Generation method and device of proxy model and readable medium

Abstract

The embodiment of the invention mainly relates to the field of state monitoring, in particular to a proxy model generation method and device and a readable medium. The method comprises the following steps: establishing a simulation model of a physical quantity of an object to be monitored; determining the positions of a plurality of sensors on an object to be monitored; the plurality of sensors comprise virtual sensors and / or physical sensors; through the simulation model, obtaining a distribution diagram of multiple groups of physical quantities and physical quantity data of each sensor in the multiple sensors along with time change; inputting the distribution diagram of the multiple groups of physical quantities into a first model for training to obtain a first agent model; establishing an agent model of the sensor according to the physical quantity data of each sensor in the plurality of sensors along with the time change; and combining the proxy model of the sensor with the first proxy model to obtain a second proxy model.

Description

Method, device and readable medium for generating proxy model Technical Field

[0001] The embodiments of the present application mainly relate to the field of state monitoring, and in particular to a method, device and readable medium for generating an agent model. Background Art

[0002] Major failures in the manufacturing process can not only lead to degraded process performance but can also result in catastrophic consequences and significant economic losses. Therefore, monitoring for anomalies in the manufacturing process is crucial for the safe and reliable operation of the plant. In the process industry, a process typically includes many key variables, some of which can be difficult and expensive to measure. For example, in the hot rolling process, the temperature distribution of the steel strip is crucial to its quality. However, temperature sensors can only monitor the temperature values ​​at a limited number of points on the strip surface, which is far from sufficient for comprehensive anomaly monitoring.

[0003] Summary of the Invention

[0004] Embodiments of the present application provide a method, device, and readable medium for generating a proxy model, which are used to quickly and comprehensively predict physical quantities of relevant dynamic changes of an object to be monitored.

[0005] In a first aspect, a method for generating a proxy model is provided, comprising: establishing a simulation model of a physical quantity of an object to be monitored; determining the positions of a plurality of sensors on the object to be monitored; the plurality of sensors comprising virtual sensors and / or physical sensors; obtaining, through the simulation model, a plurality of groups of distribution diagrams of the physical quantity, and the physical quantity data of each of the plurality of sensors varying over time; inputting the plurality of groups of distribution diagrams of the physical quantity into a first model for training to obtain a first proxy model; establishing a proxy model of the sensor based on the physical quantity data of each of the plurality of sensors varying over time; and combining the proxy model of the sensor with the first proxy model to obtain a second proxy model.

[0006] In a second aspect, a device for generating an agent model is provided, comprising at least one memory and at least one processor, wherein: the at least one memory is used to store a computer program; and the at least one processor is used to call the computer program stored in the at least one memory to execute each step of the method provided in the first aspect.

[0007] According to a third aspect, a computer-readable medium is provided, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the processor executes the steps of the method provided in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The following figures are intended only to illustrate and explain the embodiments of the present application and are not intended to limit the scope of the embodiments of the present application.

[0009] FIG1 is a flow chart of a method for generating an agent model according to an embodiment of the present application;

[0010] FIG2 is a schematic diagram of a device for generating a proxy model according to an embodiment of the present application.

[0011] Description of Reference Numerals

[0012] 100: Generation method of proxy model 101-106: Method steps

[0013] 20: Proxy model generation device 21: Memory 22: Processor DETAILED DESCRIPTION

[0014] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that discussing these embodiments is merely to enable those skilled in the art to better understand and implement the subject matter described herein, and is not intended to limit the scope of protection, applicability, or examples set forth in the claims. The functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the embodiments of the present application. Various examples may omit, replace, or add various processes or components as needed. For example, the described method may be performed in an order different from the described order, and various steps may be added, omitted, or combined. In addition, the features described relative to some examples may also be combined in other examples.

[0015] As used herein, the term "including" and its variations are open terms meaning "including but not limited to". The term "based on" means "based at least in part on". The terms "one embodiment" and "an embodiment" mean "at least one embodiment". The term "another embodiment" means "at least one other embodiment". The terms "first", "second", etc. may refer to different or the same objects. Other definitions may be included below, whether explicit or implicit. Unless the context clearly indicates otherwise, the definition of a term is consistent throughout the specification.

[0016] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0017] FIG1 is a flow chart of a method for generating a proxy model according to an embodiment of the present application. As shown in FIG1 , the method 100 for generating a proxy model includes:

[0018] Step 101: Establish a simulation model of a physical quantity of an object to be monitored.

[0019] Among them, the physical quantity can be: temperature, stress, deformation, concentration or speed, etc. The user can make targeted selections according to actual needs. Optionally, before step 101, the process parameters of the object to be monitored and the range of the process parameters can also be determined first, wherein the process parameters include: material properties, material initial state, environmental parameters or process control parameters. Then, based on the process parameters and the range of the process parameters, a sample pool of process parameters is established through a sampling method, such as Latin hypercube sampling (LHS). Optionally, the ranges of some process parameters include numerical ranges, while the ranges of other process parameters include multiple options. In one scenario, assuming that the process parameters of the object to be monitored include a, b and c, which correspond to the ranges of a: [2, 5], b: {material 1, material 2, material 3}, c: [0°, 40°]. Through the sampling method, multiple process parameter values / options are extracted to obtain multiple samples. For example, the first sample: a=2, b=material 2, c=20°, the second sample: a=3, b=material 1, c=10°,... the 100th sample: a=5, b=material 3, c=30°, and the first sample to the 100th sample constitute a sample pool of process parameters.

[0020] Step 102: Determine the positions of multiple sensors on an object to be monitored, wherein the multiple sensors include virtual sensors and / or physical sensors.

[0021] In the embodiments of the present application, physical sensors may or may not be introduced. In one embodiment, the physical quantity data of the object to be monitored obtained through the simulation model that changes with time are input into the optimizer. The positions of N virtual sensors are output through the optimizer. The positions of the N virtual sensors include: the positions corresponding to the peak, valley or highest gradient value in the physical quantity, N≥1. In one embodiment, the positions of the physical sensors on the object to be monitored, the physical quantity data of the physical sensors that change with time, and the physical quantity data of the object to be monitored obtained through the simulation model that changes with time are input into the optimizer. The positions of N virtual sensors are output through the optimizer; wherein the positions of the N virtual sensors include: the positions corresponding to the peak, valley or highest gradient value in the physical quantity, N≥1. Optionally, before inputting into the optimizer, first, the optimization target of the optimizer and the weight of each optimization target are determined, wherein the optimization target includes at least one of the following:

[0022] Minimize the distance from the sensor position to the extreme point;

[0023] Maximize the gradient value of the physical quantity at the sensor position;

[0024] Maximize the distance between different sensors.

[0025] In this way, the N virtual sensors are distributed at relatively more critical positions, thus laying the foundation for the accuracy of the corresponding proxy model.

[0026] Step 103 : obtaining, through the simulation model, distribution diagrams of multiple groups of physical quantities and the physical quantity data of each of the multiple sensors that changes over time.

[0027] Optionally, the simulation model may be used to obtain the physical quantity data corresponding to each sensor of the multiple sensors changing over time and each sample in the sample pool.

[0028] Step 104 : Inputting multiple groups of distribution diagrams of the physical quantities into the first model for training to obtain a first proxy model.

[0029] Optionally, the first model is a neural network model, such as a convolutional neural network (CNN) or a variational autoencoder (VAE).

[0030] In one embodiment, a temperature simulation model of a steel strip is first established during the hot rolling process. This model can be used to generate multiple sets of temperature distribution maps, or thermograms, at different times. These thermograms are then input into a neural network model for training to generate a first proxy model, a static proxy model, that can replace the simulation model. Compared to the simulation model, the static proxy model is more convenient, requires less computing power, and produces outputs that are close to those of the simulation model.

[0031] Step 105 : establishing a proxy model of the sensor based on the physical quantity data of each sensor that changes over time.

[0032] Alternatively, a proxy model for the sensor can be established based on the physical quantity data corresponding to each sample in the sample pool and varying over time for each of the multiple sensors. In one embodiment, the physical quantity data corresponding to each sample in the sample pool and varying over time for each of the multiple sensors can be input into a neural network model and trained using a VAE method to generate the proxy model for the sensor. This approach allows for rapid reconstruction of 2D / 3D physical fields.

[0033] Optionally, the physical quantity data F corresponding to each sample in the sample pool may be obtained based on the time-varying physical quantity data of the physical sensor in the plurality of sensors. s1 ,F s2,…,F sk ,F si , establish a proxy model F of the physical sensor si (t, p)', where t represents time and p represents the value of the process parameter. Optionally, the physical quantity data F corresponding to each sample in the sample pool can be obtained based on the time-varying physical sensor in the multiple sensors. s1 ,F s2 ,…,F sk ,F si , establish a proxy model F of the physical sensor si (t,p)', where t represents time and p represents the value of the process parameter.

[0034] Step 106 : Combine the sensor's proxy model with the first proxy model to obtain a second proxy model.

[0035] Optionally, after step 106, the values ​​of the process parameters of the monitored object can be input into the second proxy model; alternatively, the time-varying physical quantity data from the physical sensors of the monitored object and the values ​​of the process parameters can be input into the second proxy model. Subsequently, the second proxy model outputs a distribution graph of the time-varying physical quantities of the monitored object, thereby achieving real-time distribution graph prediction of the relevant physical quantities of the monitored object.

[0036] Optionally, the first proxy model is represented by the following formula: static '(X)=F”(F1,F2,…,F m ,F s1 ,F s2 ,…,F sk ), where X represents a two-dimensional or three-dimensional position vector, F static '(X) represents the physical quantity distribution data. The second proxy model is represented by the following formula: F'(X,t,p)=F”(F1'(t,p),F2'(t,p),…,F m '(t,p),F s1 ,F s2 ,…,F sk ), where X represents a two-dimensional or three-dimensional position vector, and F' represents physical quantity distribution data.

[0037] In this embodiment, a static proxy model for a physical quantity of the monitored object and proxy models for multiple sensors on the monitored object are separately established. These two proxy models are then combined to create a dynamic proxy model. This dynamic proxy model enables rapid, comprehensive, and accurate real-time prediction of the distribution of the relevant physical quantity. Furthermore, by simultaneously outputting relevant training data from the simulation model to train both the static proxy model and the sensor proxy models, this not only decouples temporal and spatial variables but also significantly reduces data costs.

[0038] In one embodiment, when it is monitored that the difference between the first physical quantity data output by the proxy model of the physical sensor among the multiple sensors and the actual value of the physical sensor is greater than a preset threshold, the simulation model can be calibrated first, and then steps 103 to 106 can be performed; or, the first physical sensor can be installed at the position of the first virtual sensor among the multiple sensors, and the first physical quantity data of the first virtual sensor can be replaced with the physical quantity data of the first physical sensor. Optionally, the simulation model can be calibrated by adjusting the calculation parameters of the simulation model, such as grid density, time step or turbulence model. By establishing a proxy model of the physical sensor and comparing it with the actual value of the physical sensor in the real scene, the simulation model is calibrated to further improve the accuracy of the distribution map of the physical quantity of the monitored object that changes over time through the second proxy model proposed in the embodiment of the present application.

[0039] In one embodiment, when a physical sensor fails, the physical quantity data obtained by the proxy model of the physical sensor can replace the physical quantity data obtained by the physical sensor, thereby further ensuring the stability and flexibility of the embodiment of the present application during operation.

[0040] The present application also provides an electronic device 200. FIG2 is a schematic diagram of electronic device 200 according to an embodiment of the present application. As shown in FIG2 , electronic device 200 includes a processor 202 and a memory 201. Memory 201 stores instructions, wherein the instructions, when executed by processor 202, implement method 100 as described above.

[0041] Among them, at least one processor 202 may include a microprocessor, an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a central processing unit (CPU), a graphics processing unit (GPU), a state machine, etc. Embodiments of computer-readable media include, but are not limited to, floppy disks, CD-ROMs, magnetic disks, memory chips, ROMs, RAMs, ASICs, configured processors, all-optical media, all tapes or other magnetic media, or any other medium from which a computer processor can read instructions. In addition, various other forms of computer-readable media can send or carry instructions to a computer, including routers, private or public networks, or other wired and wireless transmission devices or channels. Instructions may include code in any computer programming language, including C, C++, C language, Visual Basic, Java, and JavaScript.

[0042] In addition, an embodiment of the present application further provides a computer-readable medium having computer-readable instructions stored thereon. When the computer-readable instructions are executed by a processor, the processor executes the aforementioned method 100. Embodiments of computer-readable media include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROMs, CD-Rs, CD-RWs, DVD-ROMs, DVD-RAMs, DVD-RWs, DVD+RWs), magnetic tapes, non-volatile memory cards, and ROMs. Optionally, the computer-readable instructions can be downloaded from a server computer or a cloud via a communication network.

[0043] It should be noted that not all steps and modules in the above processes and system structure diagrams are required, and certain steps or modules can be omitted according to actual needs. The execution order of each step is not fixed and can be adjusted as needed. The system structure described in the above embodiments can be a physical structure or a logical structure, that is, some modules may be implemented by the same physical entity, or some modules may be implemented by multiple physical entities, or may be implemented by certain components in multiple independent devices.

Claims

1. A method for generating a proxy model, characterized in that: include: - establishing (101) a simulation model of a physical quantity of an object to be monitored; - determining (102) the positions of a plurality of sensors on an object to be monitored; The plurality of sensors include: virtual sensors and / or physical sensors; - obtaining (103) a plurality of distribution diagrams of the physical quantities and the physical quantity data of each of the plurality of sensors varying over time through the simulation model; - inputting (104) the plurality of distribution maps of the physical quantities into a first model for training to obtain a first proxy model; - establishing (105) a proxy model of the sensor based on the physical quantity data of each sensor that changes over time; - Combining (106) the sensor's proxy model with the first proxy model to obtain a second proxy model.

2. The method according to claim 1, characterized in that - Before said establishing (101) a simulation model of a physical quantity of an object to be monitored, said method further comprises: - Determine the process parameters of the object to be monitored and the range of the process parameters, wherein the process parameters include: material properties, initial material state, environmental parameters or process control parameters; - establishing a sample pool of the process parameters by a sampling method according to the process parameters and the range of the process parameters; - obtaining (103) the physical quantity data of each of the plurality of sensors that changes over time through the simulation model, including: - obtaining, by means of the simulation model, physical quantity data corresponding to each sensor of the plurality of sensors changing over time and each sample in the sample pool; - establishing (105) a proxy model of the sensor based on the physical quantity data of each sensor that changes over time, including: - establishing a proxy model of the sensor according to the physical quantity data corresponding to each sensor of the plurality of sensors changing over time and each sample in the sample pool.

3. The method according to claim 1, characterized in that After obtaining the second proxy model, the method further includes: - inputting the value of the process parameter of the object to be monitored into the second agent model; or, - inputting the physical quantity data and process parameter values ​​of the physical sensor of the object to be monitored that change over time into the second agent model; - outputting a distribution diagram of the physical quantity of the object to be monitored that changes over time through the second agent model.

4. The method according to claim 2, wherein establishing a proxy model of the sensor based on the physical quantity data corresponding to each sample in the sample pool and each sensor changing over time comprises: - According to the physical quantity data F1, F2, ..., F corresponding to each sample in the sample pool and changing over time in the virtual sensor of the plurality of sensors m .F i , establish the agent model F of the virtual sensor i '(t, p), where t represents time and p represents the value of the process parameter.

5. The method according to claim 4, wherein establishing a proxy model of the sensor based on the physical quantity data corresponding to each sample in the sample pool and each sensor changing over time comprises: - According to the physical quantity data F corresponding to each sample in the sample pool and the changes over time of the physical sensors in the multiple sensors s1 ,F s2 ,…,F sk ,F si , establish the proxy model F of the physical sensor si (t, p)', where t represents time and p represents the value of the process parameter.

6. The method according to claim 5, characterized in that The first agent model is represented by the following formula: -F static '(X)=F”(F1,F2,…,F m ,F s1 ,F s2 ,…,F sk ), where X represents the position vector, F static '(X) represents the physical quantity distribution data; The second proxy model is represented by the following formula: -F'(X,t,p)=F”(F1’(t,p),F2’(t,p),…,F m '(t,p),F s1 ,F s2 ,…,F sk ), where X represents the position vector and F' represents the physical quantity distribution data.

7. The method according to claim 1, characterized in that The first model is a neural network model.

8. The method according to claim 1, characterized in that The determining (102) of the positions of a plurality of sensors on an object to be monitored comprises: - Inputting the physical quantity data of the monitored object that changes over time obtained by the simulation model into the optimizer; - Outputting the positions of N virtual sensors through the optimizer; wherein the positions of the N virtual sensors include: positions corresponding to peak values, valley values, or highest gradient values ​​in the physical quantity; and N≥1.

9. The method according to claim 1, characterized in that The determining (102) of the positions of a plurality of sensors on an object to be monitored comprises: - inputting the position of the physical sensor on the object to be monitored, the physical quantity data of the physical sensor that changes over time, and the physical quantity data of the object to be monitored that changes over time obtained by the simulation model into the optimizer; - Outputting the positions of N virtual sensors through the optimizer; wherein the positions of the N virtual sensors include: positions corresponding to peak values, valley values, or highest gradient values ​​in the physical quantity; and N≥1.

10. The method according to claim 8 or 9, characterized in that Before inputting to the optimizer, the method further includes: - Determine the optimization objectives of the optimizer and the weight of each optimization objective; wherein the optimization objectives include at least one of the following: - Minimize the distance from the sensor position to the extreme point; - Maximize the physical quantity gradient value at the sensor position; - Maximize the distance between different sensors.

11. The method according to claim 5, characterized in that When the difference between the first physical quantity data output by the proxy model of the physical sensor among the multiple sensors and the real value of the physical sensor is greater than a preset threshold, - calibrate the simulation model and then execute steps 103 to 106; or, - installing a first physical sensor at the location of a first virtual sensor among the plurality of sensors, and replacing first physical quantity data of the first virtual sensor with the physical quantity data of the first physical sensor.

12. A device for generating an agent model, characterized in that: The invention comprises at least one memory (21) and at least one processor (22), wherein: The at least one memory (21) is used to store a computer program; The at least one processor (22) is configured to call a computer program stored in the at least one memory (21) to execute the method for generating an agent model according to any one of claims 1 to 11.

13. A computer-readable storage medium having a computer program stored thereon; characterized in that: The computer program can be executed by a processor and implements the method for generating an agent model according to any one of claims 1 to 11.

Images