being fed in, the amount of waste being fed in, the amount of air supplied, the movement of the movable grate, etc. The amount of combustible gas burned, etc., was calculated by running a simulation using the gas-phase combustion model 320M. The following is entered.

[0066] The grate combustion model 310M uses this input information to perform the simulation. This allows the grate combustion model 310M to calculate observable information and unobservable state variables in the incinerator. In the example shown in Figure 3B, the observable information is as follows: • Temperature of the flammable gas, • Amount of water evaporation, These are calculated and, as unobservable state variables, • Composition of waste put in, • Distribution of waste dropped into the furnace, • Weight of waste per grate, • Waste temperature per grate, • Radiant heat from the fire, • Amount of water vapor, • Amount of combustible gas generated, • Composition of flammable gases, • Fixed carbon combustion amount, • Amount of waste moved, This shows how the calculations were performed.

[0067] In this context, observable information refers to information for which both measured values ​​obtained during the incineration process of the incinerator and calculated values ​​obtained by the grate combustion model 310M exist. In contrast, unobservable state variables (information) refer to information for which calculated values ​​obtained by the grate combustion model 310M exist, but measured values ​​obtained during the incineration process of the incinerator do not exist.

[0068] These state variables (e.g., water vapor amount, combustible gas generation amount, fixed carbon combustion amount) calculated by the grate combustion model 310M are input into the gas-phase combustion model 320M.

[0069] As shown in Figure 3B, the gas-phase combustion model 320M includes: • State variables calculated by the grate combustion model 310M (e.g., water vapor amount, combustible gas generation amount, fixed carbon combustion amount), Examples of measurements taken during the incineration process of an incinerator include the amount of air supplied to burn the combustible gas generated by combustion, The following is entered.

[0070] The gas-phase combustion model 320M performs a simulation using this input information. This allows the gas-phase combustion model 320M to calculate observable information and unobservable state variables in the incinerator. In the example in Figure 3B, the observable information is as follows: • Combustion control temperature, • Gas concentration, These are calculated and, as unobservable state variables, • Amount of combustible gas burned, • Composition of post-combustion gas, • Amount of heat removed from the furnace body, This shows how the calculations were performed.

[0071] In this context, observable information refers to information for which both measured values ​​obtained during the incineration process of the incinerator and calculated values ​​obtained by the gas-phase combustion model 320M exist. In contrast, unobservable state variables (information) refer to information for which calculated values ​​obtained by the gas-phase combustion model 320M exist, but for which no measured values ​​are obtained during the incineration process of the incinerator.

[0072] These state variables (e.g., combustion rate of combustible gas) calculated by the gas-phase combustion model 320M are input into the grate combustion model 310M and the boiler heat recovery / power generation model 330M.

[0073] As shown in Figure 3B, the boiler heat acquisition / power generation model 330M includes: • State variables calculated by the gas-phase combustion model 320M (e.g., amount of combustible gas burned), etc. Examples of measurements taken during the incineration process of an incinerator include exhaust gas flow rate, exhaust gas temperature, etc. The following is entered.

[0074] The boiler heat recovery / power generation model 330M performs a simulation using this input information. This allows the boiler heat recovery / power generation model 330M to calculate observable information and unobservable state variables in the incinerator. In the example in Figure 3B, the observable information is as follows: • Boiler heat output, • Amount of power generation, These are calculated and, as unobservable state variables, • Heat recovery efficiency, • Amount of heat removed from the furnace body, This shows how the calculations were performed.

[0075] In this context, observable information refers to information for which both measured values ​​obtained during the incineration process of the incinerator and calculated values ​​obtained by the boiler heat recovery / power generation model 330M exist. In this context, unobservable state variables (information) refer to information for which calculated values ​​obtained by the boiler heat recovery / power generation model 330M exist, but measured values ​​obtained during the incineration process of the incinerator do not exist.

[0076] <Functional configuration of the learning device (parameter tuning phase)> Next, the functional configuration of the learning device 120 in the parameter adjustment phase will be described. Figure 4 is a diagram showing an example of the functional configuration of the learning device in the parameter adjustment phase. As described above, a learning program is installed in the learning device 120, and when this program is executed, the learning device 120 functions as a measurement value acquisition unit 400 and a parameter adjustment unit 410 in the parameter adjustment phase.

[0077] The measurement value acquisition unit 400 acquires measurement values ​​measured during the incineration process of the incinerator from the plant control system 110 and stores the acquired measurement values ​​in the measurement value storage unit 450.

[0078] The parameter adjustment unit 410 includes a grate combustion model 310M, an adjustment unit 420, a gas phase combustion model 320M, an adjustment unit 430, a boiler heat recovery / power generation model 330M, and an adjustment unit 440.

[0079] The adjustment unit 420 adjusts the parameters of the grate combustion model 310M based on the calculated values ​​obtained by the grate combustion model 310M performing a simulation and the measured values ​​stored in the measured value storage unit 450.

[0080] The adjustment unit 430 adjusts the parameters of the gas-phase combustion model 320M based on the calculated values ​​obtained by the gas-phase combustion model 320M performing a simulation and the measured values ​​stored in the measured value storage unit 450.

[0081] The adjustment unit 440 adjusts the parameters of the boiler heat recovery / power generation model 330M based on the calculated values ​​obtained by the boiler heat recovery / power generation model 330M running a simulation and the measured values ​​stored in the measured value storage unit 450.

[0082] <Details of parameter adjustment processing by the learning device's parameter adjustment unit> Next, we will explain in detail the parameter adjustment process performed by the parameter adjustment unit 410 of the learning device 120 during the parameter adjustment phase.

[0083] (1) Details of the grate combustion model and parameter adjustment process by the adjustment unit First, we will explain the details of the parameter adjustment process by the grate combustion model 310M and the adjustment unit 420, which are included in the parameter adjustment unit 410. Figure 5 is the first diagram illustrating the overview of the parameter adjustment process.

[0084] As shown in Figure 5, among the measurement values ​​stored in the measurement value storage unit 450, the measurement values ​​used when the simulation is performed by the grate combustion model 310M (for example, the composition of the waste to be introduced and the measurement value of the amount of waste) are input to the grate combustion model 310M. This allows the grate combustion model 310M to perform the simulation under default parameters. At this time, the grate combustion model 310M performs the simulation. • Among the measurement values ​​stored in the measurement value storage unit 450, the calculated values ​​corresponding to measurement values ​​other than those used when running the simulation (for example, measurement values ​​other than the composition of the waste to be introduced and the amount of waste to be introduced) (for example, the calculated value of the amount of water evaporation), The result is calculated and output to the adjustment unit 420.

[0085] Furthermore, as shown in Figure 5, among the measurement values ​​stored in the measurement value storage unit 450, the measurement values ​​corresponding to the calculated values ​​calculated by the grate combustion model 310M (for example, the measurement value of water evaporation) are input to the adjustment unit 420. Note that the measurement values ​​corresponding to the calculated values ​​calculated by the grate combustion model 310M refer to the measurement values ​​measured in the incineration process of the incinerator for which there are calculated values ​​calculated by the grate combustion model 310M.

[0086] The adjustment unit 420 includes a Kalman filter 421. The Kalman filter 421 is The calculated values ​​(calculated water evaporation amounts) obtained using the 310M grate combustion model are: • Of the measurement values ​​stored in the measurement value storage unit 450, the measurement value corresponding to the calculated value obtained by the grate combustion model 310M (measurement value of water evaporation amount) is selected. The parameters of the grate combustion model 310M are updated to approximate the desired result. This generates the adjusted grate combustion model 310M'.

[0087] Although not shown in Figure 5, when the grate combustion model 310M performs a simulation, state variables (e.g., combustible gas combustion rate) calculated by the adjusted gas-phase combustion model 320M' are input in addition to the measured values. However, at the stage of adjusting the parameters of the grate combustion model 310M, the parameters of the gas-phase combustion model 320M have not yet been adjusted. Therefore, it is assumed that predetermined values ​​are input to the grate combustion model 310M here.

[0088] (2) Details of the gas-phase combustion model and parameter adjustment process by the adjustment unit Next, the details of the parameter adjustment process by the gas-phase combustion model 320M and the adjustment unit 430, which are included in the parameter adjustment unit 410, will be explained. Figure 6 is a second diagram illustrating the overview of the parameter adjustment process.

[0089] As shown in Figure 6, among the measurement values ​​stored in the measurement value storage unit 450, the measurement values ​​used when the simulation is performed by the gas-phase combustion model 320M (e.g., amount of air) are input to the gas-phase combustion model 320M. In addition, state variables calculated by the simulation performed by the adjusted grate combustion model 310M' (e.g., amount of water evaporation (amount of water vapor)) are input to the gas-phase combustion model 320M. As a result, the gas-phase combustion model 320M performs the simulation under default parameters. At this time, the gas-phase combustion model 320M performs the simulation. • Among the measurement values ​​stored in the measurement value storage unit 450, the calculated values ​​corresponding to measurement values ​​other than those used when running the simulation (for example, measurement values ​​other than the air volume measurement), (for example, the calculated value of the combustion control temperature), The result is calculated and output to the adjustment unit 430.

[0090] Furthermore, as shown in Figure 6, among the measurement values ​​stored in the measurement value storage unit 450, the measurement values ​​corresponding to the calculated values ​​calculated by the gas-phase combustion model 320M (for example, the measurement value of the combustion control temperature) are input to the adjustment unit 430. Note that the measurement values ​​corresponding to the calculated values ​​calculated by the gas-phase combustion model 320M refer to the measurement values ​​measured in the incineration process of the incinerator for which there are calculated values ​​calculated by the gas-phase combustion model 320M.

[0091] The adjustment unit 430 includes a Kalman filter 431. The Kalman filter 431 is The calculated value (calculated combustion control temperature) obtained by the gas-phase combustion model 320M is, • Of the measurement values ​​stored in the measurement value storage unit 450, the measurement value corresponding to the calculated value calculated by the gas-phase combustion model 320M (measurement value of combustion control temperature) is selected. Update the parameters of gas-phase combustion model 320M to bring it closer to the desired result. This will generate the adjusted gas-phase combustion model 320M'.

[0092] (3) Details of the boiler heat recovery / power generation model and parameter adjustment process by the adjustment unit Next, we will explain the details of the parameter adjustment process by the boiler heat recovery / power generation model 330M and the adjustment unit 440, which are included in the parameter adjustment unit 410. Figure 7 is a third diagram illustrating the overview of the parameter adjustment process.

[0093] As shown in Figure 7, among the measurement values ​​stored in the measurement value storage unit 450, the measurement values ​​used when the simulation is performed by the boiler heat recovery / power generation model 330M (for example, the exhaust gas flow rate) are input to the boiler heat recovery / power generation model 330M. In addition, state variables (for example, the amount of combustible gas burned) calculated by the simulation performed by the adjusted gas phase combustion model 320M' are input to the boiler heat recovery / power generation model 330M. As a result, the boiler heat recovery / power generation model 330M performs the simulation under default parameters. At this time, the boiler heat recovery / power generation model 330M performs the simulation. • Among the measurement values ​​stored in the measurement value storage unit 450, the calculated values ​​corresponding to measurement values ​​other than those used when running the simulation (for example, measurement values ​​other than the exhaust gas flow rate measurement), (for example, the calculated value of boiler heat acquisition), The result is calculated and output to the adjustment unit 440.

[0094] Furthermore, as shown in Figure 7, among the measurement values ​​stored in the measurement value storage unit 450, the measurement values ​​corresponding to the calculated values ​​calculated by the boiler heat recovery / power generation model 330M (for example, the measured value of boiler heat recovery) are input to the adjustment unit 440. Note that the measurement values ​​corresponding to the calculated values ​​calculated by the boiler heat recovery / power generation model 330M refer to the measurement values ​​measured in the incineration process of the incinerator for which there are calculated values ​​calculated by the boiler heat recovery / power generation model 330M.

[0095] The adjustment unit 440 includes a Kalman filter 441. The Kalman filter 441 is The calculated value (calculated boiler heat gain) obtained using the boiler heat recovery / power generation model 330M is, • Of the measured values ​​stored in the measured value storage unit 450, the measured value corresponding to the calculated value calculated by the boiler heat recovery / power generation model 330M (measured value of boiler heat recovery) is selected. Update the parameters of the boiler heat recovery / power generation model 330M to bring it closer to the desired result. This will generate the adjusted boiler heat recovery / power generation model 330M'.

[0096] <Functional configuration of the learning device (predictive model construction phase)> Next, we will describe the functional configuration of the learning device 120 in the prediction model construction phase. Figure 8 shows an example of the functional configuration of the learning device in the prediction model construction phase. As mentioned above, a learning program is installed in the learning device 120, and when this program is executed, the learning device 120 functions as a measurement value acquisition unit 400 and a learning unit 800 in the prediction model construction phase. Of these, the measurement value acquisition unit 400 has already been explained using Figure 4, so its explanation will be omitted here.

[0097] The learning unit 800 includes a tuned grate combustion model 310M', a tuned gas-phase combustion model 320M', a tuned boiler heat acquisition / power generation model 330M', and a prediction model learning unit 810.

[0098] The adjusted grate combustion model 310M' is, • When running the simulation, the measured values ​​to be used are read from the measured value storage unit 450. • Observable state variables from one cycle prior, used when running the simulation, are obtained from the adjusted gas-phase combustion model 320M'. The simulation is executed. This causes the tuned grate combustion model 310M' to calculate state variables. The tuned grate combustion model 310M' also outputs to the tuned gas-phase combustion model 320M' the unobservable state variables from the calculated state variables that the tuned gas-phase combustion model 320M' will use when it runs its simulation. Furthermore, the tuned grate combustion model 310M' outputs to the prediction model learning unit 810 the unobservable state variables from the calculated state variables that the prediction model learning unit 810 will use when it learns its prediction model.

[0099] Note that an unobservable state variable from one cycle prior refers to an unobservable state variable calculated during the simulation execution one cycle prior, when the simulation is run at a predetermined interval.

[0100] The tuned vapor phase combustion model 320M' is, • When running the simulation, the measured values ​​to be used are read from the measured value storage unit 450. • Observable state variables from one cycle prior, used when running the simulation, are obtained from the adjusted grate combustion model 310M'. The simulation is executed. This causes the tuned gas-phase combustion model 320M' to calculate state variables. The tuned gas-phase combustion model 320M' also outputs the unobservable state variables from the calculated state variables that the tuned boiler heat recovery / power generation model 330M' will use when it runs its simulation to the tuned boiler heat recovery / power generation model 330M'. Furthermore, the tuned gas-phase combustion model 320M' outputs the unobservable state variables from the calculated state variables that the prediction model learning unit 810 will use when it trains its prediction model to the prediction model learning unit 810.

[0101] The tuned boiler heat recovery / power generation model 330M' is, • When running the simulation, the measured values ​​to be used are read from the measured value storage unit 450. • Observable state variables from one cycle prior, used when running the simulation, are obtained from the adjusted gas-phase combustion model 320M'. The simulation is executed. This causes the adjusted boiler heat recovery / power generation model 330M' to calculate state variables. The adjusted boiler heat recovery / power generation model 330M' also outputs to the prediction model learning unit 810 the unobservable state variables from the calculated state variables that the prediction model learning unit 810 will use when learning the prediction model.

[0102] The prediction model learning unit 810 obtains unobservable state variables to be used when training the prediction model from the tuned grate combustion model 310M', the tuned gas-phase combustion model 320M', and the tuned boiler heat acquisition / power generation model 330M'. The prediction model learning unit 810 also reads measured values ​​to be used when training the prediction model from the measured value storage unit 450.

[0103] The prediction model learning unit 810 uses the acquired unobservable state variables and the read-out measured values ​​as training data to train the prediction model and construct a trained prediction model.

[0104] <Details of the learning process by the predictive model learning unit of the learning device> Next, we will describe the details of the learning process performed by the predictive model learning unit 810 of the learning device 120 during the predictive model construction phase. Figure 9 is a diagram showing a specific example of the learning process performed by the learning device during the predictive model construction phase. As shown in Figure 9, during the predictive model construction phase, the predictive model learning unit 810 has a predictive model 900 and a comparison / modification unit 910.

[0105] As shown in Figure 9, among the measurement values ​​stored in the measurement value storage unit 450, the measurement values ​​used by the adjusted grate combustion model 310M' when running the simulation are input to the adjusted grate combustion model 310M'. In addition, the unobservable state variables calculated by the adjusted gas-phase combustion model 320M' one cycle earlier are input to the adjusted grate combustion model 310M'. As a result, the adjusted grate combustion model 310M' runs the simulation and calculates the unobservable state variables.

[0106] The unobservable state variables calculated by the adjusted grate combustion model 310M' are input into the prediction model 900. Additionally, the unobservable state variables calculated by the adjusted grate combustion model 310M' are input into the adjusted gas-phase combustion model 320M'.

[0107] As shown in Figure 9, among the measurement values ​​stored in the measurement value storage unit 450, the measurement values ​​used when the simulation is performed by the adjusted gas-phase combustion model 320M' are input to the adjusted gas-phase combustion model 320M'. In addition, the unobservable state variables calculated by the adjusted grate combustion model 310M' one cycle earlier are also input to the adjusted gas-phase combustion model 320M'. As a result, the adjusted gas-phase combustion model 320M' performs the simulation and calculates the unobservable state variables.

[0108] Unobservable state variables calculated by the tuned gas-phase combustion model 320M' are input into the prediction model 900. Additionally, unobservable state variables calculated by the tuned gas-phase combustion model 320M' are input into the tuned boiler heat gain / power generation model 330M'.

[0109] As shown in Figure 9, among the measured values ​​stored in the measured value storage unit 450, the measured values ​​used when the simulation is performed by the adjusted boiler heat recovery / power generation model 330M' are input into the adjusted boiler heat recovery / power generation model 330M'. In addition, unobservable state variables calculated by the adjusted gas-phase combustion model 320M' one cycle earlier are also input into the adjusted boiler heat recovery / power generation model 330M'. As a result, the adjusted boiler heat recovery / power generation model 330M' performs the simulation and calculates the unobservable state variables.

[0110] Unobservable state variables calculated by the adjusted boiler heat gain / power generation model 330M' are input into the prediction model 900.

[0111] As shown in Figure 9, the prediction model 900 is • Unobservable state variables calculated by the adjusted grate combustion model 310M', • Unobservable state variables calculated by the adjusted gas-phase combustion model 320M' • Unobservable state variables calculated by the adjusted boiler heat gain / power generation model 330M' • Measurement values ​​other than those input into the adjusted grate combustion model 310M' to the adjusted boiler heat recovery / power generation model 330M', which are used by the prediction model 900 when making predictions. The system acquires data and predicts process values ​​used to control the incineration process. The process values ​​predicted by the prediction model 900 are input to the comparison / modification unit 910. The comparison / modification unit 910 also receives process values ​​from the measurement values ​​stored in the measurement value storage unit 450 that are used to control the incineration process.

[0112] As a result, the comparison / modification unit 910 calculates the error between the process value predicted by the prediction model 900 and the measured process value stored in the measurement value storage unit 450, and updates the model parameters of the prediction model based on the calculated error. As a result, the prediction model learning unit 810 constructs a trained prediction model.

[0113] <Processing flow by the learning device during the learning phase> Next, we will explain the processing flow by the learning device 120 during the learning phase. Figure 10 is a flowchart showing the processing flow during the learning phase.

[0114] In step S1001, the learning device 120 acquires measurement values ​​from the plant control system 110.

[0115] In step S1002, the learning device 120 adjusts the parameters of the physical model using the acquired measurements and generates an adjusted physical model. The learning device 120 stores the generated adjusted physical model in the auxiliary storage device 203.

[0116] In step S1003, the learning device 120 runs a simulation by operating the adjusted physical model using the acquired measurements and calculates unobservable state variables.

[0117] In step S1004, the learning device 120 uses the calculated unobservable state variables and acquired measurements to train a predictive model and construct a trained predictive model.

[0118] In step S1005, the learning device 120 stores the constructed trained prediction model in the auxiliary storage device 203.

[0119] <System configuration during the prediction phase of the target plant> Next, the system configuration in the prediction phase of the target plant to which the detection system according to the first embodiment is applied will be described. The "detection system" is a system that detects signs of a change in the state of the target plant (indicators that a change in state is about to occur). A change in the state of the target plant includes both the target plant becoming a normal state and the target plant becoming an abnormal state, but in the first embodiment, an "abnormality detection system" that detects signs of the target plant becoming an abnormal state will be described. Figure 11 is the first diagram showing an example of the system configuration in the prediction phase of the target plant.

[0120] As shown in Figure 11, the target plant 10 to which the anomaly detection system 1110 according to the first embodiment is applied comprises target equipment 100 and a plant control system 110. Note that the target equipment 100 and plant control system 110 shown in Figure 11 are the same as the target equipment 100 and plant control system 110 shown in Figure 1, so their description is omitted here.

[0121] The anomaly detection system 1110 comprises a pattern generation device 1120 and an anomaly detection device 1130.

[0122] The pattern generation device 1120 includes a tuned physical model that reproduces the behavior of the incineration process in an incinerator, and a trained predictive model that predicts the process values ​​of the incinerator using the calculation results of the tuned physical model. The pattern generation device 1120 operates the tuned physical model and the trained predictive model when abnormal pattern generation data is input. The tuned physical model of the pattern generation device 1120 includes a tuned grate combustion model 310M', a tuned gas-phase combustion model 320M', and a tuned boiler heat recovery / power generation model 330M'. The trained predictive model of the pattern generation device 1120 includes a trained predictive model 900'.

[0123] The abnormal pattern generation data used by the pattern generation device 1120 when operating the tuned physical model and the trained prediction model is a dataset that includes, in part, abnormal prediction data (data that indicates signs of an abnormal state) based on specific abnormal causes.

[0124] A specific cause of an anomaly refers to the event that ultimately leads to an anomaly in the process value. Furthermore, anomaly prediction data is data that shows the difference between a specific measured value or state variable before and after a change that occurs as a result of a specific cause of an anomaly, and is data that has been confirmed through simulation or other means to ultimately lead to an anomaly in the process value.

[0125] As a result, the pattern generation device 1120 can generate abnormal data patterns (measured values, state variables, process values) for various abnormal causes that may occur in the incinerator. The abnormal data patterns generated by the pattern generation device 1120 are associated with specific abnormal causes and stored in the abnormal detection device 1130.

[0126] Furthermore, within the prediction phase, the phase in which the pattern generation device 1120 generates data patterns for abnormal situations is referred to as the "pattern generation phase."

[0127] The anomaly detection device 1130 acquires measurement values ​​from the plant control system 110 that are measured in the incinerator while it is in operation. Like the pattern generation device 1120, the anomaly detection device 1130 has a tuned physical model and a trained predictive model. The anomaly detection device 1130 uses the acquired measurement values ​​to operate the tuned physical model and the trained predictive model. As a result, the anomaly detection device 1130 can generate data patterns (measurement values, state variables, process values) based on the acquired measurement values.

[0128] The anomaly detection device 1130 detects signs of an anomaly by comparing the data pattern during an anomaly with the data pattern based on the acquired measurement values. When signs of an anomaly are detected, the anomaly detection device 1130 identifies the corresponding cause of the anomaly and transmits to the plant control system 110 that signs of an anomaly have been detected (detection result) and the identified cause of the anomaly.

[0129] Furthermore, within the prediction phase, the phase in which the anomaly detection device 1130 detects signs of an anomaly based on measurements taken in the operating incinerator is referred to as the "anomaly detection phase."

[0130] In other words, in this embodiment, the prediction phase includes a pattern generation phase and an anomaly detection phase.

[0131] <Hardware configuration of pattern generation device and anomaly detection device> Next, the hardware configuration of the pattern generation device 1120 and the anomaly detection device 1130 of the anomaly detection system 1110 will be described. Figure 12 shows an example of the hardware configuration of the pattern generation device and the anomaly detection device.

[0132] As shown in Figure 12, the pattern generation device 1120 and the anomaly detection device 1130 each include a processor 1201, a memory 1202, an auxiliary storage device 1203, a connection device 1204, a communication device 1205, and a drive device 1206. The processor 1201, memory 1202, auxiliary storage device 1203, connection device 1204, communication device 1205, and drive device 1206 of the pattern generation device 1120 and the anomaly detection device 1130 are interconnected via a bus 1207.

[0133] Note that the elements included in the hardware configuration of the pattern generation device 1120 and the anomaly detection device 1130 are the same as the elements included in the hardware configuration of the learning device 120 shown in Figure 2, so their explanation is omitted here.

[0134] <Functional Configuration of Pattern Generator> Next, the functional configuration of the pattern generation device 1120 in the pattern generation phase will be described. Figure 13 is the first diagram showing an example of the functional configuration of the pattern generation device in the pattern generation phase. A pattern generation program is installed in the pattern generation device 1120. When the pattern generation program is executed, the pattern generation device 1120 functions as an abnormal pattern generation data acquisition unit 1310, a gray box model unit 1320, and an abnormal data pattern generation unit 1330.

[0135] The abnormal pattern generation data acquisition unit 1310 acquires the input abnormal pattern generation data and inputs it into the gray box model unit 1320.

[0136] The gray box model unit 1320 includes a tuned grate combustion model 310M', a tuned gas-phase combustion model 320M', a tuned boiler heat recovery / power generation model 330M', and a trained prediction model 900'. The gray box model unit 1320 operates the tuned grate combustion model 310M' to the trained prediction model 900' using abnormal pattern generation data.

[0137] The abnormal data pattern generation unit 1330 acquires the cause of the abnormality that was assumed when the abnormality prediction data included in the abnormal pattern generation data was determined. The data used when operating each model (adjusted grate combustion, gas phase combustion, boiler heat recovery / power generation, and trained prediction model) in the gray box model unit 1320 (measured values, measured values ​​with anomaly prediction data attached, or state variables with anomaly prediction data attached), • By operating each model in the gray box model unit 1320, the data (state variables, process values) calculated or predicted by each model are obtained, Based on this, anomaly data patterns are generated.

[0138] The abnormal data pattern generation unit 1330 stores the generated abnormal data pattern in the abnormal data pattern storage unit 1340, associating it with the acquired cause of the abnormality.

[0139] The adjusted grate combustion model 310M' is, • Obtain the measured values ​​(or measured values ​​with anomaly prediction data attached) used when running the simulation from the data used for generating anomaly patterns. • Observable state variables from one cycle prior (or state variables with anomaly prediction data attached) used when running the simulation are obtained from the adjusted gas-phase combustion model 320M' (or data for generating anomaly patterns). Unobservable state variables are calculated by running the simulation. As a result, the tuned grate combustion model 310M' outputs the unobservable state variables that the tuned gas-phase combustion model 320M' uses when running the simulation to the tuned gas-phase combustion model 320M'. In addition, the tuned grate combustion model 310M' outputs the unobservable state variables that the trained prediction model 900' uses when making predictions to the trained prediction model 900'.

[0140] The tuned vapor phase combustion model 320M' is, • Obtain the measured values ​​(or measured values ​​with anomaly prediction data attached) used when running the simulation from the data used for generating anomaly patterns. • Observable state variables from one cycle prior (or state variables with anomaly prediction data attached) used when running the simulation are obtained from the adjusted grate combustion model 310M' (or data for anomaly pattern generation). Unobservable state variables are calculated by running the simulation. As a result, the tuned gas-phase combustion model 320M' outputs the unobservable state variables that the tuned boiler heat recovery / power generation model 330M' uses when running the simulation to the tuned boiler heat recovery / power generation model 330M'. In addition, the tuned gas-phase combustion model 320M' outputs the unobservable state variables that the trained prediction model 900' uses when making predictions to the trained prediction model 900'.

[0141] The tuned boiler heat recovery / power generation model 330M' is, • Obtain the measured values ​​(or measured values ​​with anomaly prediction data attached) used when running the simulation from the data used for generating anomaly patterns. • Observable state variables from one cycle prior (or state variables with anomaly prediction data attached) used when running the simulation are obtained from the adjusted gas-phase combustion model 320M' (or data for generating anomaly patterns). Unobservable state variables are calculated by running the simulation. As a result, the adjusted boiler heat recovery / power generation model 330M' outputs the unobservable state variables that the trained prediction model 900' uses when making predictions to the trained prediction model 900'.

[0142] The pre-trained predictive model 900' is, • Observable state variables (or state variables with anomaly prediction data attached) used for prediction are obtained from the adjusted grate combustion model 310M' to the adjusted boiler heat recovery / power generation model 330M' (or data for anomaly pattern generation). • The measured values ​​used for prediction (or measured values ​​with anomaly prediction data attached) are obtained from the data used for generating anomaly patterns. Predict process values.

[0143] <Details of the functional configuration of the gray box model section> Next, the details of the functional configuration of the gray box model section 1320 of the pattern generation device 1120 will be described. Figures 14A and 14B are the first and second figures showing examples of the functional configuration of the gray box model section. Of these, Figure 14A is an example of a functional configuration when abnormality prediction data is added to a portion of the measured values. On the other hand, Figure 14B is an example of a functional configuration when abnormality prediction data is added to a portion of the state variables. Here, the gray box model section shown in Figure 14A will be referred to as gray box model section 1320A, and the gray box model section shown in Figure 14B will be referred to as gray box model section 1320B.

[0144] (1) When adding abnormality prediction data to a portion of the measured values As shown in Figure 14A, the gray box model section 1320A receives data for generating abnormal patterns. The data for generating abnormal patterns includes: • Measurement values ​​(time-series data) taken during operation of a normally functioning incinerator, Anomaly prediction data showing the difference between before and after a specific measurement value changes due to the occurrence of a specific anomaly, This includes the following. Furthermore, if the measurement values ​​of multiple items change due to the occurrence of a specific abnormal cause, abnormal prediction data is generated for each item. In addition, the size and frequency of occurrence of the abnormal prediction data vary, and arbitrary abnormal prediction data is generated within the expected range. Moreover, there are various causes of abnormalities, and abnormal prediction data is generated according to each cause of abnormality. As a result, the gray box model unit 1320A can be input with a variety of abnormal prediction data, including abnormal prediction data that is unlikely to occur in an incinerator in operation.

[0145] The "specific measurement values ​​that change in response to the occurrence of an abnormal cause" referred to here are basically measurement values ​​derived through simulation when an abnormality is assumed to occur inside the incinerator. However, the "specific measurement values ​​that change in response to the occurrence of an abnormal cause" may also be measurement values ​​measured when an abnormality actually occurs inside the incinerator.

[0146] Of the abnormal pattern generation data input to the gray box model unit 1320A, the measured values ​​are: • Adjusted grate combustion model 310M' • Adjusted vapor phase combustion model 320M' • Adjusted boiler heat recovery / power generation model 330M' • Pre-trained predictive model 900', It will be entered into.

[0147] Furthermore, among the abnormal pattern generation data input to the gray box model unit 1320A, the abnormal prediction data is: • Measurement values ​​entered into the adjusted grate combustion model 310M', • Measurement values ​​entered into the adjusted gas-phase combustion model 320M', • Measurement values ​​entered into the adjusted boiler heat gain / power generation model 330M', • Measurement values ​​input to the trained predictive model 900', It is added to one or more of the following measurements.

[0148] The abnormality prediction data added to the measured values ​​input to the adjusted grate combustion model 310M' includes, for example, abnormality prediction data (positive value) for the amount of waste, which is added to the measured value of the amount of waste measured in a normally functioning incinerator. This abnormality prediction data for the amount of waste is, for example, abnormality prediction data where the abnormal cause is a malfunction in the mechanism for loading waste into the incinerator.

[0149] The abnormality prediction data added to the measured values ​​input to the adjusted gas-phase combustion model 320M' includes, for example, abnormality prediction data (negative value) for air volume, which is added to the measured value of air volume measured in a normal incinerator. The abnormality prediction data for air volume is, for example, abnormality prediction data where the cause is an abnormality in the air supply mechanism, etc.

[0150] The abnormality prediction data added to the measured values ​​input into the adjusted boiler heat recovery / power generation model 330M' includes, for example, abnormality prediction data (negative value) for exhaust gas flow rate, which is added to the measured value of exhaust gas flow rate measured in a normal incinerator. The abnormality prediction data for exhaust gas flow rate is, for example, abnormality prediction data that indicates a combustion failure or the like as the cause of the abnormality.

[0151] Anomaly prediction data added to the measured values ​​input to the trained prediction model 900' includes, for example, anomaly prediction data for waste composition (positive or negative values ​​for each component) added to measured values ​​of waste composition measured in a normal incinerator. Anomaly prediction data for waste composition is, for example, anomaly prediction data where the cause of the anomaly is foreign matter contamination during waste collection.

[0152] As shown in Figure 14A, the adjusted grate combustion model 310M' is: • Measurement values ​​used when running the simulation, or measurement values ​​with anomaly prediction data added to those measurements, and • Unobservable state variables calculated by the adjusted gas-phase combustion model 320M' one cycle prior, The data is obtained, and unobservable state variables are calculated. The unobservable state variables calculated by the tuned grate combustion model 310M' are input into the tuned gas-phase combustion model 320M' and also into the trained prediction model 900'.

[0153] Furthermore, as shown in Figure 14A, the adjusted gas-phase combustion model 320M' is, • Measurement values ​​used when running the simulation, or measurement values ​​with anomaly prediction data added to those measurements, and • Unobservable state variables calculated by the adjusted grate combustion model 310M' one cycle prior, The data is obtained, and unobservable state variables are calculated. The unobservable state variables calculated by the tuned gas-phase combustion model 320M' are input into the tuned boiler heat recovery / power generation model 330M' and also into the trained prediction model 900'.

[0154] Furthermore, as shown in Figure 14A, the adjusted boiler heat acquisition / power generation model 330M' is, • Measurement values ​​used when running the simulation, or measurement values ​​with anomaly prediction data added to those measurements, and • Unobservable state variables calculated by the adjusted gas-phase combustion model 320M' one cycle prior, The data is obtained, and unobservable state variables are calculated. The unobservable state variables calculated by the adjusted boiler heat gain / power generation model 330M' are input into the trained predictive model 900'.

[0155] Furthermore, as shown in Figure 14A, the trained prediction model 900' is, • Unobservable state variables calculated by the adjusted grate combustion model 310M', • Unobservable state variables calculated by the adjusted gas-phase combustion model 320M' • Unobservable state variables calculated by the adjusted boiler heat gain / power generation model 330M' • Measurement values ​​used when making predictions, or measurement values ​​after anomaly prediction data has been added to those measurement values. Obtain the data and predict the process value.

[0156] The measured values ​​or added measured values ​​input to each model in the gray box model unit 1320A, the state variables and process values ​​calculated or predicted by each model, are stored in the abnormal data pattern storage unit 1340 as abnormal data patterns. The abnormal data patterns stored in the abnormal data pattern storage unit 1340 are, for example, ·Measurements input into the adjusted grate combustion model 310M', or the measured values ​​after anomaly prediction data has been added to the said measurements, ·Measurements input into the adjusted gas-phase combustion model 320M', or the measured values ​​after abnormality prediction data has been added to the said measurements, • The measured values ​​input into the adjusted boiler heat recovery / power generation model 330M', or the measured values ​​after anomaly prediction data has been added to those measured values, The measured values ​​input to the trained predictive model 900', or the measured values ​​after anomaly prediction data has been added to the measured values, • Unobservable state variables calculated by the adjusted grate combustion model 310M', or state variables to which anomaly prediction data has been added, • Unobservable state variables calculated by the adjusted gas-phase combustion model 320M', or state variables to which abnormality prediction data has been added, • Unobservable state variables calculated by the adjusted boiler heat recovery / power generation model 330M', or state variables to which anomaly prediction data has been added, • Process values ​​predicted by the trained predictive model 900', Includes.

[0157] (2) When abnormality prediction data is added to some of the state variables As shown in Figure 14B, the gray box model section 1320B receives data for generating abnormal patterns. The data for generating abnormal patterns includes: • Measurement values ​​(time-series data) taken during operation of a normally functioning incinerator, Anomaly prediction data showing the difference between before and after a specific state variable changes as a result of a specific anomaly occurring, This includes the following. Furthermore, if multiple state variables change as a result of a specific abnormal cause occurring, abnormal prediction data is generated for each item. The size and frequency of occurrence of the abnormal prediction data vary, and arbitrary abnormal prediction data is generated within the expected range. In addition, there are various causes of abnormalities, and abnormal prediction data is generated according to each cause of abnormality. As a result, the gray box model unit 1320B can be input with a variety of abnormal prediction data, including abnormal prediction data that is unlikely to occur in an incinerator in operation.

[0158] Of the abnormal pattern generation data input to the gray box model unit 1320B, the measured values ​​are: • Adjusted grate combustion model 310M' • Adjusted vapor phase combustion model 320M' • Adjusted boiler heat recovery / power generation model 330M' • Pre-trained predictive model 900', It will be entered into.

[0159] Furthermore, among the abnormal pattern generation data input to the gray box model unit 1320B, the abnormal prediction data is: • Unobservable state variables calculated by the adjusted grate combustion model 310M', • Unobservable state variables calculated by the adjusted gas-phase combustion model 320M', • Unobservable state variables calculated by the adjusted boiler heat gain / power generation model 330M', It is attached to one or more of the following state variables.

[0160] Anomaly prediction data added to unobservable state variables calculated by the adjusted grate combustion model 310M' includes, for example, anomaly prediction data (negative values) added to the amount of combustible gas generated, which is calculated based on measurements taken in a normal incinerator. The anomaly prediction data for the amount of combustible gas generated is, for example, anomaly prediction data where the cause is poor combustion of waste in the incinerator.

[0161] Anomaly prediction data added to unobservable state variables calculated by the adjusted gas-phase combustion model 320M' includes, for example, anomaly prediction data added to the post-combustion gas composition calculated based on measurements taken in a normal incinerator. Anomaly prediction data for post-combustion gas composition can be positive or negative depending on the component. Anomaly prediction data for post-combustion gas composition is, for example, anomaly prediction data where the cause is poor combustion of combustible gases in the incinerator.

[0162] Anomaly prediction data added to unobservable state variables calculated by the adjusted boiler heat recovery / power generation model 330M' includes, for example, anomaly prediction data (negative values) added to the heat recovery efficiency calculated based on measurements taken in a normal incinerator. Anomaly prediction data for heat recovery efficiency is, for example, anomaly prediction data where the cause is a failure of a specific component in the incinerator boiler.

[0163] As shown in Figure 14B, the adjusted grate combustion model 310M' is: • Measurement values ​​used when running the simulation, and • Unobservable state variables calculated by the adjusted gas-phase combustion model 320M' one cycle prior, or state variables with anomaly prediction data added to the said unobservable state variables. The data is obtained, and unobservable state variables are calculated. The unobservable state variables calculated by the adjusted grate combustion model 310M', or the state variables to which anomaly prediction data has been added, are input into the adjusted gas-phase combustion model 320M' and also into the trained prediction model 900'.

[0164] Furthermore, as shown in Figure 14B, the adjusted gas-phase combustion model 320M' is, • Measurement values ​​used when running the simulation, and • Unobservable state variables calculated by the adjusted grate combustion model 310M' one cycle prior, or state variables with anomaly prediction data added to the said unobservable state variables. The data is obtained, and unobservable state variables are calculated. The unobservable state variables calculated by the adjusted gas-phase combustion model 320M', or the state variables to which anomaly prediction data has been added, are input into the adjusted boiler heat recovery / power generation model 330M' and also into the trained prediction model 900'.

[0165] Furthermore, as shown in Figure 14B, the adjusted boiler heat acquisition / power generation model 330M' is, • Measurement values ​​used when running the simulation, and • Unobservable state variables calculated by the adjusted gas-phase combustion model 320M' one cycle prior, or state variables with anomaly prediction data added to the said unobservable state variables. The data is obtained, and unobservable state variables are calculated. The unobservable state variables calculated by the adjusted boiler heat recovery / power generation model 330M', or the state variables to which anomaly prediction data has been added, are input into the trained prediction model 900'.

[0166] Furthermore, as shown in Figure 14B, the trained prediction model 900' is, • Unobservable state variables calculated by the adjusted grate combustion model 310M', or state variables to which anomaly prediction data has been added. • Unobservable state variables calculated by the adjusted gas-phase combustion model 320M', or state variables to which abnormality prediction data has been added. • Unobservable state variables calculated by the adjusted boiler heat gain / power generation model 330M', or state variables to which anomaly prediction data has been added. • Measurement values ​​used when making predictions, Obtain the data and predict the process value.

[0167] The measured values ​​input to each model in the gray box model unit 1320B, the state variables calculated or predicted by each model or the added state variables, and the process values ​​are stored in the abnormal data pattern storage unit 1340 as abnormal data patterns.

[0168] <Specific examples of data patterns during abnormal situations> Next, we will explain specific examples of data patterns during abnormal conditions. Figures 15 and 16 are the first and second figures showing specific examples of data patterns during abnormal conditions. Of these, Figure 15 shows a specific example of a data pattern during abnormal conditions generated by adding abnormality prediction data to some of the measured values, and Figure 16 shows a specific example of a data pattern during abnormal conditions generated by adding abnormality prediction data to some of the state variables.

[0169] (1) Anomaly data patterns generated by adding anomaly prediction data to a portion of the measured values In Figure 15, reference numeral 1501 indicates the measured value after abnormality prediction data has been added, which is input into the adjusted gas-phase combustion model 320M'. Reference numeral 1501T illustrates the time-series data. In the time-series data shown by reference numeral 1501T, the position indicated by the vertical arrow represents the change in the measured value input into the adjusted gas-phase combustion model 320M' due to the occurrence of abnormality cause X at the timing indicated by the dashed line. It is assumed that the incinerator operator has empirically learned how the measured value input into the adjusted gas-phase combustion model 320M' changes in response to the occurrence of abnormality cause X. In this embodiment, the time-series data shown by reference numeral 1501T is generated, for example, by the operator of the pattern generation device 1120 inputting abnormality pattern generation data that reflects the experience of the incinerator operator.

[0170] In Figure 15, labels 1502 to 1504 represent unobservable state variables calculated by the adjusted grate combustion model 310M' to the adjusted boiler heat gain / power generation model 330M', respectively. Labels 1502T to 1504T illustrate the respective time-series data.

[0171] In Figure 15, the label 1505 represents the process value predicted by the trained prediction model 900'. The label 1505T illustrates the time series data.

[0172] In the example of data patterns during an anomaly shown in Figure 15, the process value does not change immediately even when anomaly cause X occurs (see reference numeral 1505T). Also, the change in the measured value after the occurrence of anomaly cause X, shown by reference numeral 1501T, is small, and the state variables shown by reference numerals 1502T and 1504T do not change. On the other hand, the state variable shown by reference numeral 1503T changes significantly as a result of the occurrence of anomaly cause X (see dotted circle).

[0173] Therefore, in the case of the abnormal data pattern shown in Figure 15, an early sign of an anomaly can be detected at the timing when a change similar to the change shown by symbol 1503T is detected in the state variable shown by symbol 1503 (i.e., before an anomaly appears in the process value). Furthermore, according to the abnormal data pattern shown in Figure 15, the cause of the anomaly X can be identified when an early sign of an anomaly is detected.

[0174] (2) Anomaly data patterns generated by adding anomaly prediction data to some of the state variables In Figure 16, reference numeral 1601 indicates the state variable after anomaly prediction data has been added to the unobservable state variable calculated by the adjusted grate combustion model 310M'. Reference numeral 1601T illustrates the time-series data. In the time-series data shown by reference numeral 1601T, the position indicated by the vertical arrow represents the change in the state variable calculated by the adjusted grate combustion model 310M' due to the occurrence of anomaly cause Y at the timing indicated by the dashed line. It is assumed that the incinerator operator has empirically learned how the state variable calculated by the adjusted grate combustion model 310M' changes in response to the occurrence of anomaly cause Y. In this embodiment, the time-series data shown by reference numeral 1601T is generated, for example, by the operator of the pattern generation device 1120 inputting anomaly pattern generation data that reflects the experience of the incinerator operator.

[0175] In Figure 16, the reference numeral 1602 indicates the measured value input to the adjusted gas-phase combustion model 320M'. The reference numeral 1602T illustrates the time-series data.

[0176] In Figure 16, labels 1603 and 1604 represent unobservable state variables calculated using the adjusted grate combustion model 310M' and the adjusted boiler heat gain / power generation model 330M', respectively. Labels 1603T to 1604T illustrate the respective time-series data.

[0177] In Figure 16, code 1605 indicates the process value predicted by the trained prediction model 900'. Code 1605T illustrates the time series data.

[0178] In the example of data patterns during an anomaly shown in Figure 16, the process value changes after a predetermined time period following the occurrence of anomaly cause Y (see reference numeral 1605T). On the other hand, among reference numerals 1601T to 1604T, only reference numeral 1601T changes due to the occurrence of anomaly cause Y. Limiting the analysis to the period before an anomaly appears in the process value, the state variable changes significantly only twice in reference numeral 1601T due to the occurrence of anomaly cause Y.

[0179] Therefore, in the case of the abnormal data pattern shown in Figure 16, a change similar to a single change in the state variable indicated by symbol 1601T can be detected, allowing for the detection of an anomaly precursor before an anomaly appears in the process value. Furthermore, according to the abnormal data pattern shown in Figure 16, the cause of the anomaly Y can be identified when an anomaly precursor is detected.

[0180] <Flow of pattern generation process by pattern generation device> Next, we will explain the flow of the pattern generation process by the pattern generation device 1120. Figure 17 is a first flowchart showing the flow of the pattern generation process by the pattern generation device.

[0181] In step S1701, the operator of the pattern generation device 1120 determines the cause of the abnormality.

[0182] In step S1702, the operator of the pattern generation device 1120 identifies the measured values ​​or state variables that change as a result of the occurrence of the determined cause of the abnormality, and generates abnormality prediction data.

[0183] In step S1703, the operator of the pattern generation device 1120 generates abnormal pattern generation data, which includes measurement values ​​taken while the incinerator is operating in a normal state and abnormality prediction data.

[0184] In step S1704, the pattern generation device 1120 uses the abnormal pattern generation data to operate each model in the gray box model unit 1320.

[0185] In step S1705, the pattern generation device 1120 obtains state variables and process values ​​calculated or predicted by operating each model in the gray box model unit 1320.

[0186] In step S1706, the pattern generation device 1120 generates an abnormal data pattern, associates it with the cause of the abnormality, and stores it in the abnormal data pattern storage unit 1340. The pattern generation device 1120 associates the abnormal data pattern with the cause of the abnormality, for example, by including time information of when the abnormality occurred in the cause of the abnormality so that it is linked to the time series data contained in the abnormal data pattern.

[0187] In step S1707, the pattern generation device 1120 determines whether or not to terminate the pattern generation process. If it determines in step S1707 to continue the pattern generation process (if the answer is NO in step S1707), it returns to step S1701. On the other hand, if it determines in step S1707 to terminate the pattern generation process (if the answer is YES in step S1707), it terminates the pattern generation process.

[0188] <Functional Configuration of Anomaly Detection Device> Next, the functional configuration of the anomaly detection device 1130 in the anomaly detection phase will be described. Figure 18 shows an example of the functional configuration of the anomaly detection device in the anomaly detection phase. An anomaly detection program is installed in the anomaly detection device 1130. When this program is executed, the anomaly detection device 1130 functions as a measurement value acquisition unit 1810, a gray box model unit 1320, a data pattern generation unit 1820, and a detection unit 1830.

[0189] The measurement value acquisition unit 1810 acquires measurement values ​​measured in the incinerator while it is in operation from the plant control system 110. The measurement values ​​acquired by the measurement value acquisition unit 1810 are input to the gray box model unit 1320 as a dataset for operating each model in the gray box model unit 1320.

[0190] The gray-box model unit 1320 includes a tuned grate combustion model 310M', a tuned gas-phase combustion model 320M', a tuned boiler heat recovery / power generation model 330M', and a trained prediction model 900'. The gray-box model unit 1320 operates the tuned grate combustion model 310M' to the trained prediction model 900' using the input dataset.

[0191] The operation of the adjusted grate combustion model 310M' and the trained prediction model 900' in the gray box model section 1320 has already been explained using Figure 13, so the explanation will be omitted here.

[0192] The data pattern generation unit 1820 is: • Data (measured values) used when operating each model in the gray box model unit 1320, • By operating each model in the gray box model unit 1320, the data calculated or predicted by each model (state variables, process values) and The data is acquired and a data pattern is generated. The data pattern generation unit 1820 notifies the detection unit 1830 of the generated data pattern.

[0193] The detection unit 1830 refers to the abnormal data patterns stored in the abnormal data pattern storage unit 1340 and compares them with the data patterns generated by the data pattern generation unit 1820. Based on this, the detection unit 1830 determines whether there are any abnormal data patterns similar to those notified by the data pattern generation unit 1820. If the detection unit 1830 detects similar abnormal data patterns, it determines that it has detected an early sign of an abnormality and notifies the plant control system 110 of the detection result. The detection unit 1830 also notifies the plant control system 110 of the cause of the abnormality, which is stored in association with the similar abnormal data patterns.

[0194] This allows operators of incinerators in operation to recognize when signs of an abnormality have been detected in the incinerator (detection result) and to understand the cause of the abnormality. As a result, operators of incinerators in operation can take measures such as performing appropriate manual or automatic operations to prevent the occurrence of abnormalities.

[0195] <Flowchart of abnormality detection processing by an abnormality detection device> Next, we will explain the flow of the anomaly detection process by the anomaly detection device 1130. Figure 19 is a flowchart showing the flow of the anomaly detection process by the anomaly detection device.

[0196] In step S1901, the abnormality detection device 1130 obtains measurement values ​​from the plant control system 110 that were measured in the incinerator while it was in operation.

[0197] In step S1902, the anomaly detection device 1130 uses the acquired measurement values ​​as a dataset and generates data patterns by operating each model in the gray box model unit 1320.

[0198] In step S1903, the abnormality detection device 1130 compares the generated data pattern with the data pattern during an abnormality, and determines whether they are similar. If it is determined in step S1903 that they are not similar (if NO in step S1903), the process proceeds to step S1907. On the other hand, if it is determined in step S1903 that they are similar (if YES in step S1903), the process proceeds to step S1904. Note that the abnormality detection device 1130 determines whether the generated data pattern is similar to the data pattern during an abnormality by calculating the similarity using, for example, a known correlation function or clustering.

[0199] In step S1904, the abnormality detection device 1130 determines that an abnormality sign has been detected.

[0200] In step S1905, the abnormality detection device 1130 specifies the cause of the abnormality stored in association with the data pattern during an abnormality that was determined to be similar.

[0201] In step S1906, the abnormality detection device 1130 transmits the fact that an abnormality sign has been detected (detection result) and the cause of the abnormality to the plant control system 110.

[0202] In step S1907, the abnormality detection device 1130 determines whether to continue the abnormality detection process. If it is determined to continue (if YES in step S1907), the process returns to step S1901. On the other hand, if it is determined not to continue (if NO in step S1907), the abnormality detection process ends.

[0203] <Summary> As is clear from the above description, the abnormality detection system 1110 according to the first embodiment includes a pattern generation device 1120 that operates in the pattern generation phase and an abnormality detection device 1130 that operates in the abnormality detection phase.

[0204] The pattern generation device 1120 A pre-tuned physical model that reproduces the behavior of the incineration process in an incinerator, and a trained predictive model that predicts the process values ​​of the incinerator using the calculation results of the pre-tuned physical model, are operated using a dataset that includes some anomaly prediction data. This generates a data pattern for anomalies in the incinerator (an example of the first data pattern).

[0205] The anomaly detection device 1130 is • Using measurements taken in an operating incinerator, a data pattern (an example of a second data pattern) is generated by running a tuned physical model and a trained predictive model. • By comparing the generated data patterns with data patterns from abnormal situations, the system detects early signs of anomalies.

[0206] Thus, the anomaly detection system 1110 according to the first embodiment generates an anomaly data pattern using a dataset that includes various anomaly prediction data, including anomaly prediction data that is unlikely to occur in an actual incinerator. The anomaly detection system 1110 according to the first embodiment then uses this anomaly data pattern to detect an anomaly.

[0207] As a result, the anomaly detection system 1110 according to the first embodiment makes it possible to detect signs of anomalies in the plant with high accuracy.

[0208] [Second Embodiment] In the first embodiment described above, the cause of the anomaly is determined, and a data pattern for the anomaly is generated using a dataset that includes some anomaly prediction data based on the cause of the anomaly. However, the configuration for generating a data pattern for the anomaly is not limited to this. For example, a configuration may be used to generate a data pattern for the anomaly by determining a process value to which anomaly data (data indicating that an abnormal state has occurred) is added, and then searching for a state variable and measured value that predicts the process value after the addition. The second embodiment will be described below, focusing on the differences from the first embodiment.

[0209] <System configuration during the prediction phase of the target plant> First, we will describe the system configuration in the prediction phase of the target plant to which the anomaly detection system according to the second embodiment is applied. Figure 20 is a second diagram showing an example of the system configuration in the prediction phase of the target plant.

[0210] As shown in Figure 20, the target plant 10 to which the anomaly detection system 2010 according to the second embodiment is applied comprises target equipment 100 and a plant control system 110. Note that the target equipment 100 and plant control system 110 shown in Figure 20 are the same as the target equipment 100 and plant control system 110 shown in Figure 1, so their description is omitted here.

[0211] The anomaly detection system 2010 comprises a pattern generation device 2020 and an anomaly detection device 1130. Note that the anomaly detection device 1130 shown in Figure 20 is the same as the anomaly detection device 1130 shown in Figure 11, and therefore its explanation is omitted here.

[0212] The pattern generation device 2020 includes a tuned physical model that reproduces the behavior of the incineration process in an incinerator, and a trained predictive model that calculates the process values ​​of the incinerator using the calculation results of the tuned physical model. The pattern generation device 1120 operates the tuned physical model and the trained predictive model to search for state variables and measured values ​​that predict the process values ​​after abnormal data has been added. The pattern generation device 2020 generates an abnormal data pattern using the process values ​​after abnormal data has been added, and the searched state variables and measured values. The abnormal data is data that shows the difference between the process values ​​under normal conditions and the process values ​​under abnormal conditions.

[0213] Similar to the first embodiment described above, the pre-tuned physical models of the pattern generation device 1120 include a pre-tuned grate combustion model 310M', a pre-tuned gas-phase combustion model 320M', and a pre-tuned boiler heat recovery / power generation model 330M'. In addition, the trained predictive models of the pattern generation device 1120 include a trained predictive model 900'.

[0214] Furthermore, similar to the first embodiment described above, the prediction phase includes a pattern generation phase and an anomaly detection phase.

[0215] <Functional Configuration of Pattern Generator> Next, the functional configuration of the pattern generation device 2020 in the pattern generation phase will be described. Figure 21 is a second diagram showing an example of the functional configuration of the pattern generation device in the pattern generation phase. The pattern generation device 2020 has a pattern generation program installed, and when this program is executed, the pattern generation device 2020 functions as a gray box model unit 2110 and an abnormal data pattern generation unit 2120.

[0216] The gray box model section 2110 includes a tuned grate combustion model 310M', a tuned gas-phase combustion model 320M', a tuned boiler heat recovery / power generation model 330M', and a trained prediction model 900'.

[0217] The gray-box model unit 2110 obtains candidate input values ​​(candidate values ​​for state variables and measured values) from the abnormal data pattern generation unit 2120. The gray-box model unit 2110 operates the model to which the candidate input values ​​are input and related models from among the adjusted grate combustion model 310M', adjusted gas-phase combustion model 320M', adjusted boiler heat recovery / power generation model 330M', and trained prediction model 900'. By operating the model and related models using the candidate input values, the gray-box model unit 2110 outputs the unobservable state variables or process values ​​calculated or predicted by the model and related models to the abnormal data pattern generation unit 2120.

[0218] For example, when the state variables and measurement values used when prediction is performed by the learned prediction model 900' are acquired as input value candidates, the gray box model unit 2110 operates the learned prediction model 900' using the acquired state variables and measurement values. Further, the gray box model unit 2110 outputs the process value predicted by the learned prediction model 900' to the abnormal data pattern generation unit 2120.

[0219] For example, assume that the state variables and measurement values used when simulation is executed by the adjusted boiler heat generation / power generation model 330M' are acquired as input value candidates. In this case, the gray box model unit 2110 operates the adjusted boiler heat generation / power generation model 330M' using the acquired state variables and measurement values. The gray box model unit 2110 outputs the unobservable state variables calculated by the adjusted boiler heat generation / power generation model 330M' to the abnormal data pattern generation unit 2120.

[0220] The abnormal data pattern generation unit 2120 includes a first input value search unit 2130_1 and a second input value search unit 2130_2, and a first abnormal pattern generation unit 2140_1 and a second abnormal pattern generation unit 2140_2.

[0221] The first input value search unit 2130_1 receives the process value after addition to which abnormal data has been added, which is determined by the operator of the pattern generation device 2020. The operator of the pattern generation device 2020 determines in advance the process value and the abnormal data to be added to the process value, and determines the process value after addition by adding the determined abnormal data to the determined process value.

[0222] The first input value search unit 2130_1 receives the input value candidate (state variables and measurement values used when prediction is performed by the learned prediction model 900') input by the operator of the pattern generation device 2020, and notifies the gray box model unit 2110.

[0223] Furthermore, the first input value search unit 2130_1 determines whether the process value with the abnormal data attached matches the process value notified by the gray box model unit 2110 in response to the notification of the input value candidate. If they do not match, the first input value search unit 2130_1 changes the input value candidate and notifies the gray box model unit 2110 of the changed input value candidate. The first input value search unit 2130_1 searches for an input value candidate (state variable and measured value) corresponding to the process value with the abnormal data attached by repeating the process until they match.

[0224] Furthermore, the first input value search unit 2130_1 determines whether the searched input value candidate matches the state variable and measured value that are expected when a specific abnormal cause occurs. If the first input value search unit 2130_1 determines that they match, • Process values ​​after abnormal data has been added and searched candidate input values ​​(state variables and measured values), • Specific causes of abnormalities, The first abnormal pattern generation unit 2140_1 is notified of this. Meanwhile, if the first input value search unit 2130_1 determines that there is no match, it notifies the second input value search unit 2130_2 of the process value to which the abnormal data has been added and the searched candidate input value.

[0225] The first abnormal pattern generation unit 2140_1 generates an abnormal data pattern, associates it with a specific abnormal cause, and stores it in the abnormal data pattern storage unit 1340. The first abnormal pattern generation unit 2140_1 is... • Process value after abnormal data has been added. • Searched candidate input values ​​(state variables and measured values), Generates abnormal data patterns that include this.

[0226] The second input value search unit 2130_2 receives the added process values ​​and searched input value candidates notified by the first input value search unit 2130_1. The second input value search unit 2130_2 receives input value candidates (measured values ​​used when the simulation is performed by the adjusted grate combustion model 310M' to the adjusted boiler heat recovery / power generation model 330M') input by the operator of the pattern generation device 2020. The second input value search unit 2130_2 notifies the gray box model unit 2110 of the received input value candidates.

[0227] As a result, the gray-box model unit 2110 calculates unobservable state variables by having the adjusted grate combustion model 310M' to the adjusted boiler heat recovery / power generation model 330M' run simulations using the notified input value candidates. The calculated unobservable state variables are then notified to the second input value search unit 2130 by the gray-box model unit 2110.

[0228] The second input value search unit 2130_2 is, • The state variables included in the searched candidate input values, as notified by the first input value search unit 2130_1, • The state variables notified by the gray box model unit 2110 in response to the notification of candidate input values, The second input value search unit 2130_2 determines whether they match or not. If they do not match, the second input value search unit 2130_2 changes the input value candidate (measured value) and notifies the gray box model unit 2110 of the changed input value candidate (measured value). The second input value search unit 2130_2 searches for an input value candidate (measured value) corresponding to the searched input value candidate (state variable) by repeating the process until they match.

[0229] Furthermore, the second input value search unit 2130_2 determines a specific cause of an anomaly corresponding to the searched input value candidate (measured value). The second input value search unit 2130_2, ·The process value after the addition of abnormal data, notified by the first input value search unit 2130_1, and the searched input value candidates (i.e., the state variables and measured values ​​used when prediction was made by the trained prediction model 900'), • The second input value search unit 2130_2 searches for candidate input values ​​(i.e., the measured values ​​used when the simulation is performed by the adjusted grate combustion model 310M' to the adjusted boiler heat recovery / power generation model 330M'), • Specific causes of abnormalities, This is notified to the second abnormal pattern generation unit 2140_2.

[0230] The second abnormal pattern generation unit 2140_2 generates an abnormal data pattern, associates it with a specific abnormal cause, and stores it in the abnormal data pattern storage unit 1340. The second abnormal pattern generation unit 2140_2 is... • Process value after abnormal data has been added. • Searched input value candidates (i.e., state variables and measurements used when predictions were made by the trained prediction model 900'), • Searched candidate input values ​​(i.e., measured values ​​used when the simulation was performed by the tuned grate combustion model 310M' to the tuned boiler heat gain / power generation model 330M'), Generates abnormal data patterns that include this.

[0231] <Overview of Pattern Generation Process> Next, we will explain the overview of the pattern generation process using the pattern generation device 2020. Figure 22 is a diagram illustrating the overview of the pattern generation process using the pattern generation device.

[0232] In Figure 22, arrow 2201 indicates an overview of the first pattern generation process performed by the pattern generation device 2020.

[0233] As shown by arrow 2201, the added process value with abnormal data, determined by the operator of the pattern generation device 2020, is input to the first input value search unit 2130_1. The first input value search unit 2130_1 searches for state variables and measured values ​​that can operate the trained prediction model 900' so that the added process value with abnormal data is predicted.

[0234] The pattern generation process shown by arrow 2201 indicates the case where the searched state variables and measured values ​​match the state variables and measured values ​​expected when abnormality cause X occurs. In this case, the first input value search unit 2130_1 stores abnormality cause X in the abnormality data pattern storage unit 1340. Also, the first abnormality pattern generation unit 2140_1, • Explored state variables and measured values, • Process value after abnormal data has been added, • Measurement values ​​taken in a normal incinerator (measurements used when simulations are performed using the adjusted grate combustion model 310M' to the adjusted boiler heat recovery / power generation model 330M'), An abnormal data pattern 2210_1 containing the abnormality is generated, associated with the abnormality cause X, and stored in the abnormality data pattern storage unit 1340.

[0235] The pattern generation process shown by arrow 2201 is performed on process values ​​to which various abnormal data have been added.

[0236] In Figure 22, arrow 2202 indicates an overview of the second pattern generation process performed by the pattern generation device 2020.

[0237] As shown by arrow 2202, the added process value with abnormal data, determined by the operator of the pattern generation device 2020, is input to the first input value search unit 2130_1. The first input value search unit 2130_1 searches for state variables and measured values ​​that can operate the trained prediction model 900' so that the added process value with abnormal data is predicted.

[0238] The pattern generation process shown by arrow 2202 indicates a case where the searched state variables and measured values ​​do not match the state variables and measured values ​​expected when abnormal cause X occurs. In this case, the second input value search unit 2130_2 searches for measured values ​​that enable the adjusted grate combustion model 310M' to the adjusted boiler heat recovery / power generation model 330M' to operate in such a way that the searched state variables are calculated.

[0239] The pattern generation process shown by arrow 2202 indicates the case where the searched measurement value matches the measurement value expected when an anomaly cause Y occurs. In this case, the second input value search unit 2130_2 stores the anomaly cause Y in the anomaly data pattern storage unit 1340. Also, the second anomaly pattern generation unit 2140_2, • The state variables and measured values ​​searched by the first input value search unit 2130_1, ·Measured values ​​searched by the second input value search unit 2130_2, • Process value after abnormal data has been added, An abnormal data pattern 2220_1 including the abnormality is generated, associated with the abnormality cause Y, and stored in the abnormal data pattern storage unit 1340.

[0240] The pattern generation process shown by arrow 2202 is performed on process values ​​to which various abnormal data have been added.

[0241] <Specific examples of data patterns during abnormal situations> Next, we will explain specific examples of data patterns during abnormal conditions. Figure 23 is the third figure showing a specific example of a data pattern during abnormal conditions.

[0242] In Figure 23, reference numeral 2301 indicates the process value after abnormal data has been added. Reference numeral 2301T illustrates the time-series data. In the time-series data shown by reference numeral 2301T, the position indicated by the vertical arrow shows that abnormal data was added to the process value measured in a normal incinerator.

[0243] In Figure 23, labels 2302 to 2304 represent state variables that enable the trained prediction model 900' to operate in such a way that it predicts process values ​​with abnormal data attached. Labels 2302T to 2304T illustrate the respective time series data.

[0244] In Figure 23, reference numeral 2305 indicates a measured value that allows the adjusted gas-phase combustion model 320M' to operate so that the unobservable state variable shown in reference numeral 2303 can be calculated. Reference numeral 2305T illustrates the time-series data. Furthermore, the time-series data shown in reference numeral 2305T indicates that it was determined to match the measured value (time-series data) that would be expected if an abnormal cause Y occurred at the timing indicated by the dashed line. In other words, reference numeral 2305T corresponds to the abnormal prediction data described in the first embodiment above.

[0245] In the example of the data pattern during an anomaly shown in Figure 23, the process value does not change significantly immediately even when anomaly cause Y occurs (see reference numeral 2301T). Furthermore, the change in the measured value after the occurrence of anomaly cause Y, shown in reference numeral 2305T, is small, and the state variables shown in reference numerals 2302T and 2304T also remain unchanged. On the other hand, the state variable shown in reference numeral 2303T changes significantly as a result of the occurrence of anomaly cause Y (see dotted circle).

[0246] Therefore, in the case of the abnormal data pattern shown in Figure 23, an early sign of an anomaly can be detected at the timing when a change similar to the change shown in code 2303T is detected in the state variable shown in code 2303 (i.e., before an anomaly appears in the process value). Furthermore, according to the abnormal data pattern shown in Figure 23, the cause of the anomaly Y can be identified when an early sign of an anomaly is detected.

[0247] <Flow of pattern generation process by pattern generation device> Next, we will explain the flow of the pattern generation process using the pattern generation device 2020. Figure 24 is a second flowchart showing the flow of the pattern generation process using the pattern generation device.

[0248] In step S2401, the operator of the pattern generation device 2020 determines the abnormal data.

[0249] In step S2402, the operator of the pattern generation device 2020 determines the process value to which abnormal data will be added, and generates the process value after the abnormal data is added.

[0250] In step S2403, the pattern generation device 2020 operates each model in the gray box model unit 1320 using the candidate input values ​​(state variables and measured values). This allows the pattern generation device 2020 to search for candidate input values ​​(state variables and measured values) that enable each model to operate so that the predicted process value matches the process value after abnormal data has been added.

[0251] In step S2404, the pattern generation device 2020 determines whether or not the cause of the abnormality has been identified based on the searched input value candidates (state variables and measured values). If it is determined in step S2404 that the cause of the abnormality has not been identified (i.e., the answer is NO in step S2404), the process proceeds to step S2405.

[0252] In step S2405, the pattern generation device 2020 operates each model in the gray box model unit 1320 using the candidate input values ​​(measured values). This allows the pattern generation device 2020 to search for candidate input values ​​(measured values) that enable each model to operate so that the calculated state variables match the state variables explored in step S2403.

[0253] On the other hand, if it is determined in step S2404 that the cause of the abnormality has been identified (i.e., if the answer in step S2404 is YES), the process proceeds to step S2406.

[0254] In step S2406, the pattern generation device 2020 generates an abnormal data pattern based on the added process value, the searched state variable, and the measured value, associates it with the cause of the abnormality, and stores it in the abnormal data pattern storage unit 1340.

[0255] In step S2407, the pattern generation device 2020 determines whether or not to terminate the pattern generation process. If it determines in step S2407 to continue the pattern generation process (if the answer is NO in step S2407), it returns to step S2401. On the other hand, if it determines in step S2407 to terminate the pattern generation process (if the answer is YES in step S2407), it terminates the pattern generation process.

[0256] <Summary> As is clear from the above description, the anomaly detection system 2010 according to the second embodiment includes a pattern generation device 2020 that operates in the pattern generation phase and an anomaly detection device 1130 that operates in the anomaly detection phase.

[0257] The pattern generation device 2020 is, The system includes a pre-tuned physical model that reproduces the behavior of the incineration process of an incinerator, and a trained predictive model that uses the calculation results of the pre-tuned physical model to predict the process values ​​of the incinerator. • By running a pre-tuned physical model and a pre-trained predictive model, a data pattern for plant anomalies (an example of the first data pattern) is generated when process values ​​with anomaly data are predicted.

[0258] Thus, the anomaly detection system 2010 according to the second embodiment generates anomaly data patterns based on process values ​​to which various anomaly data, including anomaly data that is unlikely to occur in an actual incinerator, are attached, and uses these anomaly data patterns to detect signs of anomaly.

[0259] As a result, according to the anomaly detection system 2010 of the second embodiment, it becomes possible to detect signs of anomalies in the plant with high accuracy.

[0260] [Third Embodiment] In the first embodiment described above, the case of detecting signs of an abnormality was explained, but detection is not limited to signs of an abnormality. For example, signs of the incinerator returning to a normal state may also be detected. In other words, the abnormality detection system, abnormality detection device, abnormality detection method, and abnormality detection program described in the first embodiment above may be configured as a detection system, detection device, detection method, and detection program for detecting signs of changes in the state of the plant. That is, detecting signs of changes in the state of the plant as referred to here includes both detecting signs of the plant returning to a normal state and detecting signs of the plant returning to an abnormal state.

[0261] Furthermore, normality prediction data, used to detect signs that a plant is about to return to a normal state, refers to data showing the difference before and after a specific measurement or state variable is changed so that the process value becomes the normal value or the best possible normal value. Normality prediction data (data indicating signs of returning to a normal state) is data that has been confirmed through simulations, etc., to ultimately improve the process value.

[0262] The detection system according to the third embodiment includes a pattern generation device that operates in the pattern generation phase and a state change detection device that operates in the state change detection phase.

[0263] The pattern generation device is A tuned physical model that reproduces the behavior of the plant's processing steps, and a trained predictive model that predicts the plant's process values ​​using the calculation results of the tuned physical model, are operated using a dataset that includes some data indicating signs of state changes. This generates data patterns (an example of the first data pattern) that occur when the plant undergoes a state change.

[0264] The state change detection device is • Using measurements taken during the plant's processing stages, data patterns (an example of a second data pattern) generated by running a tuned physical model and a trained predictive model are compared with data patterns observed during state changes. This allows for the detection of precursors to state changes.

[0265] As a result, the detection system according to the third embodiment makes it possible to detect signs of changes in the state of the plant with high accuracy.

[0266] [Fourth Embodiment] In the second embodiment described above, the case of detecting signs of an abnormality was explained, but detection is not limited to signs of an abnormality. For example, signs of the incinerator returning to a normal state may also be detected. In other words, the abnormality detection system, abnormality detection device, abnormality detection method, and abnormality detection program described in the second embodiment above may be configured as a detection system, detection device, detection method, and detection program for detecting signs of changes in the state of the plant. That is, detecting signs of changes in the state of the plant as referred to here includes both detecting signs of the plant returning to a normal state and detecting signs of the plant returning to an abnormal state.

[0267] Furthermore, when detecting signs that the plant is about to return to a normal state, the normal data (data indicating that the plant has returned to a normal state) refers to data showing the difference between the pre-improvement and post-improvement values ​​when the process values ​​have improved to normal values ​​or the best possible normal values.

[0268] The detection system according to the fourth embodiment includes a pattern generation device that operates in the pattern generation phase and a state change detection device that operates in the state change detection phase.

[0269] The pattern generation device is This involves running a tuned physical model that replicates the behavior of the plant's processing steps, and a trained predictive model that uses the calculation results of the tuned physical model to predict the plant's process values. This generates a data pattern (an example of the first data pattern) for when the plant's state changes, provided that process values ​​with data indicating state changes are predicted.

[0270] The state change detection device is • Using measurements taken during the plant's processing stages, data patterns (an example of a second data pattern) generated by running a tuned physical model and a trained predictive model are compared with data patterns observed during state changes. This allows for the detection of precursors to state changes.

[0271] As a result, the detection system according to the fourth embodiment makes it possible to detect signs of changes in the state of the plant with high accuracy.

[0272] [Fifth Embodiment] In each of the embodiments described above, the detection system was described as a system that detects signs of changes in the state of the target plant. However, the detection system is not limited to detecting signs of changes in the state of the target plant. For example, it may be configured to detect data patterns similar to the data patterns obtained when a tuned physical model and a trained predictive model are operated using a predetermined dataset. In other words, the detection system may function as a plant system that monitors the operating status of the plant using specific data patterns.

[0273] Specifically, the plant system according to the fifth embodiment includes a pattern generation device that operates in the pattern generation phase and a detection device that operates in the detection phase.

[0274] The pattern generation device is The system includes a generation unit that generates a first data pattern by operating a pre-tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the pre-tuned physical model, using a dataset representing a predetermined state of the plant. The system then modifies a portion of the data when operating these models with a dataset that includes a portion of the modified data.

[0275] The detection device is The system has an output unit that compares a second data pattern, generated by operating a tuned physical model and a trained predictive model using measurements taken in the plant's processing steps, with the first data pattern and outputs the comparison result. The system includes a determination unit that determines whether the first data pattern and the second data pattern are similar by calculating the similarity using known correlation functions or clustering methods based on the comparison results output from the output unit.

[0276] Alternatively, instead of a determination unit, a display unit capable of displaying the comparison results output from an output unit may be provided in the detection device, and an operator may determine similarity or non-similarity based on the comparison results displayed on the display unit.

[0277] Alternatively, the pattern generation device is The system operates a tuned physical model that reproduces the behavior of the plant's processing steps, and a trained predictive model that uses the calculation results of the tuned physical model to predict the process values ​​of the plant. This generates a first data pattern when the process values ​​for a predetermined state of the plant are predicted.

[0278] Furthermore, the detection device is The system has an output unit that compares a second data pattern, generated by operating a tuned physical model and a trained predictive model using measurements taken in the plant's processing steps, with the first data pattern and outputs the comparison result. The system includes a determination unit that determines whether the first data pattern and the second data pattern are similar by calculating the similarity using known correlation functions or clustering methods based on the comparison results output from the output unit.

[0279] Alternatively, instead of a determination unit, a display unit capable of displaying the comparison results output from an output unit may be provided in the detection device, and an operator may determine similarity or non-similarity based on the comparison results displayed on the display unit.

[0280] According to the detection system of the fifth embodiment, it becomes possible to monitor the operating status of the plant using specific data patterns.

[0281] [Sixth Embodiment] In the embodiments described above, the learning device and the pattern generation device were described as separate devices, but the learning device and the pattern generation device may be the same device. Also, in the embodiments described above, the pattern generation device and the state detection device including the anomaly detection device were described as separate devices, but the pattern generation device and the state detection device including the anomaly detection device may be the same device. Also, in the embodiments described above, the detection system including the anomaly detection system was described as being provided separately from the plant control system, but it may be provided as an integrated unit. Alternatively, some or all of the functions of the detection system including the anomaly detection system may be implemented in the plant control system.

[0282] Furthermore, although the state detection device, including the learning device, pattern generation device, and anomaly detection device, has been described in each of the above embodiments as being implemented by a single device, it may also be implemented by multiple devices. For example, each functional unit implemented by the state detection device, including the learning device, pattern generation device, and anomaly detection device, may be implemented in a distributed manner across multiple devices.

[0283] It should be noted that the present invention is not limited to the configurations shown in the above embodiments, including combinations with other elements. These aspects can be modified without departing from the spirit of the present invention and can be appropriately determined according to their application. [Explanation of symbols]

[0284] 10: Target Plant 100: Target equipment 110: Plant control system 120: Learning device 310M: Grate combustion model 310M': Adjusted grate combustion model 320M: Gas-phase combustion model 320M': Adjusted gas-phase combustion model 330M: Boiler heat recovery / power generation model 330M': Adjusted boiler heat recovery / power generation model 400: Measurement value acquisition unit 410: Parameter adjustment unit 420~440: Adjustment section 800: Learning Department 810: Predictive Model Learning Unit 900: Predictive Model 900': Pre-trained predictive model 1110: Anomaly detection system 1120: Pattern Generator 1130: Anomaly detection device 1310: Data acquisition unit for generating abnormal patterns 1320: Gray Box Model Section 1810: Measurement value acquisition unit 1820: Data pattern generation unit 1830: Detection unit 2010: Anomaly detection system 2020: Pattern Generator 2110: Gray Box Model Section 2120: Abnormal Data Pattern Generation Unit

Claims

1. A generation unit generates a first data pattern by operating a pre-tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the pre-tuned physical model, using a dataset of the plant in a predetermined state, by modifying a portion of the data and operating the model using a dataset that includes the modified data. An output unit that compares a second data pattern, generated by operating the adjusted physical model and the trained prediction model using the measured values ​​taken in the processing step of the plant, with the first data pattern and outputs the comparison result. A plant system having

2. A generation unit generates data patterns during state changes in the plant by operating a pre-tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the pre-tuned physical model, using a dataset that includes some data indicating signs of state changes. A detection unit detects signs of a state change by comparing the data pattern generated by operating the adjusted physical model and the trained prediction model using the measured values ​​taken in the processing step of the plant with the data pattern at the time of the state change. A detection system having the following features.

3. The generation unit stores the cause of the state change corresponding to a dataset that includes some data indicating the precursor of the state change, in association with the data pattern at the time of the state change. When the detection unit detects an indication of a change in state, it outputs the cause of the change in state along with the detection result. The detection system according to claim 2.

4. A generation unit generates a first data pattern when the process values ​​of the plant are predicted in a predetermined state by operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the process values ​​of the plant using the calculation results of the tuned physical model. An output unit that compares a second data pattern, generated by operating the adjusted physical model and the trained prediction model using the measured values ​​taken in the processing step of the plant, with the first data pattern and outputs the comparison result. A plant system having

5. A generation unit generates a data pattern for the state change of the plant when a process value with data indicating a state change is predicted, by operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained prediction model that predicts the process values ​​of the plant using the calculation results of the tuned physical model. A detection unit detects signs of a state change by comparing the data pattern generated by operating the adjusted physical model and the trained prediction model using the measured values ​​taken in the processing step of the plant with the data pattern at the time of the state change. A detection system having the following features.

6. The generation unit generates a data pattern during a state change by searching for state variables and measured values ​​that enable the trained prediction model to operate so that a process value with data indicating the state change is predicted. The detection system according to claim 5.

7. The generation unit stores the cause of the state change identified by searching for the state variable and measured value, in association with the data pattern at the time of the state change. When the detection unit detects an indication of a change in state, it outputs the cause of the change in state along with the detection result. The detection system according to claim 6.

8. If the cause of the state change cannot be identified by searching for the state variables and measured values, the generation unit searches for measured values ​​that enable the adjusted physical model to operate so that the searched state variables can be calculated. The detection system according to claim 6.

9. The aforementioned adjusted physical model is The values ​​are generated by inputting the measured values ​​taken in the processing step of the plant into a physical model, and then adjusting the parameters of the physical model using a Kalman filter so that the calculated values ​​calculated by the physical model approach the measured values ​​taken in the processing step of the plant. The detection system according to claim 2 or 5.

10. The aforementioned trained predictive model is This is constructed by training a predictive model using the observable state variables in the plant's processing process, which are calculated by operating the adjusted physical model using the measured values ​​taken in the plant's processing process, and the measured values ​​taken in the plant's processing process. The detection system according to claim 2 or 5.

11. The processing step of the aforementioned plant is the incineration process in the incinerator. The aforementioned adjusted physical model is an adjusted grate combustion model, an adjusted gas-phase combustion model, or an adjusted boiler heat acquisition and power generation model. The detection system according to claim 2.

12. The data pattern during the aforementioned state change is: The measured values ​​input to the aforementioned adjusted grate combustion model, or the measured values ​​after data indicating a change in state has been added to the measured values, The measured values ​​input to the aforementioned adjusted gas-phase combustion model, or the measured values ​​after data indicating a precursor to a change in state has been added to the measured values, The measured values ​​input to the adjusted boiler heat acquisition and power generation model, or the measured values ​​after data indicating a change in state has been added to the measured values, The measured values ​​input to the aforementioned trained predictive model, or the measured values ​​after data indicating a change in state has been added to the measured values, An unobservable state variable calculated by the aforementioned adjusted grate combustion model, or a state variable to which data indicating a precursor to a state change has been added, An unobservable state variable calculated by the aforementioned adjusted gas-phase combustion model, or a state variable to which data indicating a precursor to a state change has been added, An unobservable state variable calculated by the adjusted boiler heat acquisition and power generation model, or a state variable to which data indicating a precursor to a state change has been added, The process values ​​predicted by the aforementioned trained prediction model, The detection system according to claim 11, including the following:

13. The aforementioned adjusted grate combustion model is A physical model that reproduces the behavior of the incineration process of the aforementioned incinerator when the introduced waste undergoes complete combustion by sequentially going through the processes of drying, thermal decomposition, and combustion while moving on a movable grate, and is an adjusted physical model in which the parameters have been adjusted based on measurements taken during the incineration process. The detection system according to claim 11.

14. The aforementioned modified gas-phase combustion model is A physical model that reproduces the behavior of combustible gas generated by combustion in the incineration process of the aforementioned incinerator when it is combusted by supplied air, wherein the parameters of the adjusted physical model are adjusted based on measurements taken in the incineration process. The detection system according to claim 11.

15. The aforementioned adjusted boiler heat recovery / power generation model is, A physical model that reproduces the behavior of the incineration process of the aforementioned incinerator, specifically when steam generated by the absorption of heat from exhaust gas in a waste heat boiler rotates a steam turbine and generates electricity, wherein the parameters of the adjusted physical model are adjusted based on measurements taken during the incineration process. The detection system according to claim 11.

16. The aforementioned trained predictive model is This is a trained predictive model that predicts process values ​​of the incinerator, constructed by training a predictive model using the unobservable state variables in the incineration process, calculated by operating the adjusted grate combustion model, the adjusted gas-phase combustion model, and the adjusted boiler heat acquisition and power generation model using the measured values ​​taken in the incineration process of the incinerator, and the measured values ​​taken in the incineration process. The detection system according to claim 11.

17. A first data pattern is generated by operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the tuned physical model, using a dataset of the plant in a predetermined state, by modifying a portion of the data and operating the model using a dataset that includes the modified data. A step of comparing a second data pattern, generated by operating the adjusted physical model and the trained prediction model using the measured values ​​taken in the processing step of the plant, with the first data pattern, and outputting the comparison result. A detection method performed by each computer.

18. A process to generate data patterns during state changes in the plant by operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the tuned physical model, using a dataset that includes data that indicates signs of state changes; A process to detect signs of a state change by comparing the data pattern generated by operating the adjusted physical model and the trained predictive model using the measured values ​​taken in the processing step of the plant with the data pattern at the time of the state change. A detection method performed by each computer.

19. A step of generating a first data pattern when the process values ​​of the plant in a predetermined state are predicted by operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the process values ​​of the plant using the calculation results of the tuned physical model, A step of comparing a second data pattern, generated by operating the adjusted physical model and the trained prediction model using the measured values ​​taken in the processing step of the plant, with the first data pattern, and outputting the comparison result. A detection method performed by each computer.

20. A process of generating a data pattern for a change in the state of the plant when a process value with data indicating a change in state is predicted, by operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the process values ​​of the plant using the calculation results of the tuned physical model, A process to detect signs of a state change by comparing the data pattern generated by operating the adjusted physical model and the trained predictive model using the measured values ​​taken in the processing step of the plant with the data pattern at the time of the state change. A detection method performed by each computer.

21. When operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the tuned physical model, using a dataset of the plant in a predetermined state, a portion of the data is modified, and the model is operated using a dataset that includes the modified data, thereby enabling the computer of the detection device that stores the first data pattern to perform the operation. A step of comparing a second data pattern, generated by operating the adjusted physical model and the trained prediction model using the measured values ​​taken in the processing step of the plant, with the first data pattern, and outputting a comparison result. A detection program to execute.

22. By operating a pre-tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the pre-tuned physical model, with a dataset that includes some data indicating signs of state changes, the computer of the detection device that stores data patterns during state changes in the plant can be configured to: A process for detecting signs of a state change by comparing data patterns generated by operating the adjusted physical model and the trained predictive model using measurements taken in the processing steps of the plant with data patterns at the time of the state change. A detection program to execute.

23. By operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the tuned physical model, the computer of the detection device that stores the first data pattern when the process values ​​for a predetermined state of the plant are predicted, A step of comparing a second data pattern, generated by operating the adjusted physical model and the trained prediction model using the measured values ​​taken in the processing step of the plant, with the first data pattern, and outputting a comparison result. A detection program to execute.

24. By operating a tuned physical model that reproduces the behavior of the plant's processing steps and a trained predictive model that predicts the plant's process values ​​using the calculation results of the tuned physical model, when process values ​​with data indicating state changes are predicted, the computer of the detection device that stores the data pattern at the time of the plant's state change is configured to: A process for detecting signs of a state change by comparing data patterns generated by operating the adjusted physical model and the trained predictive model using measurements taken in the processing steps of the plant with data patterns at the time of the state change. A detection program to execute.

Images