CN116415394A - Two-loop equipment model simulation method integrating machine learning and mechanism model - Google Patents
Two-loop equipment model simulation method integrating machine learning and mechanism model Download PDFInfo
- Publication number
- CN116415394A CN116415394A CN202111660845.XA CN202111660845A CN116415394A CN 116415394 A CN116415394 A CN 116415394A CN 202111660845 A CN202111660845 A CN 202111660845A CN 116415394 A CN116415394 A CN 116415394A
- Authority
- CN
- China
- Prior art keywords
- model
- data
- machine learning
- calculation
- unit data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 123
- 230000007246 mechanism Effects 0.000 title claims abstract description 96
- 238000010801 machine learning Methods 0.000 title claims abstract description 78
- 238000004088 simulation Methods 0.000 title claims abstract description 74
- 230000008569 process Effects 0.000 claims abstract description 74
- 238000012545 processing Methods 0.000 claims abstract description 69
- 238000013499 data model Methods 0.000 claims abstract description 66
- 238000012549 training Methods 0.000 claims abstract description 26
- 238000013507 mapping Methods 0.000 claims abstract description 13
- 238000010977 unit operation Methods 0.000 claims abstract description 8
- 238000004364 calculation method Methods 0.000 claims description 104
- 238000012360 testing method Methods 0.000 claims description 30
- 238000005259 measurement Methods 0.000 claims description 11
- 238000012795 verification Methods 0.000 claims description 10
- 238000001914 filtration Methods 0.000 claims description 8
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000002159 abnormal effect Effects 0.000 claims description 2
- 238000004140 cleaning Methods 0.000 claims 1
- 238000001514 detection method Methods 0.000 claims 1
- 238000005516 engineering process Methods 0.000 abstract description 11
- 230000010354 integration Effects 0.000 abstract description 8
- 238000013528 artificial neural network Methods 0.000 description 7
- 238000009833 condensation Methods 0.000 description 7
- 230000005494 condensation Effects 0.000 description 7
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 7
- 230000008859 change Effects 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 238000004134 energy conservation Methods 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 3
- 239000007788 liquid Substances 0.000 description 3
- 238000007637 random forest analysis Methods 0.000 description 3
- 238000012546 transfer Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000002209 hydrophobic effect Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000001502 supplementing effect Effects 0.000 description 2
- 230000032683 aging Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000015271 coagulation Effects 0.000 description 1
- 238000005345 coagulation Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000012942 design verification Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000009834 vaporization Methods 0.000 description 1
- 230000008016 vaporization Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Feedback Control In General (AREA)
Abstract
The invention belongs to the technical field of simulation, and particularly relates to a two-loop equipment model simulation method integrating machine learning and a mechanism model. The method comprises the following steps: establishing a mechanism model of the two-loop equipment, simulating, and comparing a simulation result of the mechanism model of the two-loop equipment with unit data of the current unit operation; acquiring a historical unit data set stored in a database; inputting the historical set data set into the set data processing model; training the sample set by using a machine learning method to obtain a mapping model of input parameters and output parameters of a target process, and taking the mapping model as a data model; and integrating the data model into a mechanism model of the two-loop equipment to obtain a two-loop equipment simulation model. The advantages are that: the data model and the mechanism model established by adopting the machine learning technology have very good compatibility, and the integration difficulty of the data model and the mechanism model is greatly reduced.
Description
Technical Field
The invention belongs to the technical field of simulation, and particularly relates to a two-loop equipment model simulation method integrating machine learning and a mechanism model.
Background
Modeling simulation is an important analysis method in the engineering field. In the nuclear power field, modeling simulation is a main mode for performing power plant system operation characteristic research, design verification and operator training at present. The second loop of the power plant bears the transportation and conversion of heat energy, and is one of the most critical and complex systems of the power plant. And because the implementation of the two-circuit function mainly depends on various two-circuit thermodynamic devices such as a feedwater heater, a condenser and the like, the modeling simulation of the two-circuit device becomes particularly important.
The simulation of the two-loop equipment mostly adopts a mechanism model, and the model is established by adopting a conservation equation and an empirical formula after various assumptions are introduced to simplify the physical process. However, in practice, the physical process of the device is complex, the operation condition is variable, the application range of simplifying the assumption and the empirical formula is easily exceeded, and the characteristics of the device are changed along with the aging and the maintenance of the device, so that the influence of the factors is difficult to be reflected by a mechanism model, and the simulation result is difficult to keep high consistency with the unit data. However, with the development of the current high-precision simulation technology, such as a digital twin technology, higher requirements are put on the simulation precision of the two-loop equipment, and the traditional mechanism model is often difficult to meet.
In the prior art, although a machine learning technology can be adopted to construct a data-driven equipment model, the model has the problem of difficult coupling with other process models, and is very difficult to apply. And the problems of insufficient measuring points and low sample quality existing in the power plant data are solved, and the effect is often not ideal when the power plant data are applied to an actual power plant. The fact that the measuring points are insufficient means that fewer measuring instruments are arranged in the power plant, the parameters of equipment which can be obtained are extremely limited, and the flow or pressure parameters of equipment are not available. While low sample quality refers to overrun or continuous fluctuation of data caused by measurement errors and noise interference.
Disclosure of Invention
The invention aims to provide a two-loop equipment model simulation method integrating machine learning and a mechanism model, which combines the mechanism model and the machine learning, and combines a data driving model of a key physical process established by the machine learning under the conservation principle frame of the mechanism model to finally form the two-loop equipment simulation model with high stability, high reliability and high precision. The equipment model established by the method has very good compatibility, is very easy to couple with other process models, and can identify the change of the equipment characteristics by continuously learning new operation data of the unit, thereby improving the consistency of simulation results and unit data.
The technical scheme of the invention is as follows: a two-loop equipment model simulation method integrating machine learning and mechanism models comprises the following steps:
s1, a mechanism model of the two-loop equipment is established and simulated, and a simulation result of the mechanism model of the two-loop equipment is compared with unit data of the current unit operation to determine a target process, wherein the target process is a physical process with the largest deviation between the simulation result of the mechanism model of the two-loop equipment and the unit data of the current unit operation;
s2, acquiring a historical unit data set stored in a database, and establishing a unit data processing model according to the historical unit data set, wherein the input of the unit data processing model is unit data, and the output of the unit data processing model at least comprises input parameters and output parameters of a target process;
s3, inputting the historical unit data set into the unit data processing model, and taking a processing result of the unit data processing model as a sample set;
s4, acquiring a proper machine learning method, training the sample set by using the machine learning method to obtain a mapping model of input parameters and output parameters of a target process, and taking the mapping model as a data model;
s5, integrating the data model into a mechanism model of the two-loop equipment to obtain a two-loop equipment simulation model.
The step S2 specifically includes:
acquiring a historical unit data set stored in a database, and preprocessing the historical unit data set;
determining the output of the unit data processing model according to the input parameters and the output parameters of the target process;
and establishing a first calculation model according to the preprocessed historical unit data set, verifying the accuracy of the first calculation model, and taking the first calculation model as the unit data processing model when verification passes.
The step of obtaining the historical set data set stored in the database and the step of preprocessing the historical set data set specifically comprises the following steps:
and acquiring a historical unit data set stored in a database, merging redundant measurement point data according to a preset redundant measurement point processing rule, and denoising the redundant data set by adopting a Kalman filtering method.
Establishing a first calculation model according to the preprocessed historical set data set, verifying the accuracy of the first calculation model, and taking the first calculation model as the set data processing model specifically comprises:
establishing a first calculation model according to the preprocessed historical set data set;
obtaining a simulation result of a mechanism model of the two-loop equipment, wherein the simulation result at least comprises a unit measuring point parameter, an input parameter and an output parameter of a target process;
and inputting the unit measuring point parameters into the first calculation model to obtain a calculation result, comparing the calculation result with a simulation result, and taking the first calculation model as a unit data processing model when the comparison error is smaller than a preset value, otherwise, optimizing the first calculation model according to the comparison result until the comparison result is smaller than the preset value.
The step S4 specifically includes:
acquiring a proper machine learning method;
training the sample set by using the machine learning method to obtain a second calculation model;
and verifying the second calculation model, and taking the second calculation model as a data model when the second calculation model meets the requirement.
The method for acquiring the proper machine learning specifically comprises the following steps:
training the sample set by adopting a plurality of different machine learning methods to obtain a plurality of machine models, wherein the sample set comprises a training set and a testing set;
and testing a plurality of machine models by adopting the test set to obtain test parameters so as to select a proper machine learning method according to the test parameters.
Verifying the second calculation model, and when the second calculation model meets the requirement, taking the second calculation model as a data model specifically comprises:
and verifying the second calculation model by adopting the test set, judging whether the calculation result of the second calculation model meets the requirement, if so, taking the second calculation model as a data model, and otherwise, adjusting the parameters of the second calculation model until the calculation result of the second calculation model meets the requirement.
The invention has the beneficial effects that: the invention analyzes a two-loop equipment modeling simulation method, combines a mechanism simulation technology and a machine learning technology, and provides a two-loop equipment model simulation method integrating machine learning and a mechanism model. The specific advantages are as follows:
(1) The data model and the mechanism model established by adopting the machine learning technology have very good compatibility, and the integration difficulty of the data model and the mechanism model is greatly reduced.
(2) The novel high-precision model is formed by skillfully fusing a mechanism model and a machine learning technology, and is different from simple machine learning and simple mechanism model, but mutually supplementing each other. Wherein the mechanism model enhances the effect of machine learning. On the one hand, the mechanism model can identify abnormal unit data, and a reliable sample can be provided for machine learning. On the other hand, compared with the unit data, the mechanism model simulation result is used as a sample, and the method has the characteristics of reliable data, large sample size and wide distribution range, and can accurately evaluate the learning ability of various machine learning methods. Similarly, the machine learning can identify the change of the actual power plant equipment characteristics, and short plates of the mechanism model are compensated.
(3) The integration of the data model and the mechanism model is realized by adopting a database sharing mode, so that the online learning of the data model is supported, and the flow of replacing the data model is simplified.
Drawings
FIG. 1 is a flow chart for building a data model using machine learning techniques;
FIG. 2 is a schematic diagram of the set-up of a data processing model of a unit;
FIG. 3 is a simulation result;
FIG. 4 is a schematic diagram of the integration of a mechanism model with a data model;
FIG. 5 shows the distribution of measuring points of a six-high-pressure feedwater heater of a certain power plant;
fig. 6 shows the kalman filter effect.
Detailed Description
The invention will be described in further detail with reference to the accompanying drawings and specific examples.
The invention provides a two-loop equipment model simulation method integrating machine learning and a mechanism model, which comprises the following steps:
s1, establishing a mechanism model of two-loop equipment and simulating, and comparing a simulation result of the mechanism model of the two-loop equipment with unit data of the current unit operation to determine a target process, wherein the target process is a physical process with the largest deviation between the simulation result of the mechanism model of the two-loop equipment and the unit data of the current unit operation;
s2, acquiring a historical unit data set stored in a database, and establishing a unit data processing model according to the historical unit data set, wherein the input of the unit data processing model is unit data, and the output of the unit data processing model at least comprises input parameters and output parameters of a target process;
s3, inputting the historical unit data set into the unit data processing model, and taking a processing result of the unit data processing model as a sample set;
s4, acquiring a proper machine learning method, training the sample set by using the machine learning method to obtain a mapping model of input parameters and output parameters of a target process, and taking the mapping model as a data model;
and S5, integrating the data model into a mechanism model of the two-loop equipment to obtain a two-loop equipment simulation model.
In this embodiment, a mechanism model of the two-loop device is first established, a simulation result of the mechanism model is compared with unit data, after a target process is determined, a unit data processing model is established according to input parameters and output parameters of the target process and in combination with historical unit data, then a historical unit data set is processed by using the unit data processing model, the processing result is used as a training sample of a subsequent machine learning method, then a proper machine learning method is selected to train the training sample, and after a data model is obtained, the data model is integrated into the mechanism model, so that the two-loop device simulation model is obtained. In the embodiment, the data model and the mechanism model established by adopting the machine learning technology have very good compatibility, and the integration difficulty of the data model and the mechanism model is greatly reduced. The novel high-precision model is formed by skillfully fusing a mechanism model and a machine learning technology, and is different from simple machine learning and simple mechanism model, but mutually supplementing each other. Wherein the mechanism model enhances the effect of machine learning. Similarly, the machine learning can identify the change of the actual power plant equipment characteristics, and short plates of the mechanism model are compensated.
In a preferred embodiment, in the step S1, a physical process with the largest deviation is determined by comparing the model simulation result with the unit data of the current unit operation, where the physical process is the target process, and the theoretical model part of the target process included in the device mechanism model is simply referred to as a low-precision mechanism model. For example, in a specific embodiment, a mechanism model of the high-pressure feedwater heater is established, and by comparing the simulation result of the model with the unit data, it is determined that the short plate with the largest mechanism model is in the condensation heat exchange process of shell-side steam, so this process is taken as a target process. The target process in the mechanism model is proxied by the data model established by utilizing the machine learning technology, so that the simulation precision is improved.
How to determine whether the deviation is the largest is relevant to the actual situation. For example, for high pressure feedwater heaters, the parameters of interest are outlet temperature and water level, and the heat exchange process (which determines the water level and outlet temperature) can severely affect both parameters. If the outlet temperature deviation is larger, the error is obviously caused in the heat exchange process, and the heat exchange process comprises a two-phase heat exchange process and a single-phase heat exchange process, wherein the single-phase heat exchange calculation principle is already mature (related research shows that the accuracy of the single-phase heat exchange mechanism is far higher than that of the two-phase heat exchange mechanism), so that the process with the largest deviation or the most necessary optimization is the two-phase heat exchange mechanism, namely the target process is the two-phase heat exchange mechanism process.
Since the actual crew data does not contain the input and output parameters of the data model (obtained by machine learning techniques), a "crew data processing model" must be built to process the crew data so that a sample library for machine learning can be obtained. The purpose of the step S2 is to establish a unit data processing model, and the unit data processing model is to process unit data to obtain a training sample of a subsequent machine learning method, where the training sample corresponds to the input and output of the target process, so that the input and output of the data model obtained by training the machine learning method are similar to the input and output of the target process, and the data model can replace the target process to complete simulation, thereby achieving the purpose of improving simulation precision and reducing the integration difficulty of the data model and the machine model.
In a preferred embodiment, the step S2 specifically includes:
acquiring a historical unit data set stored in a database, and preprocessing the historical unit data set;
determining the output of the unit data processing model according to the input parameters and the output parameters of the target process;
and establishing a first calculation model according to the preprocessed historical unit data set, verifying the accuracy of the first calculation model, and taking the first calculation model as the unit data processing model when verification passes.
In this embodiment, the database is a pre-established historical unit database, and when the unit operates, operation data of different time periods of the unit are stored in the database, so that subsequent analysis is facilitated. When a historical unit data set is obtained, in order to avoid that unreasonable data in the historical unit data set affects the establishment of a final model, the historical unit data set needs to be preprocessed firstly, specifically, the historical unit data set stored in the database is obtained, and preprocessing the historical unit data set specifically includes:
and acquiring a historical unit data set stored in a database, merging redundant measurement point data according to a preset redundant measurement point processing rule, and denoising the redundant data set by adopting a Kalman filtering method.
Specifically, after a historical unit data set is obtained from a database, the redundant measurement point data are combined according to the processing rule of a unit control system on the redundant measurement points, and then the unit data are processed by adopting a Kalman filtering method to remove the influence of noise and interference. The preset redundant point processing rule is determined according to actual requirements, for example, in some embodiments, the redundant point processing rule is a processing rule that "dead pixels are removed when the two-by-two difference values are greater than 5%, and the remaining points are averaged", and after the redundant points are removed, the Kalman filtering method can be used for denoising to remove the influence of noise and interference. The kalman filtering method is the prior art, and is not described herein.
After the historical unit data set is processed, a unit data processing model can be built according to the historical unit data set, before the historical unit data set is built, input and output of the unit data processing model are required to be determined first, and in the embodiment, the input and output of the unit data processing model are determined according to input parameters and output parameters of a target process. The processing has the advantages that the input and output of the data model are similar to the target process, and the integration difficulty of the data model and the mechanism model is greatly reduced.
Further, when the first calculation model is built, the energy conservation principle and other mechanism models with higher reliability (for example, the single-phase heat exchange mechanism is higher in precision than the two-phase heat exchange mechanism, and the single-phase heat exchange mechanism can be used for building the first calculation model) are adopted to build the first calculation model, for example, in a specific embodiment, the first calculation model is built based on the energy conservation principle and the hydrophobic region convection heat exchange mechanism model, the input of the model is measurement point data of an actual unit, the output of the model comprises input and output parameters of a data model, and of course, the output of the first calculation model is not limited to the input and output parameters of the data model, but also comprises other intermediate parameters, and the specific construction process of the first calculation model depends on the specific equipment type. Specifically, the establishing a first calculation model according to the preprocessed historical set data set, verifying the accuracy of the first calculation model, and when verification passes, taking the first calculation model as the set data processing model specifically includes:
establishing a first calculation model according to the preprocessed historical set data set;
obtaining a simulation result of a mechanism model of the two-loop equipment, wherein the simulation result at least comprises a unit measuring point parameter, an input parameter and an output parameter of a target process;
and inputting the unit measuring point parameters into the first calculation model to obtain a calculation result, comparing the calculation result with a simulation result, and taking the first calculation model as a unit data processing model when the comparison error is smaller than a preset value, otherwise, optimizing the first calculation model according to the comparison result until the comparison result is smaller than the preset value.
In this embodiment, a first calculation model is first built, the output parameters of the first calculation model include the input and output parameters of the data model (i.e., the input parameters and the output parameters of the target process), then the accuracy of the first calculation model is verified, and if the requirement is not met, the first calculation model is optimized. The specific optimization process is different according to actual requirements, and in some embodiments, parameters in the first calculation model can be directly optimized according to the comparison result, so that the calculation result meets the requirements.
In this embodiment, first, a simulation result of a mechanism model of the two-loop device is extracted, and after the simulation result is processed by adopting a first calculation model, the accuracy of the first calculation model is verified by comparing errors of the calculation result and the simulation result, and if the requirement is not met, the first calculation model is optimized according to the comparison result.
Specifically, since the output parameters of the unit data processing model are some intermediate parameters during the operation of the unit, the details can be seen in fig. 2, and the simulation results include these parameters (e.g., fig. 3). Therefore, the output parameters of the simulation result (the same parameters as the unit data, such as pressure and temperature) can be used as the input parameters of the unit data processing model, the processing result (corresponding to the three boxes on the right side of fig. 2) can be obtained, and then the processing result and the simulation result can be compared.
Furthermore, in this embodiment, since the unit data does not include the input parameters and the output parameters of the target process, the simulation result is used to verify the unit data processing model. Since the basic energy conservation and a small amount of necessary mechanism models with higher reliability form a 'unit data processing model' (a mechanism model which cannot contain a target process), the relation between the input parameters of the 'data model' and the output parameters of the 'data model' is not known, but the two data are obtained directly according to the unit data at the same time, so that a simulation result is needed to verify.
In a preferred embodiment, in the step S3, after the unit data processing model is obtained, the processing of the historical unit data set may be implemented by the unit data processing model, and since the input of the unit data processing model is unit data, the output includes the input parameter and the output parameter of the target process, and thus, when the historical unit data set is input, a plurality of paired input parameters and output parameters may be obtained, and the plurality of paired input parameters and output parameters may form a sample set of the subsequent machine learning method, and may be used as a sample library for machine learning.
In addition, it should be noted that, when the set data sample is unevenly distributed or the error is too large, it is difficult to comprehensively evaluate the learning ability of the various machine learning methods on the physical characteristics of the target process. In order to solve the problem, the low-precision mechanism model is considered to be similar to the actual physical process in terms of the change characteristics although the precision is insufficient, so that the low-precision mechanism model is adopted to generate data to supplement the sample library, thereby avoiding evaluation deviation caused by poor sample quality and utilizing the unit data with poor quality in the sample library.
In a preferred embodiment, in step S4, in order to ensure accuracy of the result, a suitable machine learning method is required to be acquired first, and then a sample set is trained by adopting the method, so as to obtain a mapping model, the mapping model reflects a mapping relationship between an input parameter and an output parameter of a target process, and the output parameter calculated by the mapping model can solve the problem of larger error of the output result of the target process, so that accuracy of a simulation result of the whole mechanism model is further ensured by combining with the mechanism model. Specifically, the step S4 specifically includes:
acquiring a proper machine learning method;
training the sample set by using the machine learning method to obtain a second calculation model;
and verifying the second calculation model, and taking the second calculation model as a data model when the second calculation model meets the requirement.
In this embodiment, the method for obtaining suitable machine learning specifically includes:
training the sample set by adopting a plurality of different machine learning methods to obtain a plurality of machine models, wherein the sample set comprises a training set and a testing set;
and testing a plurality of machine models by adopting the test set to obtain test parameters so as to select a proper machine learning method according to the test parameters.
In this embodiment, the training set is first trained using a number of different machine learning methods, including but not limited to: linear regression method, decision tree regression method, random forest regression method, gradient enhanced random forest regression method, bagging regression method, K neighbor regression method, support vector machine regression method, adaboost regression method, lightGBM regression method, GBDT regression method, XGB model algorithm, MLP neural network algorithm and the like, so that a plurality of machine models can be obtained, then testing is carried out by using a testing set to obtain testing parameters, wherein the testing parameters at least comprise correlation coefficients, RMSE and R 2 The user can then select a suitable machine learning method from the test parameters, and the selection process is selected by the user independently, which is not limited in the application, for example, selecting an MLP neural network algorithm.
After the machine learning method is selected, training a sample set by adopting the machine learning method to obtain a second calculation model, verifying the second calculation model by utilizing a test set in order to ensure the calculation accuracy of the second calculation model, specifically, reserving data in a certain range by the sample set for more comprehensively testing the learning capacity of the machine learning method without using the model training, wherein the data in the certain range is called a verification set, and the rest samples are divided into a training set and a test set by adopting a cross verification method. In order to ensure accuracy of data model calculation, a second calculation model needs to be verified, specifically, the second calculation model is verified, and when the second calculation model meets requirements, the second calculation model is taken as the data model, and specifically includes:
and verifying the second calculation model by adopting the test set, judging whether the calculation result of the second calculation model meets the requirement, if so, taking the second calculation model as a data model, and otherwise, adjusting the parameters of the second calculation model until the calculation result of the second calculation model meets the requirement.
In this embodiment, when the second calculation model does not meet the requirement, an attempt is made to adjust the parameter setting of the second calculation model, for example, for the neural network algorithm, the number of nodes and the number of layers of the neural network may be adjusted, and the learning rate may be adjusted according to the change process of the loss function. If the prediction capability of the second calculation model can meet the requirement, using the second calculation model as a data model; otherwise, the extended sample data is trained by reusing the machine learning algorithm, and the data source is not limited to the set data, but also comprises experimental data or mechanism model simulation data.
After the data model is built, the input of the data model is the same as the input of the target process, and the output of the data model can replace the output of the target process, so that the problem of inaccurate simulation results caused by errors of the output of the target process is avoided. Therefore, after the data model and the mechanism model are integrated, the obtained simulation model has extremely high simulation capability, and the output data is relatively similar to the real data. In this embodiment, as shown in fig. 4, the data model and the mechanism model are integrated by sharing a database. And providing input parameters for the data driving model by using the mechanism model by using the shared database as a data transmission medium, and returning a prediction result to the mechanism model by using the data model to form an organic whole. The integration mode has the advantages of supporting online learning of the data model and simplifying the flow of replacing the data model.
In a preferred embodiment, the step S5 further includes:
updating the database, and optimizing the data model according to the historical set data set in the updated database.
In this embodiment, the latest unit data may be used as samples for online learning, or a new data model may be rebuilt based on the samples to replace the old model, so as to ensure that the model timely identifies the change of the actual power plant equipment characteristics.
For a better understanding of the present invention, the following detailed description of the technical solution of the present invention is given with reference to fig. 1 to 6, which illustrate an embodiment of the present invention:
the following describes the implementation of the present invention using a typical two-circuit device, such as a high-pressure feedwater heater:
step 1: and (3) establishing a mechanism model of the high-pressure feed water heater, and determining the condensation heat exchange process of the steam on the shell side of the short plate with the largest mechanism model by comparing and analyzing a model simulation result and unit data, so that the process is taken as a target process.
Step 2: as shown in fig. 1, a machine learning technique is used to build a data model of the target process.
Step 2.1, selecting input and output parameters of a data model:
the condensation heat exchange coefficient is a key characteristic parameter of the condensation heat exchange process, and is calculated in a mechanism model by adopting the following formula:
referring to the above, select a coagulation exchangeCoefficient of heat h con Selecting vaporization latent heat r and liquid film density rho for outputting data model l Coefficient of thermal conductivity lambda of liquid film l Dynamic viscosity eta of liquid film l External diameter d of heat transfer tube and steam temperature t s Heat transfer tube wall temperature t w And temperature difference (t) s -t w ) As input parameters.
Step 2.2 preconditioner group data:
fig. 5 shows the distribution of measuring points of a high-pressure feedwater heater in a certain power plant, and actual operation data of the feedwater heater, such as the temperature of a drainage outlet of the feedwater heater, the water level at the shell side, the inlet and outlet pressure and temperature of a heat transfer pipe, the opening of a valve on a drainage outlet pipeline, the pressure of a deaerator and a condenser and the like, are obtained from a database.
And then according to the processing rule of the DCS system for removing bad points when the two-by-two difference values of the redundant measuring points are larger than 5%, and taking the average value of the residual points, processing redundant pressure and water level measuring points.
Finally, the unit data is processed by adopting a Kalman filtering method to remove the influence of noise and interference, and FIG. 6 shows the filtering effect of the outlet pressure of the tube side of the high-pressure feedwater heater.
Step 2.3, building and verifying a unit data processing model:
as shown in FIG. 2, a unit data processing model is established based on an energy conservation principle and a hydrophobic region convection heat exchange mechanism model, wherein the input of the model is measurement point data of an actual unit, and the output of the model comprises input and output parameters of a data model.
And then extracting simulation results of the mechanism model of the high-pressure feed water heater, wherein the extracted parameters comprise unit measuring point parameters, input and output parameters of a data model and other key intermediate parameters such as shell side saturation temperature and condensation heat exchange quantity.
Finally, the unit data processing model is adopted to process the extracted simulation result, and the error of the processing result and the simulation result is compared and is far less than 1%, so that the accuracy of the unit data processing model can meet the requirement.
Step 2.4, processing the unit data:
and processing the unit data by using the established unit data processing model to obtain input and output data samples of the data model, wherein the input and output data samples can be used as a sample library for machine learning. Under the low working condition, the high-pressure feed water heater is not operated sometimes, and the processing of the unit data at the moment can cause the overrun of the calculated result and the overrun data should be removed.
Step 2.5 screening machine learning method:
and (3) taking the unit data processed in the step 2.4 as a sample library for screening the machine learning method.
In order to more comprehensively test the learning capability of the machine learning method, a certain range of data is reserved and not used for model training, the data of the part is called a verification set, and the rest samples are divided into a training set and a testing set by adopting a cross verification method. And then, establishing an input-to-output mapping model by adopting more than ten machine learning methods such as random forests, neural networks and the like, evaluating the generalization capability of the model by adopting a test set and a verification set, wherein the evaluation indexes are correlation coefficients, RMSE and R2, and the evaluation result shows that the neural network method has very strong learning capability for the condensation heat exchange process of the feedwater heater.
Step 2.6, establishing a data model:
the sample library is divided into a training set and a testing set by adopting a cross-validation method, a data model is built by adopting the neural network method screened in the step 2.5, the built model comprises 3 hidden layers, 2048 nodes in each layer, the learning rate is 0.02, and the output parameter is condensation heat exchange thermal resistance (namely 1/hcon). The model predictive capability can meet the requirements and step 3 is entered.
Step 3: integrating the data model established in the step 2 into a mechanism model.
As shown in fig. 4, the data model and the mechanism model are integrated in a manner of sharing a database. And providing input parameters for the data model by using the mechanism model through the shared database as a data transmission medium, and returning a prediction result to the mechanism model by using the data model to form a high-precision intelligent high-pressure feedwater heater model.
Claims (7)
1. A two-loop equipment model simulation method integrating machine learning and mechanism models is characterized by comprising the following steps:
s1, a mechanism model of the two-loop equipment is established and simulated, and a simulation result of the mechanism model of the two-loop equipment is compared with unit data of the current unit operation to determine a target process, wherein the target process is a physical process with the largest deviation between the simulation result of the mechanism model of the two-loop equipment and the unit data of the current unit operation;
s2, acquiring a historical unit data set stored in a database, and establishing a unit data processing model according to the historical unit data set, wherein the input of the unit data processing model is unit data, and the output of the unit data processing model at least comprises input parameters and output parameters of a target process;
s3, inputting the historical unit data set into the unit data processing model, and taking a processing result of the unit data processing model as a sample set;
s4, acquiring a proper machine learning method, training the sample set by using the machine learning method to obtain a mapping model of input parameters and output parameters of a target process, and taking the mapping model as a data model;
s5, integrating the data model into a mechanism model of the two-loop equipment to obtain a two-loop equipment simulation model.
2. The two-loop equipment model simulation method integrating machine learning and mechanism models according to claim 1, wherein the step S2 specifically includes:
acquiring a historical unit data set stored in a database, and preprocessing the historical unit data set;
determining the output of the unit data processing model according to the input parameters and the output parameters of the target process;
and establishing a first calculation model according to the preprocessed historical unit data set, verifying the accuracy of the first calculation model, and taking the first calculation model as the unit data processing model when verification passes.
3. The two-loop equipment model simulation method integrating machine learning and mechanism models according to claim 1, wherein the step of obtaining a historical set data set stored in a database and the step of preprocessing the historical set data set specifically comprises the following steps:
after the historical set data set stored in the database is obtained, null values are filled firstly based on power plant data acquisition and storage rules, abnormal value detection and data cleaning are carried out, then redundant measurement point data are combined according to preset redundant measurement point processing rules, and finally, a Kalman filtering method is adopted to carry out denoising treatment on the redundant set of data.
4. The two-loop equipment model simulation method integrating machine learning and mechanism models according to claim 1, wherein the method is characterized in that a first calculation model is built according to the preprocessed historical set data set, the accuracy of the first calculation model is verified, and when verification is passed, the method specifically comprises the steps of:
establishing a first calculation model according to the preprocessed historical set data set;
obtaining a simulation result of a mechanism model of the two-loop equipment, wherein the simulation result at least comprises a unit measuring point parameter, an input parameter and an output parameter of a target process;
and inputting the unit measuring point parameters into the first calculation model to obtain a calculation result, comparing the calculation result with a simulation result, and taking the first calculation model as a unit data processing model when the comparison error is smaller than a preset value, otherwise, optimizing the first calculation model according to the comparison result until the comparison result is smaller than the preset value.
5. The two-loop equipment model simulation method integrating machine learning and mechanism models according to claim 1, wherein the step S4 specifically includes:
acquiring a proper machine learning method;
training the sample set by using the machine learning method to obtain a second calculation model;
and verifying the second calculation model, and taking the second calculation model as a data model when the second calculation model meets the requirement.
6. The two-loop equipment model simulation method integrating machine learning and mechanism models according to claim 1, wherein the method for acquiring the proper machine learning specifically comprises the following steps:
training the sample set by adopting a plurality of different machine learning methods to obtain a plurality of machine models, wherein the sample set comprises a training set and a testing set;
and testing a plurality of machine models by adopting the test set to obtain test parameters so as to select a proper machine learning method according to the test parameters.
7. The two-loop equipment model simulation method integrating machine learning and mechanism models according to claim 1, wherein the second calculation model is verified, and when the second calculation model meets the requirement, the second calculation model is taken as a data model specifically comprising:
and verifying the second calculation model by adopting the test set, judging whether the calculation result of the second calculation model meets the requirement, if so, taking the second calculation model as a data model, and otherwise, adjusting the parameters of the second calculation model until the calculation result of the second calculation model meets the requirement.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111660845.XA CN116415394A (en) | 2021-12-31 | 2021-12-31 | Two-loop equipment model simulation method integrating machine learning and mechanism model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111660845.XA CN116415394A (en) | 2021-12-31 | 2021-12-31 | Two-loop equipment model simulation method integrating machine learning and mechanism model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116415394A true CN116415394A (en) | 2023-07-11 |
Family
ID=87055021
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111660845.XA Pending CN116415394A (en) | 2021-12-31 | 2021-12-31 | Two-loop equipment model simulation method integrating machine learning and mechanism model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116415394A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117131708A (en) * | 2023-10-26 | 2023-11-28 | 中核控制系统工程有限公司 | Modeling method and application of digital twin anti-seismic mechanism model of nuclear industry DCS equipment |
| CN117521528A (en) * | 2024-01-03 | 2024-02-06 | 中国核动力研究设计院 | Turbine equipment simulation model evolution method, device, medium and computing equipment |
-
2021
- 2021-12-31 CN CN202111660845.XA patent/CN116415394A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117131708A (en) * | 2023-10-26 | 2023-11-28 | 中核控制系统工程有限公司 | Modeling method and application of digital twin anti-seismic mechanism model of nuclear industry DCS equipment |
| CN117131708B (en) * | 2023-10-26 | 2024-01-16 | 中核控制系统工程有限公司 | Modeling method and application of digital twin anti-seismic mechanism model of nuclear industry DCS equipment |
| CN117521528A (en) * | 2024-01-03 | 2024-02-06 | 中国核动力研究设计院 | Turbine equipment simulation model evolution method, device, medium and computing equipment |
| CN117521528B (en) * | 2024-01-03 | 2024-03-15 | 中国核动力研究设计院 | Turbine equipment simulation model evolution method, device, medium and computing equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116415394A (en) | Two-loop equipment model simulation method integrating machine learning and mechanism model | |
| CN112818595B (en) | Correction method and system for digital twin model data of evaporation area of thermal power plant | |
| CN107480341A (en) | A kind of dam safety comprehensive method based on deep learning | |
| CN108803520A (en) | A kind of dynamic process monitoring method rejected based on the non-linear autocorrelation of variable | |
| CN103792933A (en) | Method for determining and tuning process characteristic parameters using a simulation system | |
| CN109784348A (en) | A kind of infrared power equipment identification and inline diagnosis method and its system | |
| CN109948863A (en) | Drainage pipeline networks inspection shaft liquid level prediction technique based on shot and long term memory models LSTM | |
| CN102880905A (en) | Online soft measurement method for normal oil dry point | |
| CN115034129B (en) | NOx emission concentration soft measurement method for thermal power plant denitration device | |
| CN107590351A (en) | A kind of centrifugal pump lift specificity analysis and the method for prediction | |
| Abdul-Wahab et al. | Optimization of multistage flash desalination process by using a two-level factorial design | |
| CN108830417A (en) | A kind of residential energy consumption prediction technique and system based on ARMA and regression analysis | |
| CN112113146B (en) | Synchronous self-adaptive check method for roughness coefficient and node water demand of water supply pipe network pipeline | |
| CN113268883A (en) | Method for predicting corrosion rate of submarine crude oil pipeline based on PCA-ABC-SVM model | |
| CN115935834A (en) | History fitting method based on deep autoregressive network and continuous learning strategy | |
| CN107885967A (en) | A kind of regression model hyperparameter optimization method | |
| Merino et al. | Real Time Optimization for steam management in an evaporation section | |
| CN117189713A (en) | Hydraulic system fault diagnosis method based on digital twin driving | |
| CN114912364A (en) | Natural gas well flow prediction method, device, equipment and computer readable medium | |
| CN109491709A (en) | A kind of software code degree of controllability integrated evaluating method based on AHP and neural network | |
| Wu et al. | Temporal convolution network‐based time frequency domain integrated model of multiple arch dam deformation and quantification of the load impact | |
| CN109920489A (en) | It is a kind of that model and method for building up are hydrocracked based on Lasso-CCF-CNN | |
| CN108572639A (en) | A kind of dynamic process monitoring method rejected based on principal component autocorrelation | |
| CN117252374A (en) | Multi-layer production yield distribution method for tight gas-combined production well based on machine learning | |
| CN116662925A (en) | Industrial process soft measurement method based on weighted sparse neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |