JP2018092511A - Operational support device, apparatus operation system, control method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a monitoring device with a simulation function to aid in the monitoring of plants, etc.SOLUTION: An operational support device comprises: a process data collecting part for collecting, for each unit of one or a plurality of apparatuses, process data relating to the operating state of the apparatuses included in the unit; a prediction model constructing part for constructing a prediction model for each unit obtained by learning based on the collected process data the characteristics of the one or a plurality of apparatuses using a neural net or deep learning; and a control part for generating control data to be inputted into the prediction model.SELECTED DRAWING: Figure 12

Description

  The present invention relates to an operation support apparatus, a device operation system, an operation method, a control method, and a program.

  Conventionally, a plant or a machine to be monitored is modeled by a physical model or a statistical method model, and used for monitoring or diagnosing the plant. For example, in Patent Document 1, normal data detected by sensors provided in equipment in a chemical plant is collected, a learning model (normal model) is created by a technique such as clustering, and the sensors are detected during plant operation. There is a description of a system for predicting plant abnormalities by comparing evaluation target data and created learning models.

Japanese Patent No. 5845374

  However, for example, a conventionally used learning model disclosed in Patent Document 1 often deteriorates in accuracy due to changes in the operating conditions and usage environment of the plant and mechanical devices. Therefore, although it is suitable for monitoring a plant or the like that repeats the same operation pattern, it is not suitable for a plant or a mechanical device in which the operation pattern is not fixed.

  Accordingly, an object of the present invention is to provide an operation support apparatus, a device operation system, an operation method, a control method, and a program that can solve the above-described problems.

According to a first aspect of the present invention, a process data collection unit that collects process data related to an operation state of a device included in the unit of one or a plurality of devices, and the process data collection unit based on the collected process data. Operation support comprising: a prediction model building unit that builds a prediction model for each unit learned by neural network or deep learning of characteristics of one or a plurality of devices; and a control unit that generates control data to be input to the prediction model Device.
According to the prediction model construction unit in the first aspect, it is possible to construct a highly accurate prediction model that can cope with changes in the operating conditions and usage environment of the device. In addition, since the behavior of the equipment can be simulated by generating and inputting arbitrary control data for a highly accurate prediction model, the behavior of the actual equipment can be changed even when the operation pattern of the equipment changes. It is possible to accurately reproduce and operate the device according to the reproduced result.

The prediction model construction unit according to the second aspect of the present invention learns the characteristics for each unit of the one or more devices based on a combination of input data and output data with respect to the unit, and within the range of the learning. The prediction model is constructed without including a control process for the one or more devices.
By constructing a model with as small a unit as possible and including no control process, it is possible to simplify the prediction model to improve accuracy, and to improve the convenience of the constructed prediction model.

The said control part in the 3rd aspect of this invention is a control apparatus which performs control with respect to the said 1 or several apparatus.
By simply connecting an operation support device to an existing control device, it is possible to simulate a device using a prediction model.

The control unit according to a fourth aspect of the present invention is an emulator that simulates a control device that controls the one or more devices.
Even in the absence of a control device (actual machine), it is possible to perform device simulation at a relatively low cost by using an emulator in which the function of the control device is mounted on a PC or the like.

The operation support apparatus according to the fifth aspect of the present invention may further include a predetermined physical model or statistical model indicating characteristics of the one or more devices.
A highly accurate and proven physical model or statistical model created by an expert or the like may be used. Thereby, the prediction precision of the behavior of an apparatus can be raised.

The operation support apparatus according to the sixth aspect of the present invention may further include a reinforcement learning unit that optimizes a control parameter in the control unit or a control process by the control unit.
With this configuration, the control parameters and control process of the control device can be optimized.

The control unit according to a seventh aspect of the present invention acquires calculation data generated by a control calculation by the control device from a storage unit of a control device that controls the one or more devices, and uses the calculation data Control data.
With this configuration, the processing of the control device can be simulated with higher accuracy.

The prediction model in the eighth aspect of the present invention may include a physical model or a statistical model for a part of the characteristics of the one or more devices.
The accuracy of the prediction model can be increased by incorporating a highly accurate and proven physical model created by an expert or the like into a part of the prediction model.

  The prediction model construction unit according to the ninth aspect of the present invention uses the combination of input data and output data for the one or more devices as learning data, and the prediction when the input to the prediction model is the learning data A prediction model in which the output from the model has the same value as the learning data may be constructed.

In the tenth aspect of the present invention, when a physical model for a part of the characteristics of the one or more devices is included in an intermediate layer of the prediction model, the characteristics of the one or more devices are compressed to a low dimension. The physical model applied to the layer to be restored and the physical model applied to the layer to be restored to a higher dimension have an inverse function relationship.
According to the ninth and tenth aspects, a prediction model can be constructed by an hourglass neural network.

  The operation support apparatus according to the eleventh aspect of the present invention compares a predicted value of process data based on the prediction model with actual process data acquired from the one or more devices corresponding to the prediction model. An evaluation unit that evaluates actual process data is further provided.

An operation support apparatus according to a twelfth aspect of the present invention is based on a process data collection unit that collects process data related to an operation state of a device included in a unit of one or a plurality of devices, and the collected process data. A prediction model building unit that builds a prediction model for each unit learned by neural network or deep learning of characteristics of the one or more devices, a predicted value of process data based on the prediction model, and the device An evaluation unit that compares the acquired actual process data with each other and evaluates the actual process data.
According to the eleventh and twelfth aspects, the state of the operating device can be evaluated.

  A thirteenth aspect of the present invention is a device operation system comprising the operation support device according to any one of the above, the one or more devices, and a control device for the one or more devices. .

According to a fourteenth aspect of the present invention, the operation support device according to any one of the above is provided together with the one or more devices and the control device for the one or more devices, and the one or more devices are provided. It is an operation method to operate the equipment.
According to the thirteenth and fourteenth aspects, a device group to be monitored and the operation support device are provided side by side to simulate the behavior of the device to be monitored by the operation support device. Can be operated.

  According to a fifteenth aspect of the present invention, the operation support apparatus collects process data relating to the operating state of the equipment included in the unit in units of one or a plurality of equipment, and based on the collected process data, the 1 It is a control method that constructs a prediction model for each unit in which characteristics of one or a plurality of devices are learned by a neural network or deep learning, and generates control data to be input to the prediction model.

  According to a sixteenth aspect of the present invention, the computer of the operation support apparatus collects process data relating to the operation state of the device included in the unit in units of one or more devices, based on the collected process data. A program for functioning as means for constructing a prediction model for each unit in which characteristics of the one or more devices are learned by neural network or deep learning, and means for generating control data to be input to the prediction model. is there.

  According to the present invention, since a highly accurate prediction model can be constructed and monitoring can be performed while performing a simulation using the prediction model, it is possible to appropriately monitor a plant or the like whose operation pattern is not fixed. Become.

It is a schematic block diagram which shows the structural example of the apparatus operation system in 1st embodiment of this invention. It is a functional block diagram which shows an example of the operation assistance apparatus in 1st embodiment of this invention. It is a schematic diagram which shows the application object of the operation assistance apparatus in 1st embodiment of this invention. It is a 1st figure which shows the example of application object of the operation assistance apparatus in 1st embodiment of this invention. It is a 2nd figure which shows the example of application object of the operation assistance apparatus in 1st embodiment of this invention. It is a 3rd figure which shows the example of application object of the operation assistance apparatus in 1st embodiment of this invention. It is a 1st figure explaining construction of the prediction model by the operation support device in a first embodiment of the present invention. It is a 2nd figure explaining construction of the prediction model by the operation support device in a first embodiment of the present invention. It is a 1st figure explaining evaluation of the process data using the prediction model by the operation assistance apparatus in 1st embodiment of this invention. It is a 2nd figure explaining evaluation of the process data using the prediction model by the operation assistance apparatus in 1st embodiment of this invention. It is a flowchart of operation | movement of the operation assistance apparatus in 1st embodiment of this invention. It is a functional block diagram which shows an example of the operation assistance apparatus in 2nd embodiment of this invention. It is a 1st figure explaining the simulation using the prediction model by the operation assistance apparatus in 2nd embodiment of this invention. It is a 2nd figure explaining the simulation using the prediction model by the operation assistance apparatus in 2nd embodiment of this invention. It is a flowchart of operation | movement of the operation assistance apparatus in 2nd embodiment of this invention. It is a functional block diagram which shows an example of the operation assistance apparatus in 3rd embodiment of this invention. It is a figure explaining the reinforcement learning by the operation assistance apparatus in 3rd embodiment of this invention. It is a functional block diagram which shows an example of the operation assistance apparatus in 4th embodiment of this invention. It is a figure explaining the simulation using the prediction model by the operation assistance apparatus in 4th embodiment of this invention. It is a flowchart of operation | movement of the operation assistance apparatus in 4th embodiment of this invention. It is a figure which shows the example of application of the operation assistance apparatus in 1st embodiment-4th embodiment of this invention. It is a figure which shows an example of the hardware constitutions of the operation assistance apparatus in 1st embodiment-4th embodiment of this invention.

<First embodiment>
Hereinafter, a device operation system according to a first embodiment of the present invention will be described with reference to FIGS.
FIG. 1 is a schematic block diagram showing a configuration example of a device operation system in the first embodiment of the present invention.
The device operation system 50 monitors the state of the device group 100, detects an abnormality, and the like. As illustrated in FIG. 1A, the device operation system 50 includes a device group 100 and an operation support device 10. The device group 100 and the operation support apparatus 10 are connected via a network, and the operation support apparatus 10 acquires process data from the devices included in the device group 100 and constructs a prediction model that reflects the characteristics of each device ( Learning mode). In addition, the operation support apparatus 10 acquires process data from the devices included in the device group 100 and evaluates the state of the devices based on the constructed learning model (operation mode). The device group 100 is a plant, for example. Hereinafter, the device group 100 may be referred to as a plant 100. Further, the operation support apparatus 10 of the present embodiment is a monitoring apparatus. The process data refers to a control signal (hereinafter referred to as control data) output from a control device that controls the device group 100, a measurement value detected by a sensor provided in the vicinity of the device, or control data or a measurement value. A set of operation values based on it.
Note that the device group 100 and the operation support device 10 are directly connected as in the device operation system 50 ′ illustrated in FIG. 1B, and the operation support device 10 acquires process data directly from the devices included in the device group 100. A configuration is also possible.

FIG. 2 is a functional block diagram showing an example of the operation support apparatus in the first embodiment of the present invention.
As shown in FIG. 2, the operation support apparatus 10 includes a process data collection unit 11, an input / output unit 12, a communication unit 13, an operation mode management unit 14, a storage unit 15, a prediction model construction unit 16, And an evaluation unit 17.
The process data collection unit 11 collects process data relating to the operation state for each unit of one or a plurality of devices.
The input / output unit 12 receives input of operation instruction information from the operator to the operation support apparatus 10 and outputs evaluation result information by the evaluation unit 17.
The communication unit 13 communicates with other devices. For example, the communication unit 13 receives process data output by each device via a communication device provided in the device group 100.
The storage unit 15 stores various data. For example, the storage unit 15 stores the process data collected by the process data collection unit 11.
Based on the process data collected by the process data collection unit 11, the prediction model construction unit 16 constructs a prediction model for each one or a plurality of devices that has learned characteristics in an operation range in which one or a plurality of devices operate. . In constructing the prediction model, the prediction model construction unit 16 uses a neural network and a deep learning technique.
The evaluation unit 17 compares the predicted value of the process data based on the prediction model with the actual process data acquired from the device, and evaluates the soundness of the actual process data.
The operation mode management unit 14 sets the operation mode of the operation support apparatus 10. Specifically, the operation mode management unit 14 switches and sets a learning mode for constructing a prediction model and an operation mode for monitoring the device group 100.

FIG. 3 is a schematic diagram showing an application target of the operation support apparatus in the first embodiment of the present invention.
FIG. 3 is a schematic diagram of the control device 20 in the device group 100 and the devices 1 to 3 that are the control targets. The control device 20 outputs control data. The device 1 inputs control data or a substance to be input (fuel, medium, current, etc. flowing between the devices) controlled based on the control data, and performs processing and operation of the device 1. The device 2 performs processing and operation of the device 2 by inputting a substance generated by the processing and operation of the device 1. The device 3 performs processing and operation of the device 3 by inputting a substance generated by the processing and operation of the device 2. Further, the control device 20 receives process data generated by the processing and operation of the device 3, generates next control data by feedback control or the like, and outputs it to the device 1 side. The configuration illustrated in FIG. 3 is an example, and the present invention is not limited to this. For example, the control device 20 may further output control data to the devices 2 and 3, or the devices 1 to 3 may be parallel to the control device 20. Also, the number of devices is not limited to three. The same applies to the schematic diagram of the plant illustrated in the following figures.

FIG. 4 is a first diagram illustrating an application target example of the operation support apparatus according to the first embodiment of the present invention.
FIG. 4 shows a schematic diagram of the gas turbine plant 100A. The control device 20 controls the flow rate of air flowing into the compressor 31D, the amount of fuel gas supplied to the combustor 31E by adjusting the opening of the fuel flow rate adjusting valves 31A to 31C, and the like, and rotationally drives the turbine 31F. Then, the generator 31G is operated. When the gas turbine plant 100A is applied to the schematic diagram of FIG. 3, for example, the combustor 31E is the device 1, the turbine 31F is the device 2, and the generator 31G is the device 3. The control device 20 acquires the output value of the generator 31G and adjusts the opening of the fuel flow rate adjustment valve 31A and the like.

FIG. 5 is a second diagram showing an application target example of the operation support apparatus in the first embodiment of the present invention.
FIG. 5 shows a schematic diagram of the steam turbine plant 100B. The steam turbine plant 100B includes a high pressure steam supply valve 32A, an intermediate pressure steam supply valve 32B, a high pressure steam turbine 32C, an intermediate pressure steam turbine 32D, a low pressure steam turbine 32E, a generator 32F, a condenser 32G, a low pressure feed water heater 32H, It includes a deaerator 32I, a feed water pump 32J, a high pressure feed water heater 32K, a boiler 32L, the control device 20, and the like. Also in the case of the steam turbine plant 100B, for example, the high pressure steam turbine 32C, the intermediate pressure steam turbine 32D, the low pressure steam turbine 32E, the generator 32F, the condenser 32G, the low pressure feed water heater 32H, the deaerator 32I, and the high pressure feed water heater. 32K or the like can be applied to the devices 1 to 3 shown in FIG. As will be described later, in the present embodiment, a prediction model in which the characteristics of each device are learned for each of the devices 1 to 3 shown in FIG. 3 is constructed, but the target of one prediction model is not limited to one device. In FIG. 5, for example, the condenser 32G and the low-pressure feed water heater 32H may be grouped and handled as the device 1 to construct the prediction model.

FIG. 6 is a third diagram showing an application target example of the operation support apparatus in the first embodiment of the present invention.
In FIGS. 4 and 5, the plant is exemplified as an application target of the operation support apparatus 10. However, the application target of the operation support apparatus 10 is not limited to a plant, and may be a single machine apparatus including a plurality of devices. As an example of the mechanical device, FIG. 6 shows a conceptual diagram of the engine. The device group 100C is an engine including an intake / exhaust system 33A, a fuel system 33B, a lubrication system 33C, and a cooling system 33D. In the case of the device group 100C (engine), for example, the intake / exhaust system 33A, the fuel system 33B, the lubrication system 33C, the cooling system 33D, and the like can be applied to the devices 1 to 3 shown in FIG.

FIG. 7 is a first diagram illustrating the construction of a prediction model by the operation support apparatus in the first embodiment of the present invention.
As shown in FIG. 7, in the operating plant 100, the operation support apparatus 10 acquires input data and output data for the device 1, and constructs a prediction model 161 that is a learning model indicating the characteristics of the device 1. Similarly, the operation support apparatus 10 acquires the input / output data of the device 2 and obtains a prediction model 162 that indicates the characteristics of the device 2. The operation support apparatus 10 acquires the input and output data of the device 3 and indicates the prediction model 163 that indicates the characteristics of the device 3. To construct. More specifically, the process data collection unit 11 acquires input data and output data for the device 1 and stores them in the storage unit 15. Further, the prediction model construction unit 16 constructs a prediction model 161 based on the input / output data stored in the storage unit 15 by using a machine learning (neural network, deep learning) technique. The same applies to the prediction models 162 and 163.

  Next, the construction of the prediction model by the prediction model construction unit 16 will be described. Conventionally, a model constructed by a physical model or a statistical method is applied to equipment or the like provided in a plant to be useful for monitoring, an abnormality sign, or the like. However, models constructed using these methods can simulate the exact behavior of a device within a certain range, but if the operating conditions and usage environment deviate from that range, there will be a large discrepancy from the actual behavior of the device. It is common to become. Then, when an unexpected event occurs, the accuracy of monitoring using the model and the abnormality sign deteriorates. In addition, in order to accurately model the behavior of the plant, complicated analysis and tuning are usually required. In the present embodiment, the prediction model construction unit 16 learns the characteristics of the equipment 1 and the like using a neural network or the like as a learning method using time-series input / output data of the equipment 1 and the like collected during operation of the plant 100. To do. Thereby, the prediction model reflecting the correlation between the process data, the response delay, and the dynamic characteristics can be constructed only by giving the time-series input / output data of the devices 1 to 3 of the plant 100. Learning methods used by the prediction model construction unit 16 include, for example, AANN (Auto Associative Neural Network), RNN (Recurrent Neural Network), LSTM (Long short-term memory), which is an extended model of RNN, and the like. A deep learning method such as a GRU (Gated Recurrent Unit) having the above performance. It has been found that by building a prediction model using these learning methods, a prediction model that can be adapted to a wider range of operating conditions and usage environments can be constructed.

  Further, the prediction model construction unit 16 learns a prediction model by using one or a plurality of devices as a unit (a group of a plurality of devices is described as a component), and using input data and output data to that device or component as learning data. . For example, the prediction model construction unit 16 learns input and output time series data of the low-pressure feed water heater 32H stored in the storage unit 15 for the low-pressure feed water heater 32H alone in FIG. A 32H prediction model may be constructed. In addition, the prediction model construction unit 16 inputs and degass the condenser 32G for one component in which the three devices of the condenser 32G, the low pressure feed water heater 32H, and the deaerator 32I in FIG. 5 are combined. The time series data of the output of the device 32I may be learned to construct a prediction model for this component.

If the range of the target device to be modeled is widened, it is difficult to specify the location of occurrence when an abnormality occurs. Therefore, the range of the target device to be modeled is preferably as small as possible (for example, one device).
In addition, only the device to be controlled independent of the control device is modeled. This is because the model becomes complicated if a control process is included in the model. The operation of the control device is well known by the manufacturer and does not need to be modeled. Therefore, in this embodiment, model construction is performed for the range excluding the control process by the control device. For example, in FIG. 5, when adjusting the opening of the high-pressure steam supply valve 32A according to the output (temperature, etc.) of the high-pressure steam turbine 32C, the high-pressure steam supply valve 32A and the high-pressure steam turbine 32C are one component. Since the control process for adjusting the opening degree of the high-pressure steam supply valve 32A is included inside, no model is constructed in units of this range.

Next, an example of the learning method will be described.
FIG. 8 is a second diagram for explaining the construction of the learning model by the operation support apparatus in the first embodiment of the present invention.
FIG. 8 is a configuration diagram of a learning model by AANN. The data flow from the input layer L1 to the intermediate layer L3 indicates a process in which the feature to be modeled is dimensionally compressed to a low dimension, and the data flow from the intermediate layer L3 to the output layer L5 is restored to a high dimension. Shows the process. When the prediction model construction unit 16 constructs a prediction model for the combustor 31E in FIG. 4, for example, the fuel flow rate of the main system, the fuel flow rate of the pilot system, the fuel flow rate of the top hat system, and the inflow from the compressor 31D The air flow rate (four pieces of input data) and the inlet temperature of the turbine 31F (one piece of output data) are input to the input layer L1 as learning data. Then, the prediction model construction unit 16 learns the coupling coefficient of each network so that the error between the learning data input from the input layer L1 and the output data output from the output layer L5 becomes small (becomes 0). A method such as a back propagation method is used for learning the coupling coefficient. When the learning is completed, one set of data having the same value as the learning data input to the input layer L1 is output from the output layer L5 of the prediction model 161.

  Further, since the prediction model 161 learns the characteristics of the device 1 by dimensional compression, even if a set of input data including an outlier is input, the prediction model 161 is not affected by the outlier so that the output model L5 is not affected. It is configured to output output data. For example, the prediction model 161 configured to input learning data (10, 10, 10, 10) and output (10, 10, 10, 10) is the learning data (20, 10, 10, 10). ) Is input, the influence of the outlier “20” is reduced and, for example, (11, 10, 10, 10) is output. At this time, a difference occurs between the learning data input to the input layer L1 and the data output from the output layer L5. When the divergence occurs, it can be determined that some abnormality has occurred. As will be described later, the evaluation unit 17 uses this property to set the output data from the output layer L5 when the input / output data set of the equipment of the plant 100 is input to the input layer L1 of the prediction model 161. Then, it is determined whether or not an abnormality has occurred in the device 1 by comparing with the set of input data.

In the figure, the node group L2A of the intermediate layer L2 surrounded by a broken line is a physical model or a statistical model. This physical model or the like is a model created by an expert or the like based on the physical and mechanical characteristics of the target device. When there is an appropriate (proven) physical model that is known in advance for some of the characteristics of the device 1, the physical model can be incorporated into a part of the intermediate layer of the prediction model 161. Further, the node group L4A of the intermediate layer L4 is a node group corresponding to the node group L2A. For example, when the physical model applied to the intermediate layer L2 is represented by the function y = f (x), the physical model applied to the intermediate layer L4 is represented by the inverse function x = f ⁻¹ (y). It means that you can.

  The set of learning data given to the input layer L1 includes the value of the control data output by the control device 20, the measured value measured on the input side or the output side of the device, or the calculation based on them. Value (mass flow calculated from volume flow).

  8 explains the construction of the prediction model by AANN. For example, in the case of RNN or LSTM, for example, input data (for example, the value of control data) for the device 1 is given to the input layer, and the device 1 Learning is performed so that data having the same value as output data (for example, a measured value on the downstream side of the device 1) is output. The error back-propagation method can also be used for learning in this case.

FIG. 9 is a first diagram illustrating process data evaluation using a prediction model by the operation support apparatus according to the first embodiment of the present invention.
As shown in FIG. 9, in the operating plant 100, the operation support apparatus 10 acquires input data and output data for the device 1 and evaluates whether these values are healthy. Specifically, the evaluation unit 17 performs evaluation using the prediction model 161. For example, when the prediction model 161 is the model by AANN illustrated in FIG. 8, the evaluation unit 17 sets the evaluation data (for example, three systems of fuel flow rates in the above example, The air inflow amount and the turbine inlet temperature) are input to the input layer L1 of the prediction model 161 and compared with the prediction data output from the output layer L5. For example, the evaluation unit 17 evaluates that the state of the device 1 is abnormal if the difference between the two is equal to or greater than a threshold value. Alternatively, when the prediction model 161 is a model constructed by RNN, for example, the evaluation unit 17 inputs the input data of the device 1 to the input layer of the prediction model 161, and the prediction data output from the output layer of the prediction model 161 The output data (actually measured value) of the device 1 is compared, and if the difference between the two is equal to or greater than the threshold value, it is evaluated that the state of the device 1 is abnormal.

The evaluation unit 17 determines that the deviation between the value of the prediction data and the actual measurement value is a threshold value, and compares the cumulative sum of deviations with a predetermined threshold value, and is abnormal if the cumulative sum of deviations is equal to or greater than the threshold value. You may evaluate. Further, the evaluation unit 17 may evaluate, for example, the prediction data when the deviation change is larger than that of the past deviation history by using a statistical method such as a sequential probability ratio test or ChangeFinder. FIG. 10 shows an example of an abnormality evaluation method.
FIG. 10 is a second diagram illustrating process data evaluation using a prediction model by the operation support apparatus according to the first embodiment of the present invention.
The vertical axis in FIG. 10 indicates the integrated value of the deviation between the predicted data and the actual process data, and the horizontal axis indicates the measurement time. Errors are accumulated as the amount of measurement data increases (time elapses). FIG. 10 shows an upper limit value (Un) and a lower limit value (Ln) of the error integrated value. For example, the evaluation unit 17 determines that the deviation (Sn) between the predicted data and the actual process data is equal to or higher than the upper limit value, and is normal when the deviation is equal to or lower than the lower limit value.

  As described above, the operation support apparatus 10 according to the present embodiment acquires process data from the devices or components included in the plant 100 and constructs a prediction model corresponding to each individual device or component (learning mode). And the operation support apparatus 10 acquires process data from the apparatus or component with which the plant 100 is equipped, and evaluates based on the prediction model which constructed | assembled the acquired process data (operation mode). Next, the overall operation of the operation support apparatus 10 will be described with reference to FIG.

FIG. 11 is a flowchart of the operation of the operation support apparatus in the first embodiment of the present invention.
In the following description, the case where the prediction model 161 of the device 1 is constructed and the process data is evaluated based on the prediction model 161 will be described as an example.
First, the process data collection unit 11 acquires process data of the device 1 from the operating plant 100 (step S11). Specifically, the process data collection unit 11 acquires input data and output data of the device 1. The process data collection unit 11 stores the acquired process data in the storage unit 15 in association with the acquisition time (step S12). Next, the operation mode management unit 14 determines whether or not the operation mode of the operation support apparatus 10 is the “learning mode” (step S13). For example, the current operation mode is recorded in the storage unit 15 by the operation of the operator, and the operation mode management unit 14 determines whether it is “learning mode” or “operation mode” based on this recording. In the learning mode (step S13; Yes), the prediction model construction unit 16 constructs a prediction model 161 (step S14). Specifically, the prediction model construction unit 16 constructs a model using a neural network or a deep learning method using the time-series input / output process data stored in the storage unit 15 as learning data. By using these methods, it is possible to construct a highly accurate prediction model 161 reflecting the dynamics and non-linearity of the plant and the mechanical device. The prediction model construction unit 16 records the constructed prediction model 161 in the storage unit 15. Next, the operation mode management unit 14 determines whether or not to end the operation of the operation support apparatus 10 (step S17). For example, when the operator performs an end instruction operation, the input / output unit 12 receives the instruction operation and outputs it to the operation mode management unit 14. The operation mode management unit 14 determines whether to end the operation support apparatus 10 based on the presence / absence of the instruction operation. When the process ends (step S17; Yes), the flowchart ends. When the operation is continued (step S17; No), the processing from step S11 is repeated.

  On the other hand, when the determination result in step S13 is the operation mode (step S13; No), the evaluation unit 17 inputs process data to the prediction model 161 recorded in the storage unit 15. The prediction model 161 calculates prediction data (step S15). The evaluation unit 17 acquires a prediction result (prediction data) based on the prediction model 161. The evaluation unit 17 compares the predicted data with the process data (evaluation data) and evaluates the process data (step S16). For example, the evaluation unit 17 evaluates that the evaluation data is abnormal if the difference between the prediction data and the process data is greater than or equal to a threshold value. The input / output unit 12 outputs and displays the evaluation result by the evaluation unit 17 on a display, for example. Next, when the operation mode management unit 14 determines whether the operation of the operation support apparatus 10 is finished and does not finish, the processing from step S11 is repeated.

  According to the present embodiment, it is possible to construct a highly accurate prediction model that reflects correlation between process data, device dynamics such as response delay and dynamic characteristics, and nonlinearity. In addition, it is possible to construct a highly accurate prediction model that can cope with changes in the operating conditions of the equipment and the usage environment. Thereby, the process state estimation accuracy of a plant or a mechanical device can be improved, and more appropriate monitoring can be performed.

<Second embodiment>
Hereinafter, a device operation system according to a second embodiment of the present invention will be described with reference to FIGS.
The operation support apparatus 10 </ b> A according to the second embodiment has a simulation function for simulating the behavior of the plant 100. When the operation support apparatus 10A is used, it is possible to operate and monitor the plan 100 while simulating the behavior of the plant 100. The operation support apparatus 10A of the second embodiment is a monitoring apparatus with a simulation function.

FIG. 12 is a functional block diagram showing an example of the operation support apparatus in the second embodiment of the present invention.
Of the configurations according to the second embodiment of the present invention, the same components as those constituting the operation support apparatus 10 according to the first embodiment are denoted by the same reference numerals, and description thereof is omitted. As illustrated, the operation support apparatus 10A includes a process data collection unit 11, an input / output unit 12, a communication unit 13, an operation mode management unit 14, a storage unit 15, a prediction model construction unit 16, and an evaluation unit 17. And a control simulation unit 18.
The control simulation unit 18 generates simulation control data to be input to the prediction model 161 and the like constructed by the prediction model construction unit 16 and inputs the simulation control data to the prediction model 161 and the like. Moreover, the simulation process data which the prediction model 161 grade | etc., Outputs with respect to the input is acquired, and new simulation control data is produced | generated. In this way, the control simulation unit 18 simulates the behavior of the plant 100.

FIG. 13 is a first diagram illustrating a simulation using a prediction model by the operation support apparatus according to the second embodiment of the present invention.
As illustrated in FIG. 13, the operation support apparatus 10A acquires a demand for the plant 100 (the control apparatus 20), an operation condition of the plant 100, and a use environment. Then, the operation support apparatus 10A simulates the behavior of the plant 100. In the figure, prediction models 161, 162, and 163 are prediction models constructed in the same manner as in the first embodiment, and are stored in the storage unit 15. Also, as shown in the drawing, the plurality of prediction models 161 to 163 are connected to the input / output between the control simulation unit 18 and the respective prediction models 161 to 163 by the control simulation unit 18 based on the configuration of the devices 1 to 3. The

The control simulation unit 18 of the operation support apparatus 10 </ b> A has a function as an emulator that performs the same control as the control apparatus 20. Further, the control simulation unit 18 simulates the flow of process data among the control device 20, the prediction model 161, the prediction model 162, and the prediction model 163.
That is, when the operation support apparatus 10A obtains an operation condition such as a demand for the plant 100 (for example, power to be generated if it is a power plant), the control simulation unit 18 generates simulated control data according to the demand, It outputs to the prediction model 161 which modeled the apparatus 1. FIG. The prediction model 161 predicts the output value of the device 1 according to the input simulation control data, and outputs the prediction data. The control simulation unit 18 acquires this prediction data and outputs it to the prediction model 162. The prediction model 162 acquires this prediction data, predicts the output value of the device 2, and outputs the prediction data. Similarly, the control simulation unit 18 outputs the prediction data output from the prediction model 162 to the prediction model 163. The prediction model 163 predicts the output value of the device 3 based on the prediction data and outputs the prediction data. The control simulation unit 18 acquires the prediction data output from the prediction model 163, performs feedback control or the like, and generates the next simulation control data.

On the other hand, the evaluation unit 17 receives the input / output data of the device 1, the input / output data of the device 2, and the input / output data of the device 3 when an operation corresponding to the simulation content is performed, for example, via the process data collection unit 11. get. Alternatively, the evaluation unit 17 acquires reference data (data serving as an evaluation criterion) input by the operator via the input / output unit 12. Then, the evaluation unit 17 compares the input / output data (or reference data) of the device 1 with the input / output data of the prediction model 161 and evaluates the state of the prediction model 161 generated by the simulation. Similarly, the evaluation unit 17 evaluates whether the prediction model 162 and the prediction model 163 are abnormal. As a result, the operator can check whether the simulated contents are correct.
Thereafter, this process is repeated. It is considered that the higher the accuracy of the prediction models 161 to 163, the smaller the deviation between the result of the simulation performed by the operation support apparatus 10A and the behavior shown by the plant 100.

For example, the input / output unit 12 acquires control data from the control simulation unit 18, input data output to the prediction models 161 to 163, and output data output from the prediction models 161 to 163, and outputs and displays them on the display device. Also good. The operator compares the simulated process data by the operation support apparatus 10A (control simulation unit 18) displayed on the display device with the actual process data obtained from the plant 100 and the reference data input by the operator, and determines the state of the plant 100. Can be monitored.
Note that the prediction models 161 to 163 need not all be constructed by the prediction model construction unit 16, but include physical models (including statistical models) created by experts based on physical and mechanical characteristics. ). As in the first embodiment, a physical model created by an expert may be incorporated in a part of the prediction models 161 to 163.

According to the simulation function of the present embodiment, the process state of each device 1 to 3 is estimated with high accuracy by simulating the behavior of the plant 100 using the prediction model 161 and the like constructed by the prediction model construction unit 16. be able to. Further, by comparing the result of highly accurate simulation with actual process data obtained from the plant 100, it is possible to grasp how far the actual state of the device 1 etc. is from the state on the simulation. it can.
In addition, by simulating simulated process data on each of the devices 1 to 3 obtained from the simulation results, the simulated process data of the entire plant 100 reflecting the correlation between the process data, the dynamics of the devices, and the non-linearity is referenced. However, since the actual plant 100 can be monitored, more appropriate monitoring can be performed.

  Further, according to the simulation function of the present embodiment, the operation simulation of the plant 100 can be performed by the control simulation unit 18 while simulating a state different from the current state of the plant 100. For example, it is assumed that process data suspected of being abnormal is detected in the device 1. If the control simulation part 18 is used in such a scene, the operator can consider how to return the apparatus 1 to a normal state. For example, the operator inputs operation instruction information that returns the output data of the device 1 (prediction model 161) to the operation support apparatus 10A. Then, the input / output unit 12 receives the operation instruction information and outputs it to the control simulation unit 18. The control simulation unit 18 generates control data corresponding to the operation instruction information and outputs the control data to the prediction model 161. The operator looks at the output data of the prediction model 161 and confirms the behavior of the prediction model 161 (device 1). If the output data of the prediction model 161 is normal, it can be confirmed that the operation instruction information is appropriate. Then, the operator can perform the confirmed operation on the actual plant 100 to return the behavior of the device 1 to a normal state. As described above, since the operation can be verified in advance by the simulation function, the convenience when an abnormality occurs is improved.

FIG. 14 is a second diagram illustrating a simulation using a prediction model by the operation support apparatus according to the second embodiment of the present invention.
As a modification of the present embodiment, as shown in FIG. 14, for example, the actual plant 100, simulation by the control simulation unit 18, and actual process data are input to the prediction model 161 and the like as in the first embodiment. The process of obtaining the prediction data may be performed in parallel. More specifically, the operation support apparatus 10 </ b> A acquires actual process data related to the devices 1 to 3 of the plant 100 while simulating the behavior of the plant 100, and evaluates the actual process data. That is, in parallel with the simulation model 18 that simulates the operation state different from the actual plant 100 between the prediction models 161 to 163, the process data collection unit 11 is connected to the device as in the first embodiment. 1 actual process data is acquired, and the evaluation unit 17 inputs the actual process data to the prediction model 161. Then, the evaluation unit 17 compares the actual process data with the evaluation data, and evaluates the state of the device 1 based on the degree of divergence between the two data.

  If the evaluation of the actual process data by the evaluation unit 17 is performed in this way, the operation monitoring of the plant 100 is performed while simulating a state different from the current state of the plant 100 using the control simulation unit 18. Convenience in the scene is improved. For example, it is assumed that process data suspected of being abnormal in the device 1 is detected by the evaluation of the evaluation unit 17. In such a situation, the operator can verify what operation is performed to return the device 1 to a normal state by causing the control simulation unit 18 to perform a simulation. In the meantime, the evaluation unit 17 continues to evaluate the device 1 and the like, so the operator can leave the monitoring of the plant 100 to the operation support apparatus 10A and concentrate on countermeasures against abnormalities.

  Note that, when the operation method illustrated in FIG. 13 is used, the configuration of the operation support apparatus 10A may be configured by removing the evaluation unit 17 from the block diagram illustrated in FIG. That is, in this case, in parallel with the operation of the plant 100, a simulation by the control simulation unit 18 is performed, and the operator monitors the operation of the plant 100 while referring to the simulation result.

Next, an example of operation | movement of 10 A of operation assistance apparatuses of this embodiment is demonstrated.
FIG. 15 is a flowchart of the operation of the operation support apparatus in the second embodiment of the present invention.
The processing similar to that in FIG. 11 will be briefly described. First, the process data collection unit 11 acquires process data of the device 1 from the operating plant 100 (step S11). The process data collection unit 11 stores the acquired process data in the storage unit 15 in association with the acquisition time (step S12). Next, the operation mode management unit 14 determines whether or not the operation mode of the operation support apparatus 10 is the “learning mode” (step S13). In the learning mode (step S13; Yes), the prediction model construction unit 16 constructs a prediction model 161 (step S14). The prediction model construction unit 16 records the constructed prediction model 161 in the storage unit 15. The operation in the above learning mode is the same as in the first embodiment.
Next, the operation mode management unit 14 determines whether or not to end the operation of the operation support apparatus 10 (step S17). When the process ends (step S17; Yes), the flowchart ends. When the operation is continued (step S17; No), the processing from step S11 is repeated.

  On the other hand, when the determination result in step S13 is the operation mode (step S13; No), the control simulation unit 18 transmits the information input by the operator, such as operating conditions, usage environment, and operation instructions, via the input / output unit 12. get. The operating condition information is, for example, the magnitude of electric power to be output if the plant 100 is a power plant, and the operating environment is, for example, the atmospheric temperature. The control simulation unit 18 performs a control calculation based on the acquired operating condition information (step S131). For example, the control simulation unit 18 calculates and generates control data instructing the device 1 and outputs the control data to the prediction model 161. Next, the prediction model 161 and the like calculate prediction data (step S15). Next, the evaluation unit 17 acquires a prediction result (prediction data) based on the prediction model 161, and evaluates the prediction data by comparing with, for example, reference data input by the operator (step S161). For example, the evaluation unit 17 evaluates that the prediction data is abnormal if the difference between the prediction data and the reference data is greater than or equal to a threshold value. As a result, the operator can confirm whether or not the operation instruction input by the operator is a desired operation instruction. Next, the operation mode management unit 14 determines the end of the operation of the operation support apparatus 10A. If the operation mode management unit 14 does not end the operation, the processing from step S131 is repeated. That is, the simulation by the control simulation unit 18 can be repeated independently of the actual plant side.

  According to the present embodiment, in parallel with the operation of the plant 100, the behavior of the plant 100 can be simulated based on a highly accurate prediction model constructed in the learning mode. Moreover, the state evaluation of the apparatus 1 etc. can be performed by comparing the simulation result by each prediction model 161 etc. with the actual process data and reference data regarding each apparatus 1 etc. with which the plant 100 is equipped. In addition, since a simulation using the high-precision prediction model 161 and the like can be performed independently of the plant 100, for example, even in a scene where an operation that is significantly different from the normal operation pattern is required, a simulation will be performed in advance. Since the operation of the plant 100 can be performed while confirming whether the operation to be performed is appropriate, the operability is improved.

<Third embodiment>
Hereinafter, a device operation system according to a third embodiment of the present invention will be described with reference to FIGS.
The operation support apparatus 10B according to the third embodiment further has a reinforcement learning function. When a simulation is performed by the control simulation unit 18 included in the second embodiment, the presence of a control parameter to be improved may be noticed. In the present embodiment, the reinforcement learning function can improve the control parameter and the operation method.

FIG. 16 is a functional block diagram showing an example of the operation support apparatus in the third embodiment of the present invention.
Among the configurations according to the third embodiment of the present invention, the same components as those constituting the operation support apparatus 10A according to the second embodiment are denoted by the same reference numerals, and description thereof is omitted. As illustrated, the operation support apparatus 10B includes a process data collection unit 11, an input / output unit 12, a communication unit 13, an operation mode management unit 14, a storage unit 15, a prediction model construction unit 16, and an evaluation unit 17. And a control simulation unit 18 and a reinforcement learning unit 18B.
The reinforcement learning unit 18B sets an evaluation function for the result of a series of simulations by the control simulation unit 18, gives a good evaluation to a simulation suitable for a certain purpose, and gives a low evaluation to a result other than that. Thus, it has a function of learning control parameters and control processes suitable for the purpose.

FIG. 17 is a diagram for explaining reinforcement learning by the operation support apparatus according to the third embodiment of the present invention.
As shown in FIG. 17, the operation support apparatus 10B includes a reinforcement learning unit 18B. The reinforcement learning unit 18 </ b> B acquires the operating conditions and usage environment input by the control simulation unit 18, simulation control data output by the control simulation unit 18, and input / output data of the prediction models 161, 162, and 163, and stores them in the storage unit 15. . Then, learning is performed to optimize the control parameters inside the control simulation unit 18. For example, optimization indicators such as minimizing emissions, maximizing output (efficiency), minimizing fuel consumption and cost, and minimizing deviations between target values and actual values for certain control results Is given to the reinforcement learning unit 18B. Then, the reinforcement learning unit 18B learns time-series simulated process data accumulated during the simulation, and calculates a control process optimized for a given index and a control parameter for realizing the control process. The reinforcement learning unit 18B updates the control parameters stored in the control simulation unit 18 with the calculated control parameters. For example, the control simulation unit 18 says that the valve opening degree of the valve provided in the plant 100 for the first 60 seconds is 50% based on the conventional control process, and the valve opening degree is 80% for the next 90 seconds. When the control is being executed, the reinforcement learning unit 18B performs reinforcement learning, for example, setting the valve opening for the first 90 seconds to 60% and the valve opening for the next 60 seconds to 80%. It is learned that the control process is the most efficient control process when a predetermined output is obtained. Then, the reinforcement learning unit 18B sets the first “60 seconds” among the internal control parameters to “90 seconds”, the first valve opening “50%” to “60%”, and the next “90 seconds” to “60”. Update to "second". Whether or not the update is actually performed may be automatically updated, or may be updated only when the input / output unit 12 receives an update permission operation by the operator.
The operator can examine and implement the same update to the actual control device 20 with reference to the update result by the reinforcement learning unit 18B.

According to this embodiment, in addition to the effects of the first embodiment and the second embodiment, optimization of control parameters and control processes can be realized.
In addition to the update by the reinforcement learning unit 18B, the control parameters of the control simulation unit 18 may be configured to be updated from the outside by an operation instruction from an operator or the like.

<Fourth embodiment>
Hereinafter, a device operation system according to a fourth embodiment of the present invention will be described with reference to FIGS.
FIG. 18 is a functional block diagram showing an example of the operation support apparatus in the fourth embodiment of the present invention.
Among the configurations according to the fourth embodiment of the present invention, the same components as those constituting the operation support apparatus 10A according to the second embodiment are denoted by the same reference numerals, and description thereof is omitted. As illustrated, the operation support apparatus 10C includes a process data collection unit 11C, an input / output unit 12, a communication unit 13, an operation mode management unit 14, a storage unit 15, a prediction model construction unit 16, and an evaluation unit 17. And a control simulation unit 18.
The process data collection unit 11C acquires control calculation data stored in the internal memory of the control device 20 of the plant 100 and process data related to the devices 1 to 3.
The virtual plant of the plant 100 simulated by the control simulation unit 18 in the operation support device 10A of the second embodiment and the operation support device 10B of the third embodiment does not necessarily reflect the current operation state of the plant 100, and is completely independent. I was simulating the condition. On the other hand, in the operation support apparatus 10C of the fourth embodiment, the process data collection unit 11C periodically acquires process data from the plant 100 and reflects the value in the virtual plant simulated by the control simulation unit 18. The virtual environment of the plant 100 that faithfully reproduces the operation state of the plant 100 that changes from moment to moment can be realized (digital twin).

FIG. 19 is a diagram for explaining a simulation using a prediction model by the operation support apparatus in the fourth embodiment of the present invention.
As illustrated in FIG. 19, the operation support apparatus 10 </ b> C periodically tracks process data from the plant 100. Normally, control devices such as plants are duplicated, and each of the two control devices is provided with a tracking function so that the same state can be maintained. Using this tracking function, the operation support apparatus 10 </ b> C (process data collection unit 11 </ b> C) acquires process data from the plant 100. In particular, the process data collection unit 11 </ b> C according to the present embodiment is not limited to process data measured between the control device 20 and the devices 1 to 3, but is also calculated by control calculations recorded in the internal memory of the control device 20. Data can be acquired. For example, if the control logic of the control device 20 includes a first-order lag element or an integral element, these values are recorded in the internal memory. The process data collection unit 11C acquires control calculation data in the internal memory of the control device 20 by the tracking function, and the control simulation unit 18 performs control calculation using the acquired value. Thereby, the control simulation unit 18 can simulate control closer to the actual control device 20. Further, it is possible to omit the implementation of logic having a first-order lag element or the like in the control simulation unit 18.

  The control simulation unit 18 outputs the input data of the device 1 acquired by the process data collection unit 11C to the prediction model 161. Similarly, the control simulation unit 18 outputs the input data of the devices 2 and 3 acquired by the process data collection unit 11C to the prediction models 162 and 163, respectively. By periodically outputting actual process data to the prediction model 161 and the like, the control simulation unit 18 can perform a simulation closer to reality. Furthermore, when the operator inputs operation instruction information from the input / output unit 12, a simulation closer to the reality of the plant 100 for an operation to be performed in the future can be performed. During simulation for operation confirmation, reflection of the process data acquired by the process data collection unit 11C in the simulation space may be stopped.

Next, an example of the operation of the operation support apparatus 10C according to the present embodiment will be described.
FIG. 20 is a flowchart of the operation of the operation support apparatus in the fourth embodiment of the present invention.
The process similar to that in FIG. 15 will be briefly described. First, the process data collection unit 11C acquires process data of the equipment 1 and the like, and control calculation data in the internal memory of the control device 20 from the operating plant 100 (step S111). The process data collection unit 11C stores the acquired process data and the like in the storage unit 15 in association with the acquisition time (step S12). Next, the operation mode management unit 14 determines whether or not the operation mode of the operation support apparatus 10 is the “learning mode” (step S13). In the learning mode (step S13; Yes), the prediction model construction unit 16 constructs the prediction model 161 and the like (step S14). The prediction model construction unit 16 records the constructed prediction model 161 and the like in the storage unit 15. The operation in the above learning mode is the same as in the first embodiment.
Next, the operation mode management unit 14 determines whether or not to end the operation of the operation support apparatus 10C (step S17). When the process ends (step S17; Yes), the flowchart ends. When the operation is continued (step S17; No), the processing from step S11 is repeated.

  On the other hand, when the determination result in step S13 is the operation mode (step S13; No), the control simulation unit 18 transmits the information input by the operator, such as operating conditions, usage environment, and operation instructions, via the input / output unit 12. get. The control simulation unit 18 performs a control calculation based on the acquired information such as operating conditions and the internal data (control calculation data) of the control device 20 acquired in step S111 (step S132). By using the internal data of the control device 20, the control process can be simulated with high accuracy. The control simulation unit 18 calculates control data instructed to the device 1 and outputs it to the prediction model 161 and the like. Alternatively, in the case of timing immediately after the input / output data (actual process data) of the latest device 1 or the like is tracked, the latest tracked actual process data is output to the prediction model 161 or the like. Next, the prediction model 161 and the like calculate prediction data (step S15). Next, the evaluation unit 17 acquires prediction data based on the prediction model 161 and the like, and actual process data of the device 1 and the like acquired by the process data collection unit 11C. Then, the evaluation unit 17 compares the predicted data with the actual process data and evaluates the actual process data (step S162). For example, the evaluation unit 17 evaluates that the actual process data is abnormal if the difference between the predicted data and the actual process data is equal to or greater than a threshold value. Next, the operation mode management unit 14 determines whether or not the operation support device 10C has ended. If the operation mode management unit 14 does not end the operation, the processing from step S111 is repeated.

According to the present embodiment, it is possible to estimate the operation state of the plant 100 with high accuracy by performing the simulation by periodically tracking the control simulation unit 18 and the prediction model 161 and the like with the plant 100.
Apart from the description of the flowchart, the operator may perform arbitrary operation instruction information (simulated operation) on the operation support apparatus 10C to perform the state prediction simulation.

  Furthermore, a fatigue deterioration model corresponding to the usage environment and operating conditions for each device is constructed (a fatigue deterioration model created by an expert or the like) may be stored, and the fatigue deterioration model is stored in the storage unit 15 to deteriorate the device 1 or the like. You may comprise so that the simulation which considered the degree can be performed.

  Moreover, about the control simulation part 18 of 2nd embodiment-4th embodiment, about the function which emulates operation | movement of the control apparatus 20, control apparatus 20 'similar to the control apparatus 20 currently used in the plant 100 is not shown. (Actual machine) may be applied. In this case, the control simulation unit 18 simulates the exchange of process data between the control device 20 ′ and the prediction models 161 to 163. When such a usage method is applied, for example, the standby side control device (20 ′) of the control devices 20 that are duplicated in a plant that actually operates is easily connected to the operation support devices 10A to 10C. A monitoring device with a simulation function can be added. Hereinafter, application examples of the operation support apparatuses 10 to 10C of the first embodiment to the fourth embodiment will be described.

FIG. 21 is a diagram illustrating an application example of the operation support apparatus in the first to fourth embodiments of the present invention.
As shown in FIG. 21, the plant 100 is operating at the power plant 70. The plant 100 includes a control device 20, a device 1, a device 2, ..., a device n. The operator H of the plant 100 performs operation monitoring of the plant 100 while operating the terminal 30. The terminal 30 is connected to the control device 20, and the operator H inputs operation instruction information to the control device 20 through the terminal 30. Further, the operator H monitors various actual process data including measurement values detected by sensors provided in the plant 100 via the terminal 30. Further, the terminal 30 is connected to the operation support apparatus 10 (or operation support apparatuses 10A to 10D) installed in a remote place, for example, via a network.

  The operation support apparatus 10 is installed in a remote control room 60, for example. The operation support apparatus 10 is connected to the plant 100 via a network. The operation support apparatus 10 constructs the prediction models 161 to 16n in the learning mode. When the construction of the prediction models 161 to 16n is completed and the operation is performed in the operation mode, the operation support apparatus 10 acquires the process data of the plant 100 via the network, and the prediction data and the process based on the acquired process data, the prediction model 161, and the like. The state of equipment 1 etc. of plant 100 is evaluated by comparing with data. The operation support apparatus 10 transmits the evaluation result to the terminal 30 via the network. In addition, the operation support apparatus 10 outputs the evaluation result to the terminal 40. The terminal 30 displays the evaluation result by the operation support apparatus 10. The operator H performs operation monitoring of the plant 100 while confirming the evaluation result. When operation support devices 10 </ b> A to 10 </ b> C having a simulation function are installed in the control room 60 instead of the operation support device 10, the operator H inputs operation instruction information for the control simulation unit 18 to the terminal 30. The operation instruction information by the operator H is received by the communication unit 13 via the network, and the input / output unit 12 receives the operation instruction information. The control simulation unit 18 performs a simulation based on the operation instruction information. The communication unit 13 transmits the simulation result to the terminal 30. The terminal 30 displays the simulation result. The operator H can confirm the simulation result for the operation instruction made by the operator H.

  On the other hand, the operator I of the control room 60 can grasp the state of the plant 100 while referring to the evaluation result displayed on the terminal 40 while being in a remote place. Further, when there is an inquiry from the operator H, it is possible to give operational advice to the local operator H by referring to the evaluation result displayed on the terminal 40. When the operator I inputs operation instruction information for the control simulation unit 18 to the terminal 40, the control simulation unit 18 acquires the operation instruction information via the input / output unit 12 and performs a simulation. The simulation result is output to the terminal 40, and the operator I confirms the simulation result.

  In the configuration as shown in FIG. 21, the operator I is an engineer on the manufacturer side who has specialized knowledge about the plant 100, the control room 60 is a central control room provided on the manufacturer side, and the operator H is the power plant 70. Suppose you are an employee. Then, the manufacturer can monitor the state of the plant 100 by himself / herself while leaving the operation of the plant 100 which is his / her product to the delivery destination user (power plant 70). According to the operation support apparatus 10, the operator I (manufacturer side), such as process data transmitted to a computer (operation support apparatus 10) installed in the control room 60 of the company, the prediction model 161 that inputs the process data, and the like. The state of the plant 100 can be monitored by the output data. On the other hand, the operator H (user side) can refer to the evaluation result by the evaluation unit 17 transmitted by the operation support apparatus 10 without installing the operation support apparatus 10 in the power plant 70, and can be used for operation monitoring of the plant 100. it can. Further, it is not necessary for the user side to construct the prediction model 161 or the like, or to perform maintenance of the operation support apparatus 10, and it is possible to leave all the work and the like unnecessary for operation monitoring of the plant 100 to the manufacturer side.

  In addition, when any of the operation support devices 10A to 10C is installed instead of the operation support device 10 in FIG. 21, the operator I (manufacturer side) operates the operation support device 10A and the like to simulate the plant 100. be able to. For example, when there is an inquiry about an operation method from the operator H (user side), the operator I confirms that there is no problem in the operation to be performed by the simulation function provided by the operation support apparatus 10A or the like. Since it is possible to answer to the operator H whether such operation should be performed, it is effective for safe operation of the plant 100. In addition, for example, when the operator H has made an incorrect operation, the operation can be reproduced by the control simulation unit 18, so that the understanding of the phenomenon occurring in the field is deepened and appropriate support is provided. Can be provided.

  On the other hand, the operator H of the power plant 70 can also operate the actual plant 100 after confirming the result of the operation to be performed by the simulation function, so that the daily operation can be performed more appropriately. it can. Further, for example, when inquiring about a certain operation to the manufacturer, the operator H can inquire while inputting the operation instruction information to the terminal 30 and causing the control simulation unit 18 to execute the simulation. At this time, on the manufacturer side, the operator I can refer to the contents of the simulation by the operator H via the terminal 40. The operator H and the operator I can communicate with each other while mutually confirming one simulation result executed by the operation support apparatus 10A or the like. As a result, for example, an inexperienced operator H can monitor the operation of the plant 100 with peace of mind. Note that two operation support devices 10A and the like can be installed on the control room 60 side, one can be shared with the user side, and one can be dedicated to the manufacturer side. Alternatively, a plurality of operation support devices 10 (for example, 10-1 to 10-3) may be installed in the control room 60, and different plants 100 (for example, 100-1 to 100-3) may be assigned to each for monitoring. Good. Further, the control device 20 may be installed in the control room 60.

FIG. 23 is a diagram illustrating an example of a hardware configuration of the operation support apparatus according to the first to fourth embodiments of the present invention.
The computer 900 is, for example, a PC (Personal Computer) or a server terminal device including a CPU 901, a main storage device 902, an auxiliary storage device 903, an input / output interface 904, and a communication interface 905. The above-described operation support apparatuses 10 to 10C are mounted on the computer 900. The operation of each processing unit described above is stored in the auxiliary storage device 903 in the form of a program. The CPU 901 reads a program from the auxiliary storage device 903, develops it in the main storage device 902, and executes the above processing according to the program. Further, the CPU 901 ensures a storage area corresponding to the storage unit 15 in the main storage device 902 according to the program. In addition, the CPU 901 ensures a storage area for storing data being processed in the auxiliary storage device 903 according to the program.

  In at least one embodiment, the auxiliary storage device 903 is an example of a tangible medium that is not temporary. Other examples of the tangible medium that is not temporary include a magnetic disk, a magneto-optical disk, a CD-ROM, a DVD-ROM, and a semiconductor memory connected via the input / output interface 904. When this program is distributed to the computer 900 via a communication line, the computer 900 that has received the distribution may develop the program in the main storage device 902 and execute the above processing. The program may be for realizing a part of the functions described above. Further, the program may be a so-called difference file (difference program) that realizes the above-described function in combination with another program already stored in the auxiliary storage device 903.

  In addition, it is possible to appropriately replace the components in the above-described embodiments with known components without departing from the spirit of the present invention. The technical scope of the present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the spirit of the present invention. The control simulation unit 18 is an example of a control unit.

1, 2, 3 ... equipment 100 ... equipment group (plant)
10, 10A, 10B, 10C ... operation support apparatuses 11, 11C ... process data collection unit 12 ... input / output unit 13 ... communication unit 14 ... operation mode management unit 15 ... storage unit 16 ... Prediction model construction unit 17 ... Evaluation unit 18 ... Control simulation unit 18B ... Reinforcement learning unit 20 ... Control device 30, 40 ... Terminal 50, 50 '... Device operation system

Claims (16)

A process data collection unit that collects process data related to the operating state of the equipment included in the unit in one or more equipment units;
A prediction model construction unit that constructs a prediction model for each unit in which characteristics of the one or more devices are learned by neural network or deep learning based on the collected process data;
A control unit that generates control data to be input to the prediction model;
An operation support apparatus comprising: The prediction model construction unit learns the characteristic based on a combination of input data and output data for the unit for each unit of the one or a plurality of devices, and the range of the learning for the one or a plurality of devices Building the predictive model without the control process;
The operation support apparatus according to claim 1. The control unit is a control device that controls the one or more devices.
The operation support apparatus according to claim 1 or 2. The control unit is an emulator that simulates a control device that controls the one or more devices.
The operation support apparatus according to claim 1 or 2. Further comprising a predetermined physical model or statistical model indicative of characteristics of the one or more devices;
The operation support apparatus according to any one of claims 1 to 4. A reinforcement learning unit that optimizes a control parameter in the control unit or a control process by the control unit;
The operation support apparatus according to any one of claims 1 to 5. The control unit obtains calculation data generated by a control calculation by the control device from a storage unit of a control device that controls the one or more devices, and generates control data using the calculation data.
The operation support apparatus according to any one of claims 1 to 6. The prediction model includes a physical model for some of the characteristics of the one or more devices.
The operation support apparatus according to any one of claims 1 to 7. The prediction model construction unit uses a combination of input data and output data for the one or more devices as learning data, and outputs from the prediction model when the input to the prediction model is the learning data. Build a prediction model that has the same value as the data,
The operation support apparatus according to any one of claims 1 to 8. When the intermediate layer of the prediction model includes a physical model for some of the characteristics of the one or more devices, a physical model applied to the layer that compresses the characteristics of the one or more devices to a low dimension; There is an inverse function relationship with the physical model applied to the layer that restores to higher dimensions.
The operation support apparatus according to claim 9. An evaluation unit that evaluates the actual process data by comparing the predicted value of the process data based on the prediction model and the actual process data acquired from the one or more devices corresponding to the prediction model;
The operation support apparatus according to any one of claims 1 to 10, further comprising: A process data collection unit that collects process data related to the operating state of the equipment included in the unit in one or more equipment units;
A prediction model construction unit that constructs a prediction model for each unit in which characteristics of the one or more devices are learned by neural network or deep learning based on the collected process data;
An evaluation unit that evaluates the actual process data by comparing the predicted value of the process data based on the prediction model and the actual process data acquired from the device;
An operation support apparatus comprising: The operation support apparatus according to any one of claims 1 to 12,
The one or more devices and a control device for the one or more devices;
A device operation system comprising: The operation support apparatus according to any one of claims 1 to 11,
Along with the one or more devices and the control device for the one or more devices, the one or more devices are operated.
Operation method. Operation support device
Collect process data on the operating status of the equipment contained in the unit of one or more equipment,
Based on the collected process data, construct a prediction model for each unit in which the characteristics of the one or more devices are learned by neural network or deep learning,
Generating control data to be input to the prediction model;
Control method. The computer of the operation support device
Means for collecting process data relating to the operating state of equipment included in one or more equipment units;
Means for constructing a prediction model for each unit in which characteristics of the one or more devices are learned by neural network or deep learning based on the collected process data;
Means for generating control data to be input to the prediction model;
Program to function as.

Images