A secondary circuit equipment model simulation method integrating machine learning and mechanism model

Abstract

The present invention belongs to the field of simulation technology, and specifically relates to a secondary circuit equipment model simulation method that integrates machine learning and mechanism models. The method comprises the following steps: establishing a mechanism model of a secondary circuit equipment and performing simulation, comparing the simulation result of the mechanism model of the secondary circuit equipment with the unit data of the current unit operation; obtaining a historical unit data set stored in a database; inputting the historical unit data set into the unit data processing model; training the sample set using a machine learning method to obtain a mapping model of the input parameters and output parameters of the target process, and using the mapping model as a data model; integrating the data model into the mechanism model of the secondary circuit equipment to obtain a simulation model of the secondary circuit equipment. The method has the advantages that the data model established using machine learning technology has very good compatibility with the mechanism model, which greatly reduces the difficulty of integrating the data model with the mechanism model.

Description

Technical Field

[0001] The present invention belongs to the field of simulation technology, and in particular relates to a secondary circuit equipment model simulation method integrating machine learning and mechanism model. Background Art

[0002] Modeling and simulation are important analytical methods in the field of engineering. In the field of nuclear power, modeling and simulation are the main methods for studying the operating characteristics of power plant systems, design verification, and operator training. The secondary circuit of a power plant is responsible for the transportation and conversion of thermal energy and is one of the most critical and complex systems in a power plant. Because the realization of the secondary circuit function mainly depends on various secondary circuit thermal equipment such as feedwater heaters and condensers, modeling and simulation of secondary circuit equipment becomes particularly important.

[0003] The simulation of secondary circuit equipment mostly adopts mechanism model, which is to establish the simulation model by introducing various assumptions to simplify the physical process and using conservation equations and empirical formulas. However, in reality, the physical process of the equipment is complex and the operating conditions are changeable, which can easily exceed the scope of application of simplified assumptions and empirical formulas. Moreover, as the equipment ages and is repaired, its characteristics will change. The mechanism model is difficult to reflect the influence of these factors, so it is difficult for the simulation results to maintain a high degree of consistency with the unit data. However, with the development of current high-precision simulation technologies such as digital twin technology, higher requirements are placed on the simulation accuracy of secondary circuit equipment, which traditional mechanism models often cannot meet.

[0004] In existing technologies, although machine learning technology can be used to construct data-driven equipment models, this model is difficult to couple with other process models and is very difficult to apply. Moreover, due to the problems of insufficient measurement points and low sample quality in power plant data, the effect is often not ideal when applied to actual power plants. Insufficient measurement points refers to the fact that there are few measuring instruments arranged in the power plant, and the equipment parameters that can be obtained are extremely limited, often even the flow or pressure parameters of the equipment are not available. Low sample quality refers to the fact that the data exceeds the limit or continues to fluctuate due to measurement errors and noise interference. Summary of the invention

[0005] The purpose of the present invention is to provide a secondary circuit equipment model simulation method that integrates machine learning and mechanism model, combines mechanism model and machine learning, and under the framework of the conservation principle of the mechanism model, combines the data-driven model of the key physical process established by machine learning, and finally forms a secondary circuit equipment simulation model with high stability, high reliability and high precision. The equipment model established by this method has very good compatibility, is very easy to couple with other process models, and can also continuously learn new operating data of the unit, identify changes in equipment characteristics, and improve the consistency of simulation results with unit data.

[0006] The technical solution of the present invention is as follows: A secondary circuit equipment model simulation method integrating machine learning and mechanism model comprises the following steps:

[0007] S1: Establishing a mechanism model of the secondary circuit equipment and performing simulation, comparing the simulation result of the mechanism model of the secondary circuit equipment with the unit data of the current unit operation to determine the target process, wherein the target process is the physical process with the largest deviation between the simulation result of the mechanism model of the secondary circuit equipment and the unit data of the current unit operation;

[0008] 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 includes at least input parameters and output parameters of a target process;

[0009] S3: inputting the historical unit data set into the unit data processing model, and taking the processing result of the unit data processing model as a sample set;

[0010] S4: obtaining a suitable machine learning method, using the machine learning method to train the sample set to obtain a mapping model of input parameters and output parameters of the target process, and using the mapping model as a data model;

[0011] S5 integrates the data model into the mechanism model of the secondary circuit equipment to obtain a simulation model of the secondary circuit equipment.

[0012] The step S2 specifically includes:

[0013] Acquire a historical unit data set stored in a database, and preprocess the historical unit data set;

[0014] Determining the output of the unit data processing model according to the input parameters and output parameters of the target process;

[0015] A first calculation model is established according to the preprocessed historical unit data set, and the accuracy of the first calculation model is verified. When the verification passes, the first calculation model is used as the unit data processing model.

[0016] The acquiring of the historical unit data set stored in the database and preprocessing of the historical unit data set specifically include:

[0017] The historical unit data set stored in the database is obtained, and after merging the redundant measurement point data according to the preset redundant measurement point processing rules, the Kalman filtering method is used to denoise the redundantly processed data set.

[0018] Establishing a first calculation model according to the preprocessed historical unit data set, and verifying the accuracy of the first calculation model, and when the verification passes, using the first calculation model as the unit data processing model specifically includes:

[0019] Establishing a first calculation model according to the preprocessed historical unit data set;

[0020] Acquire simulation results of the mechanism model of the secondary circuit equipment, wherein the simulation results at least include unit measurement point parameters, input parameters and output parameters of the target process;

[0021] The unit measurement point parameters are input into the first calculation model to obtain calculation results, and the calculation results are compared with the simulation results. When the comparison error is less than a preset value, the first calculation model is used as the unit data processing model; otherwise, the first calculation model is optimized according to the comparison result until the comparison result is less than a preset value.

[0022] The step S4 specifically includes:

[0023] Get the right machine learning methods;

[0024] Using the machine learning method to train the sample set to obtain a second computing model;

[0025] The second computing model is verified, and when the second computing model meets the requirements, the second computing model is used as a data model.

[0026] The obtaining of a suitable machine learning method specifically includes:

[0027] Using several different machine learning methods to train the sample set to obtain multiple machine models, wherein the sample set includes a training set and a test set;

[0028] The test set is used to test the plurality of machine models to obtain test parameters, so as to select a suitable machine learning method according to the test parameters.

[0029] The second computing model is verified, and when the second computing model meets the requirements, the second computing model is used as a data model, specifically including:

[0030] The test set is used to verify the second computing model to determine whether the calculation result of the second computing model meets the requirements. If so, the second computing model is used as the data model; otherwise, the parameters of the second computing model are adjusted until the calculation result of the second computing model meets the requirements.

[0031] The beneficial effect of the present invention is that the present invention analyzes the modeling and simulation method of secondary circuit equipment, combines mechanism simulation with machine learning technology, and proposes a secondary circuit equipment model simulation method that integrates machine learning and mechanism model. The specific advantages are as follows:

[0032] (1) The data model established using machine learning technology has very good compatibility with the mechanism model, which greatly reduces the difficulty of integrating the data model and the mechanism model.

[0033] (2) The mechanism model and machine learning technology are cleverly integrated, which is different from pure machine learning and pure mechanism model. Instead, they complement each other to form a new type of high-precision model. The mechanism model strengthens the effect of machine learning. On the one hand, the mechanism model can identify abnormal unit data and provide reliable samples for machine learning. On the other hand, compared with unit data, the simulation results of the mechanism model are used as samples with reliable data, large sample size and wide distribution range, which can accurately evaluate the learning ability of various machine learning methods. Similarly, machine learning can identify changes in the characteristics of actual power plant equipment, making up for the shortcomings of the mechanism model.

[0034] (3) The integration of data models and mechanism models is achieved by using a shared database, which not only supports online learning of data models but also simplifies the process of replacing data models. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 Flowchart for building data models using machine learning techniques;

[0036] Figure 2 Ideas for establishing data processing model for the unit;

[0037] Figure 3 is the simulation result;

[0038] Figure 4 Provide a schematic diagram for integrating the mechanism model with the data model;

[0039] Figure 5 The measurement point distribution of the No. 6 high-pressure feedwater heater in a power plant;

[0040] Figure 6 This is the Kalman filter effect. DETAILED DESCRIPTION

[0041] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] The present invention provides a secondary circuit equipment model simulation method integrating machine learning and mechanism model, comprising the following steps:

[0043] S1. Establish a mechanism model of the secondary circuit equipment and perform simulation, compare the simulation result of the mechanism model of the secondary circuit equipment with the unit data of the current unit operation to determine the target process, wherein the target process is the physical process with the largest deviation between the simulation result of the mechanism model of the secondary circuit equipment and the unit data of the current unit operation;

[0044] S2. Acquire a historical unit data set stored in a database, and establish 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 includes at least input parameters and output parameters of a target process;

[0045] S3, inputting the historical unit data set into the unit data processing model, and using the processing result of the unit data processing model as a sample set;

[0046] S4. Obtain a suitable machine learning method, and use the machine learning method to train the sample set to obtain a mapping model of input parameters and output parameters of the target process, and use the mapping model as a data model;

[0047] S5. Integrate the data model into the mechanism model of the secondary circuit equipment to obtain a simulation model of the secondary circuit equipment.

[0048] In this embodiment, a mechanism model of the secondary circuit equipment is first established, and the simulation results of the mechanism model are compared with the unit data. After determining the target process, the unit data processing model is established according to the input parameters and output parameters of the target process in combination with the historical unit data. Then, the unit data processing model is used to process the historical unit data set, and the processing results are used as training samples for subsequent machine learning methods. Then, a suitable machine learning method is selected to train the training samples. After obtaining the data model, the data model is integrated into the mechanism model, thereby obtaining the simulation model of the secondary circuit equipment. In this embodiment, the data model established by machine learning technology has very good compatibility with the mechanism model, which greatly reduces the difficulty of integrating the data model with the mechanism model. The mechanism model and machine learning technology are cleverly integrated, which is different from simple machine learning and simple mechanism model, but complement each other to form a new type of high-precision model. Among them, the mechanism model strengthens the effect of machine learning. Similarly, machine learning can identify changes in the characteristics of actual power plant equipment, making up for the shortcomings of the mechanism model.

[0049] In a preferred embodiment, in step S1, the physical process with the largest deviation is determined by comparing the model simulation results with the unit data of the current unit operation. This physical process is the target process. The theoretical model part of the target process contained in the equipment mechanism model is referred to as a low-precision mechanism model. Since the simulation deviation of the low-precision mechanism model is large, it is necessary to use machine learning technology to establish a data model to replace the low-precision mechanism model to ensure the accuracy of the simulation results. For example, in a specific embodiment, a mechanism model of a high-pressure feedwater heater is established. By comparing the model simulation results with the unit data, it is determined that the biggest shortcoming of the mechanism model is the condensation heat exchange process of the shell-side steam, so this process is used as the target process. Subsequently, the data model established by machine learning technology is used to represent the target process in the mechanism model, thereby improving the accuracy of the simulation.

[0050] Among them, how to determine whether the deviation is the largest is related to the actual situation. For example, for a high-pressure feedwater heater, the parameters that need to be paid attention to are the outlet temperature and water level, and the heat exchange process (which determines the water level and outlet temperature) will seriously affect these two parameters. If the outlet temperature deviation is large, it is obviously caused by errors in the heat exchange process. Since the heat exchange process includes two-phase heat exchange and single-phase heat exchange, the calculation principle of single-phase heat exchange is already very mature (relevant research shows that the accuracy of the single-phase heat exchange mechanism is much higher than that of the two-phase heat exchange mechanism), the process with the largest deviation or the most necessary optimization is the two-phase heat exchange mechanism, that is, the target process is the two-phase heat exchange mechanism process.

[0051] Since the actual unit data does not contain the input and output parameters of the data model (obtained by machine learning technology), it is necessary to establish a "unit data processing model" to process the unit data, so as to obtain a sample library for machine learning. Among them, the purpose of step S2 is to establish a unit data processing model. The function of the unit data processing model is to obtain training samples for subsequent machine learning methods by processing the unit data. This training sample corresponds to the input and output of the target process, so that the input and output of the data model trained by the machine learning method are similar to the input and output of the target process, so that the data model can replace the target process to complete the simulation, thereby achieving the purpose of improving the simulation accuracy and reducing the difficulty of integrating the data model with the machine model.

[0052] In a preferred embodiment, the step S2 specifically includes:

[0053] Acquire a historical unit data set stored in a database, and preprocess the historical unit data set;

[0054] Determining the output of the unit data processing model according to the input parameters and output parameters of the target process;

[0055] A first calculation model is established according to the preprocessed historical unit data set, and the accuracy of the first calculation model is verified. When the verification passes, the first calculation model is used as the unit data processing model.

[0056] In this embodiment, the database is a pre-established historical unit database. When the unit is running, the operating data of the unit in different time periods are stored in the database to facilitate subsequent analysis. After the historical unit data set is obtained, in order to avoid the influence of unreasonable data therein on the establishment of the final model, it is necessary to first pre-process the historical unit data set. Specifically, the historical unit data set stored in the database is obtained, and the pre-processing of the historical unit data set specifically includes:

[0057] The historical unit data set stored in the database is obtained, and after merging the redundant measurement point data according to the preset redundant measurement point processing rules, the Kalman filtering method is used to denoise the redundantly processed data set.

[0058] Specifically, after obtaining the historical unit data set from the database, the redundant measurement point data is first merged according to the processing rules of the unit control system for redundant measurement points, and then the Kalman filter method is used to process the unit data to remove the influence of noise and interference. Among them, the preset redundant measurement point processing rules are determined according to actual needs. For example, in some embodiments, the redundant measurement point processing rules are "bad points are removed when the difference between the two is greater than 5%, and the remaining points are averaged". After removing the redundant points, the Kalman filter method can be used for denoising to remove the influence of noise and interference. Among them, the Kalman filter method is a prior art and will not be repeated here.

[0059] After the historical unit data set is processed, the unit data processing model can be established based on the historical unit data set. Before the establishment, the input and output of the unit data processing model must be determined first. In this embodiment, the input and output of the unit data processing model are determined based on the input parameters and output parameters of the target process. The advantage of this processing is that the input and output of the data model are similar to the target process, which greatly reduces the difficulty of integrating the data model with the mechanism model.

[0060] Furthermore, when establishing the first calculation model, the energy conservation principle and other mechanism models with higher reliability are used (for example, the single-phase heat exchange mechanism mentioned above has higher accuracy than the two-phase heat exchange mechanism, so the single-phase heat exchange mechanism can be selected to construct the first calculation model) to establish the first calculation model. For example, in a specific embodiment, based on the energy conservation principle and the hydrophobic zone convection heat exchange mechanism model, the first calculation model is established. The input of the model is the measurement point data of the actual unit, and the output includes the input and output parameters of the data model. Of course, the output of the first calculation model is not limited to the input and output parameters of the data model, and it can also include other intermediate parameters. The specific construction process depends on the specific equipment type. Specifically, the first calculation model is established according to the pre-processed historical unit data set, and the accuracy of the first calculation model is verified. When the verification is passed, the first calculation model is used as the unit data processing model, which specifically includes:

[0061] Establishing a first calculation model according to the preprocessed historical unit data set;

[0062] Acquire simulation results of the mechanism model of the secondary circuit equipment, wherein the simulation results at least include unit measurement point parameters, input parameters and output parameters of the target process;

[0063] The unit measurement point parameters are input into the first calculation model to obtain calculation results, and the calculation results are compared with the simulation results. When the comparison error is less than a preset value, the first calculation model is used as the unit data processing model; otherwise, the first calculation model is optimized according to the comparison result until the comparison result is less than a preset value.

[0064] In this embodiment, a first calculation model is first established, and the output parameters of the first calculation model include the input and output parameters of the data model (i.e., the input parameters and output parameters of the target process), and then the accuracy of the first calculation model is verified. If it does not meet the requirements, the first calculation model is optimized. The specific optimization process varies according to actual needs. In some embodiments, the parameters in the first calculation model can be directly optimized according to the comparison results, so that its calculation results meet the requirements.

[0065] In this embodiment, the simulation results of the secondary circuit equipment mechanism model are first extracted and processed using the first calculation model. The accuracy of the first calculation model is verified by comparing the errors between the calculation results and the simulation results. If the requirements are not met, the first calculation model is optimized based on the comparison results.

[0066] Specifically, since the output parameters of the unit data processing model are some intermediate parameters during the unit operation, details can be found in Figure 2 , and the simulation results include these parameters (such as Figure 3Therefore, the output parameters of the simulation results (the same parameters as the unit data, such as pressure, temperature, etc.) can be used as the input parameters of the unit data processing model, and the processing results (corresponding to Figure 2 The three boxes on the right side of the diagram) can then be used to compare the processed results with the simulated results.

[0067] Further, in this embodiment, since the unit data does not contain the input parameters and output parameters of the target process, the simulation results are used to verify the unit data processing model. Since the basic energy conservation and a small number of necessary mechanism models with higher reliability constitute the "unit data processing model" (which cannot contain the mechanism model of the target process), the relationship between the "input parameters of the data model" and the "output parameters of the data model" is not known, but the two types of data are obtained simultaneously directly based on the unit data, so the simulation results are needed for verification. Because the simulation data is calculated by the mechanism model, the mapping relationship between the "input parameters of the data model" and the "output parameters of the data model" is clearly known, but the "unit data processing model" does not know this relationship. Therefore, by processing the output parameters of the simulation through the "unit data processing model", these two parameters will be obtained. By comparing with the simulation data, it can be known whether the two parameters obtained by processing accurately reflect the mapping relationship between the two.

[0068] In a preferred embodiment, in step S3, after the unit data processing model is obtained, the unit data processing model can be used to process the historical unit data set. Since the input of the unit data processing model is the unit data, and the output includes the input parameters and output parameters of the target process, after the historical unit data set is input, multiple pairs of input parameters and output parameters can be obtained. The multiple pairs of input parameters and output parameters constitute a sample set for the subsequent machine learning method, which can be used as a sample library used for machine learning.

[0069] In addition, it should be noted that when the unit data samples are unevenly distributed or the error is too large, it is difficult to fully evaluate the learning ability of various machine learning methods on the physical characteristics of the target process. In order to solve this problem, considering that the low-precision mechanism model is not accurate enough, but the change characteristics should be similar to the actual physical process, a low-precision mechanism model is used to generate data to supplement the sample library, which not only avoids the evaluation bias caused by poor sample quality, but also utilizes the poor quality unit data in the sample library.

[0070] In a preferred embodiment, in step S4, in order to ensure the accuracy of the results, it is necessary to first obtain a suitable machine learning method, and then use the method to train the sample set to obtain a mapping model. This mapping model reflects the mapping relationship between the input parameters and output parameters of the target process. The output parameters calculated by this mapping model can solve the problem of large errors in the output results of the target process, thereby further ensuring the accuracy of the simulation results of the entire mechanism model by combining with the mechanism model. Specifically, step S4 specifically includes:

[0071] Get the right machine learning methods;

[0072] Using the machine learning method to train the sample set to obtain a second computing model;

[0073] The second computing model is verified, and when the second computing model meets the requirements, the second computing model is used as a data model.

[0074] In this embodiment, the method of obtaining a suitable machine learning method specifically includes:

[0075] Using several different machine learning methods to train the sample set to obtain multiple machine models, wherein the sample set includes a training set and a test set;

[0076] The test set is used to test the plurality of machine models to obtain test parameters, so as to select a suitable machine learning method according to the test parameters.

[0077] In this embodiment, firstly, several different machine learning methods are used to train the training set, wherein the machine learning methods include but are not limited to: linear regression method, decision tree regression method, random forest regression method, gradient enhanced random forest regression method, bagging regression method, K nearest neighbor regression method, support vector machine regression method, Adaboost regression method, LightGBM regression method, GBDT regression method, XGB model algorithm, MLP neural network algorithm, etc., so that multiple machine models can be obtained, and then the test set is used for testing to obtain test parameters, wherein the test parameters at least include correlation coefficient, RMSE, R 2 After that, the user can independently select the appropriate machine learning method according to the test parameters. This selection process is a user's own choice and this application does not limit it. For example, the MLP neural network algorithm is selected.

[0078] When the machine learning method is selected, the machine learning method can be used to train the sample set to obtain the second computing model. In order to ensure the accuracy of the calculation of the second computing model, it is also necessary to use the test set to verify the second computing model. Specifically, in order to more comprehensively test the learning ability of the machine learning method, the sample set retains a certain range of data that is not used for model training. This part of the data is called the validation set, and the remaining samples are divided into training sets and test sets using cross-validation. In order to ensure the accuracy of the data model calculation, it is also necessary to verify the second computing model. Specifically, the verification of the second computing model, when the second computing model meets the requirements, the second computing model is used as the data model. Specifically includes:

[0079] The test set is used to verify the second computing model to determine whether the calculation result of the second computing model meets the requirements. If so, the second computing model is used as the data model; otherwise, the parameters of the second computing model are adjusted until the calculation result of the second computing model meets the requirements.

[0080] In this embodiment, when the second computing model does not meet the requirements, the parameter settings of the second computing model are adjusted. For example, for the neural network algorithm, the number of nodes and the number of neural network layers can be adjusted, and the learning rate can be adjusted according to the change process of the loss function. If the prediction ability of the second computing model can meet the requirements, the second computing model is used as the data model; otherwise, the sample data is expanded and re-trained using the machine learning algorithm. The data source is not limited to the unit data, but also includes experimental data or mechanism model simulation data.

[0081] When the data model is established, its input is the same as the input of the target process, and its output can replace the output of the target process, thereby avoiding the problem of inaccurate simulation results caused by errors in the output of the target process. Therefore, after integrating the data model and the mechanism model, the obtained simulation model has extremely high simulation capabilities, and the output data is closer to the real data. In this embodiment, Figure 4 As shown in the figure, the data model and the mechanism model are integrated by using a shared database. The shared database is used as the data transmission medium, the mechanism model provides input parameters for the data-driven model, and the data model returns the prediction results to the mechanism model, forming an organic whole. The advantage of this integration method is that it not only supports online learning of the data model, but also simplifies the process of replacing the data model.

[0082] In a preferred embodiment, the step S5 further includes:

[0083] The database is updated, and the data model is optimized according to the historical unit data set in the updated database.

[0084] In this embodiment, the latest unit data can be used as samples for online learning, or a new data model can be rebuilt based on these samples to replace the old model to ensure that the model can timely identify changes in the actual power plant equipment characteristics.

[0085] In order to better understand the present invention, the following Figures 1 to 6 , a specific embodiment is given to explain the technical solution of the present invention in detail:

[0086] The specific implementation process of the present invention is described below by taking a typical secondary circuit device such as a high-pressure feedwater heater as an example:

[0087] Step 1: Establish a mechanism model of the high-pressure feedwater heater. By comparing the model simulation results with the unit data, it is determined that the biggest shortcoming of the mechanism model lies in the condensation heat transfer process of the shell-side steam, so this process is taken as the target process.

[0088] Step 2: If Figure 1 As shown, machine learning technology is used to establish a data model of the target process.

[0089] Step 2.1 Select the input and output parameters of the data model:

[0090] The condensation heat transfer coefficient is the key characteristic parameter of the condensation heat transfer process and is calculated using the following formula in the mechanism model:

[0091]

[0092] Refer to the above formula and select the condensation heat transfer coefficient h con As the output of the data model, select the latent heat of vaporization r, liquid film density ρ l , liquid film thermal conductivity λ l 、Liquid film dynamic viscosity η l , outer diameter of heat transfer tube d, steam temperature t s , heat transfer tube wall temperature t w and temperature difference (t s -t w ) as input parameter.

[0093] Step 2.2 Preprocessing unit data:

[0094] Figure 5 The measurement point distribution of the No. 6 high-pressure feedwater heater in a power plant was used to obtain the actual operating data of the feedwater heater from the database, such as the feedwater heater drain outlet temperature, shell side water level, heat transfer tube inlet and outlet pressure and temperature, valve opening on the drain outlet pipeline, deaerator and condenser pressure, etc.

[0095] Then, according to the DCS system's processing rule for redundant measuring points: "when the difference between any two points is greater than 5%, the bad points are removed, and the average value of the remaining points is taken", the redundant pressure and water level measuring points are processed.

[0096] Finally, the Kalman filter method is used to process the unit data to remove the influence of noise and interference. Figure 6 This is the filtering effect of the outlet pressure on the tube side of the high-pressure feedwater heater.

[0097] Step 2.3 Establish and verify the unit data processing model:

[0098] like Figure 2 As shown, based on the energy conservation principle and the convection heat transfer mechanism model of the drain zone, a unit data processing model is established. The input of the model is the measurement point data of the actual unit, and the output includes the input and output parameters of the data model.

[0099] Then the simulation results of the high-pressure feedwater heater mechanism model are extracted. The extracted parameters include the unit measurement point parameters, the input and output parameters of the data model, and other key intermediate parameters such as the shell side saturation temperature and condensation heat transfer capacity.

[0100] Finally, the unit data processing model is used to process the extracted simulation results. By comparing the errors between the processing results and the simulation results, the error is much less than 1%. Therefore, the accuracy of this unit data processing model can meet the requirements.

[0101] Step 2.4: Processing unit data:

[0102] The established unit data processing model is used to process the unit data, and the input and output data samples of the data model are obtained, which can be used as the sample library used for machine learning. Under low operating conditions, the high-pressure feedwater heater is sometimes not running. When processing the unit data at this time, the calculation results will exceed the limit, and the data exceeding the limit should be eliminated.

[0103] Step 2.5 Screening of machine learning methods:

[0104] The group data processed in step 2.4 is used as a sample library for screening machine learning methods.

[0105] In order to more comprehensively test the learning ability of machine learning methods, a certain range of data is reserved and not used for model training. This part of the data is called the validation set, and the remaining samples are divided into training sets and test sets using cross-validation. Then, more than ten machine learning methods such as random forests and neural networks are used to establish input-to-output mapping models, and the test set and validation set are used to evaluate the generalization ability of the model. The evaluation indicators are correlation coefficient, RMSE, and R2. From the evaluation results, it can be seen that for the condensation heat exchange process of the feedwater heater, the neural network method has a very strong learning ability.

[0106] Step 2.6: Create a data model:

[0107] The sample library is divided into training set and test set by cross validation method, and the data model is established by the neural network method selected in step 2.5. The established model contains 3 hidden layers, 2048 nodes in each layer, the learning rate is 0.02, and the output parameter is the condensation heat transfer resistance (i.e. 1 / hcon). The model prediction ability can meet the requirements, and go to step 3.

[0108] Step 3: Integrate the data model established in step 2 into the mechanism model.

[0109] like Figure 4 As shown in the figure, the data model and the mechanism model are integrated by means of a shared database. With the shared database as the data transmission medium, the mechanism model provides input parameters for the data model, and the data model returns the prediction results to the mechanism model, thus forming a high-precision intelligent high-pressure feedwater heater model.

Claims

1. A secondary circuit equipment model simulation method integrating machine learning and mechanism model, characterized in that: The steps include: S1: Establishing a mechanism model of the secondary circuit equipment and performing simulation, comparing the simulation result of the mechanism model of the secondary circuit equipment with the unit data of the current unit operation to determine the target process, wherein the target process is the physical process with the largest deviation between the simulation result of the mechanism model of the secondary circuit 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 includes at least input parameters and output parameters of a target process; S3: inputting the historical unit data set into the unit data processing model, and taking the processing result of the unit data processing model as a sample set; S4: obtaining a machine learning method, using the machine learning method to train the sample set to obtain a mapping model of input parameters and output parameters of the target process, and using the mapping model as a data model; S5 integrates the data model into the mechanism model of the secondary circuit equipment to obtain a secondary circuit equipment simulation model. The data model and the mechanism model are integrated in a shared database manner, with the shared database as the data transmission medium. The mechanism model provides input parameters for the data-driven model, and the data model returns the prediction results to the mechanism model to form an organic whole.

2. A secondary circuit equipment model simulation method integrating machine learning and mechanism model as claimed in claim 1, characterized in that: The step S2 specifically includes: Acquire a historical unit data set stored in a database, and preprocess the historical unit data set; Determining the output of the unit data processing model according to the input parameters and output parameters of the target process; A first calculation model is established according to the preprocessed historical unit data set, and the accuracy of the first calculation model is verified. When the verification passes, the first calculation model is used as the unit data processing model.

3. A secondary circuit equipment model simulation method integrating machine learning and mechanism model as claimed in claim 1, characterized in that: The acquiring of the historical unit data set stored in the database and preprocessing of the historical unit data set specifically include: After obtaining the historical unit data set stored in the database, first fill in the empty values ​​based on the power plant data collection and storage rules, perform outlier detection and data cleaning, then merge the redundant measurement point data according to the preset redundant measurement point processing rules, and finally use the Kalman filtering method to denoise the redundantly processed data set.

4. A secondary circuit equipment model simulation method integrating machine learning and mechanism model as claimed in claim 2, characterized in that: Establishing a first calculation model according to the preprocessed historical unit data set, and verifying the accuracy of the first calculation model, and when the verification passes, using the first calculation model as the unit data processing model specifically includes: Establishing a first calculation model according to the preprocessed historical unit data set; Acquire simulation results of the mechanism model of the secondary circuit equipment, wherein the simulation results at least include unit measurement point parameters, input parameters and output parameters of the target process; The unit measurement point parameters are input into the first calculation model to obtain calculation results, and the calculation results are compared with the simulation results. When the comparison error is less than a preset value, the first calculation model is used as the unit data processing model; otherwise, the first calculation model is optimized according to the comparison result until the comparison result is less than a preset value.

5. A secondary circuit equipment model simulation method integrating machine learning and mechanism model as claimed in claim 1, characterized in that: The step S4 specifically includes: Get machine learning methods; Using the machine learning method to train the sample set to obtain a second computing model; The second computing model is verified, and when the second computing model meets the requirements, the second computing model is used as a data model.

6. A secondary circuit equipment model simulation method integrating machine learning and mechanism model as claimed in claim 5, characterized in that: The method for obtaining machine learning specifically includes: Using several different machine learning methods to train the sample set to obtain multiple machine models, wherein the sample set includes a training set and a test set; The test set is used to test the plurality of machine models to obtain test parameters, so as to select a machine learning method according to the test parameters.

7. A secondary circuit equipment model simulation method integrating machine learning and mechanism model as claimed in claim 6, characterized in that: The second computing model is verified, and when the second computing model meets the requirements, the second computing model is used as a data model, specifically including: The test set is used to verify the second computing model to determine whether the calculation result of the second computing model meets the requirements. If so, the second computing model is used as the data model; otherwise, the parameters of the second computing model are adjusted until the calculation result of the second computing model meets the requirements.

Images