Modelica-based economizer parameter identification hybrid modeling method

Abstract

The invention discloses an economizer parameter identification hybrid modeling method based on Modelica, and the method comprises the steps: carrying out the main mechanism modeling of an economizer through a Modelica language, carrying out the parameter identification data driving modeling of a feed water flow resistance coefficient through a DNN neural network model for a water pressure drop problem which is difficult to predict in the mechanism prediction of the economizer, and exporting a DNN network parameter, and integrating the DNN neural network model into the mechanism model of the economizer. According to the method, a universal mechanism model capable of explaining the economizer is carried out by means of Modelica, meanwhile, the prediction precision of the mechanism model is improved by means of a data-driven model, finally, parameters of the data-driven model are exported and written into the mechanism model, and efficient application of engineering is achieved.

Description

Technical Field

[0001] This application relates to the technical field of thermal power plant unit equipment modeling, and particularly to a hybrid modeling method for economizer parameter identification based on Modelica. Background Art

[0002] Modelica is a component-oriented modeling language dedicated to multi-domain physical system modeling. Its mechanism model is based on physical laws and describes the dynamic behavior of the system through equations, which is very suitable for simulating the interactions of complex systems such as thermal, fluid, and electromechanical systems. Modelica supports a component-oriented modeling method, allowing equipment and systems to be modeled as independent modules (or components), and each module represents a unit equipment (such as pumps, valves, boilers, etc.). These modules can be reused in different systems, greatly improving the reusability and maintainability of the model. However, there may be idealized situations in the mechanism model, resulting in the problem that the model prediction does not match the actual situation.

[0003] The mechanism-data hybrid modeling combines the interpretability of the physical mechanism model with the accuracy of the data-driven model, which can effectively improve the prediction ability and robustness of the model. The physical mechanism model provides the basic laws of the system to ensure the generality of the model, while the data-driven model can capture complex non-linear relationships to make up for the deficiencies of the physical model, thereby improving the overall performance. The current mechanism-data hybrid models mainly use various interfaces to achieve information transfer between the two models. However, the mechanism models and data-driven models on different platforms have great limitations in the actual application process, especially in the system modeling of Modelica unit equipment models, which will make the system extremely redundant. Summary of the Invention

[0004] The purpose of the embodiments of this application is to provide a hybrid modeling method for economizer parameter identification based on Modelica to solve the technical problem of the transferability of the hybrid model in actual applications in related technologies.

[0005] According to the first aspect of the embodiments of this application, a hybrid modeling method for economizer parameter identification based on Modelica is provided, characterized by including: Using Modelica to model the heat transfer process between the flue gas and water in the economizer and the energy and momentum balance of the two fluid streams to obtain a mechanism model; Calculating the ideal feed water flow resistance coefficient using the first on-site data, where the first on-site data includes the flue gas pressure drop at the inlet and outlet of the economizer, the feed water flow rate, the feed water temperature, the outlet water temperature, and the height difference between the inlet and outlet of the economizer; Standardize the second-site data, and then use the PCC and MI methods to calculate the input features of the DNN model by weighted calculation; the second-site data includes the furnace coal quantity, feed water flow rate, temperature, pressure, inlet flue gas temperature, pressure, and flow rate. Use the input features of the DNN model and the ideal feed water flow resistance coefficient as the output features to train the DNN model, and use Bayesian optimization to tune the structure and hyperparameters of the DNN model. Export the weights, biases, and activation functions of each layer of neurons in the tuned DNN model. Write the weights, biases, and activation functions into the mechanism model, create a new function in the mechanism model, and set multiple parameters to store the weight and bias matrices, and the mean and variance of the 6D input features. Use the Modelica language to reconstruct the mapping relationship between the neurons of each layer of the DNN model in the newly created function. Call the newly created function through the mechanism model to realize the parameter identification of the feed water flow resistance coefficient using the DNN model in the mechanism model.

[0006] Optionally, use Modelica to construct equations for the heat transfer and momentum balance of the flue gas and water in the economizer, including: Obtain two fluid streams of flue gas and water, represented by subscripts g and w respectively. The three parameters of the two fluid streams, temperature, pressure, and flow rate, are represented by T, p, and qm respectively. The input and output fluid streams are distinguished by subscripts i and o. The heat transfer equation for the flue gas and water is constructed as: ; ; Among them, is the average temperature difference between the two fluid streams, is the heat transfer power, and the coefficient is the heat transfer coefficient; Construct the flue gas momentum balance equation: ; Among them, is the fixed pressure drop generated in the flue gas section due to reasons such as height, is the flow resistance coefficient of the flue gas, is the flue gas density; The momentum balance equation of water: ; Among them, is the height difference between the inlet and outlet of the economizer fluid stream water, is the parameter required for parameter identification using the DNN model; Construct the energy balance equation: ; Among them, is the internal volume of the economizer, is the specific heat of the flue gas.

[0007] Optionally, a DNN model is used to identify the parameters of the feed water flow resistance coefficient in the mechanism model. The standardized training data is input. The DNN model includes an input layer, a hidden layer, and an output layer. In the DNN model, the input feature dimension is 6, including the furnace coal amount, the economizer feed water flow rate, pressure, the inlet flue gas temperature, pressure, and flow rate. The output feature dimension is 1, which is the feed water flow resistance coefficient. The number of intermediate hidden layers and the number of neurons in each layer are obtained through Bayesian optimization. In the DNN model, data is transmitted from the input layer to the hidden layer, and thus passes through all hidden layers and finally reaches the output layer to complete a forward pass. Each neuron receives the output from the previous layer, and after weighted summation, the output result is obtained through the activation function ReLU.

[0008] Optionally, the reconstructed mapping relationship is as follows: represents the result after the model input is standardized, represents the mapping result of the nth layer of neurons, represents the weight matrix of the nth layer of neurons; is the bias matrix of the nth layer of neurons, represents the final mapping result, that is, the predicted result of the required feed water flow resistance coefficient, where represents taking the larger value of each element in matrix A compared with 0, that is, the calculation method of ReLU. There is no ReLU in the last output layer, so there is only a linear mapping. ReLU is the activation function.

[0009] According to the second aspect of the embodiments of the present application, a Modelica-based economizer parameter identification hybrid modeling device is provided, including: A modeling module for using Modelica to model the heat transfer process between the flue gas and water in the economizer and the energy and momentum balance of the two fluid streams to obtain a mechanism model; A first calculation module for calculating the ideal feed water flow resistance coefficient using the first on-site data, where the first on-site data includes the flue gas pressure drop at the inlet and outlet of the economizer, the feed water flow rate, the feed water temperature, the outlet water temperature, and the height difference between the inlet and outlet of the economizer; A second calculation module, configured to standardize the second on-site data and then use two methods, PCC and MI, to calculate the input features of the DNN model by weighted calculation; the second on-site data includes the coal amount in the furnace, the feed water flow rate, temperature, pressure, the inlet flue gas temperature, pressure, and flow rate. A training and optimization module, configured to use the input features of the DNN model and the ideal feed water flow resistance coefficient as the output features to train the DNN model, and use Bayesian optimization to tune the structure and hyperparameters of the DNN model. An export module, configured to export the weights, biases, and activation functions of each layer of neurons of the tuned DNN model. A writing module, configured to write the weights and biases into the mechanism model, create a new function in the mechanism model, and set multiple parameters to store the weight and bias matrices, and the mean and variance of the 6-dimensional input features. A reconstruction module, configured to use the Modelica language to reconstruct the mapping relationship between each layer of neurons of the DNN model in the newly created function. A calling module, configured to call the newly created function through the mechanism model to implement parameter identification of the feed water flow resistance coefficient using the DNN model in the mechanism model.

[0010] According to the third aspect of the embodiments of the present application, there is provided an electronic device, characterized by including: One or more processors; A memory, configured to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in the first aspect.

[0011] According to the third aspect of the embodiments of the present application, there is provided a computer-readable storage medium, on which computer instructions are stored, and when the instructions are executed by a processor, the steps of the method as described in the first aspect are implemented.

[0012] The technical solutions provided by the embodiments of the present application may include the following beneficial effects: The present application adopts the mechanism modeling method of Modelica, and the obtained final unit device model can be migrated to various similar scenarios, with strong reusability and expandability.

[0013] Adopts a data-driven parameter identification method, which overcomes the problem of low model prediction accuracy existing in the mechanism model.

[0014] Export the data-driven model and write it into the function of Modelica, realizing an integrated unit device model of the mechanism data hybrid model on the Modelica platform, and solving the problem that the current hybrid model needs an interface to read the model in actual engineering applications.

[0015] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and should not limit this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings herein are incorporated into and constitute a part of this specification, showing embodiments consistent with this application, and are used together with the specification to explain the principles of this application.

[0017] Figure 1 is a flowchart of a Modelica-based hybrid modeling method for economizer parameter identification according to an exemplary embodiment.

[0018] FIG. 2 is a schematic diagram comparing the model prediction results after using the present invention with the mechanism model prediction results without parameter identification according to an exemplary embodiment.

[0019] Figure 3 is a schematic structural diagram of a Modelica-based hybrid modeling device for economizer parameter identification according to an exemplary embodiment. DETAILED DESCRIPTION

[0020] Exemplary embodiments will be described in detail herein, and examples thereof are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. On the contrary, they are merely examples of devices and methods consistent with some aspects of this application as detailed in the appended claims.

[0021] The terms used in this application are for the purpose of describing particular embodiments only and are not intended to limit this application. The singular forms "a", "the", and "said" used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0022] It should be understood that although the terms first, second, third, etc. may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "when" or "upon determining".

[0023] Figure 1 It is a flowchart of a hybrid modeling method for economizer parameter identification based on Modelica shown according to an exemplary embodiment. As Figure 1 shown, this method is applied to a terminal and may include the following steps: Step S1: Use Modelica to model the heat transfer process between the flue gas and water in the economizer and the energy and momentum balance of the two fluid streams to obtain a mechanism model; Specifically, the mechanism model mainly consists of the following heat transfer equation, flue gas momentum balance equation, water momentum balance equation, and energy balance equation. The construction of each equation will be elaborated in detail below.

[0024] Create a new model in Modelica, define the flow rate, temperature, and pressure of the inlet and outlet fluid streams of the flue gas and water, and write the following equations into the equation of the model. First, the two fluid streams of the flue gas and water are represented by subscripts g and w respectively. The three parameters of the two fluid streams, temperature, pressure, and flow rate, are represented by T, p, and qm respectively. The inlet and outlet fluid streams are distinguished by subscripts i and o. The heat transfer equation is: where is the average temperature difference between the two fluid streams, is the heat transfer power, is the heat transfer area, and the coefficient is fitted through training data.

[0025] Flue gas momentum balance equation: where is the fixed pressure drop generated in the flue gas section due to reasons such as height, is the flow resistance coefficient of the flue gas, is the flue gas density.

[0026] Water momentum balance equation: where is the height difference between the inlet and outlet of the water in the economizer fluid stream.

[0027] Energy balance equation: is the internal volume of the economizer, is the specific heat of the flue gas.

[0028] As a non-causal modeling method, Modelica uses equations to describe the heat exchange process of the economizer. In the actual use of the model, as long as a sufficient number of model boundary conditions are known, the remaining boundary conditions can be calculated, rather than being limited to determining the outlet flow through the inlet flow only.

[0029] Step S2: Calculate the ideal flow resistance coefficient of the feed water using the first on-site data, which is collected by the sensors of the thermal power plant. The first on-site data includes flue gas pressure drop, feed water flow rate, feed water temperature, outlet water temperature, and height difference between the inlet and outlet of the economizer. Specifically, calculate the ideal flow resistance coefficient of the feed water for each group using the first on-site data. The specific calculation formula is as follows: The density of water is calculated through the average temperature and pressure. In this way, the ideal heat transfer coefficient of this process can be calculated, making the prediction of the DNN model tend to the desired ideal value.

[0030] Step S3: Standardize the second on-site data, which is also collected by the factory sensors, and then use the PCC and MI methods to calculate the input features of the DNN model by weighted calculation. The second on-site data includes the amount of coal in the furnace, feed water flow rate, temperature, pressure, inlet flue gas temperature, pressure, and flow rate. Specifically, standardize the training data of the data-driven model: Use the PCC and MI methods, two methods for calculating data correlation, for the standardized data: This is the partial correlation coefficient of XY after controlling for variable Z, where represents the Pearson correlation coefficient between XY, represents the Pearson correlation coefficient between YZ, represents the Pearson correlation coefficient between XZ. The calculated result reflects the linear relationship between XY after excluding the influence of Z. The calculation result of the partial correlation coefficient is a number in [-1, 1]. -1 indicates negative linear correlation, 1 indicates linear correlation, and 0 indicates linear independence. The MI calculation formula is as follows: where the mutual information is the joint distribution of xy and and the marginal distribution The relative entropy. The calculated result of mutual information is a positive number greater than 0 and does not exceed the entropies of x and y themselves .

[0031] The sum of the two items is calculated with weights as the final evaluation index, and the weighting formula is as follows: where represents the evaluation score of the nth input feature, represents the partial correlation coefficient between the nth input feature and the output after controlling all other input features, and the last item represents the ratio of the mutual information between the input feature and Y to the entropy of Y itself. Thus, through the size relationship between them, and the number n of input features, a comprehensive judgment is made to select appropriate input features for subsequent neural network training and prediction. In this case, the input features are the boiler coal quantity, economizer feed water flow rate, feed water pressure, flue gas flow rate, inlet flue gas temperature, inlet flue gas pressure, which may be related to the mechanism analysis and the feed water flow resistance coefficient. Excluding the feed water temperature from these seven items, the remaining six items are used as the input features of the DNN network. In this case, the output feature to be regression predicted is the feed water flow resistance coefficient of the economizer.

[0032] In data-driven modeling, selecting input features is crucial for model performance. The mechanism analysis method based on physical, chemical, or engineering science models provides a clear theoretical basis for feature selection and combines domain knowledge. However, these models usually oversimplify the system and may ignore important features or treat them as constants in practice, which may hinder the effectiveness of data-driven models. In addition, some variables are far from the input in the system, so it is difficult to quantify their impact on the output. MI and PCC measure the statistical correlation between two variables, indicating how much the uncertainty of one variable is reduced by knowing the other variable. It can capture both linear and complex non-linear relationships between input variables and target variables. It makes fewer assumptions about the data distribution and is applicable to various data types, including continuous and discrete variables. This can also quantitatively measure the importance of different input variables to the target variable.

[0033] Step S4: Use the input features of the DNN model and the ideal feed water flow resistance coefficient as the output feature to train the DNN model, and use Bayesian optimization to tune the structure and hyperparameters of the DNN model; Specifically, a DNN model is used to perform parameter identification on the feed water flow resistance coefficient in the mechanism model. The standardized training data is input. The DNN model includes an input layer, a hidden layer, and an output layer. In the DNN model, the input features have a dimension of 6 for S3, including the furnace coal amount, economizer feed water flow rate and pressure, inlet flue gas temperature, pressure, and flow rate. The output feature dimension is 1, which is the feed water flow resistance coefficient. The number of intermediate hidden layers and the number of neurons in each layer are obtained through Bayesian optimization. In the DNN model, data is transmitted from the input layer to the hidden layer, and thus passes through all hidden layers and finally reaches the output layer to complete a forward pass. Each neuron receives the output from the previous layer, and after weighted summation, the output result is obtained through the ReLU activation function.

[0034] Use Bayesian optimization to tune the hyperparameters of the DNN network. Bayesian hyperparameter tuning simulates the objective function by constructing a surrogate model, that is, the relationship between the hyperparameters of the model and the evaluation index MSE. This includes setting: The learning rate is between 10 -6 -10 -1 ; The number of hidden layers is between 1 and 5 layers; The number of neurons in each layer is between 32 and 128.

[0035] The hyperparameter space defined in this way provides a search range for optimization. Subsequently, the objective function objective is defined. In this function, a set of hyperparameter combinations is input each time it is called. A DNN model is constructed with these parameters, and the Adam optimizer is used. The performance is measured by the MSE value of the validation set during model training. The objective function is called multiple times to evaluate different hyperparameter combinations. The surrogate model is a Gaussian process, and the exploration and exploitation are balanced through the acquisition function in each iteration, gradually narrowing the range to find a better solution. Finally, the optimal hyperparameter combination is output. The final hyperparameters obtained by searching in this case are 3 hidden layers, the number of neurons are 69, 126, 126 respectively, and the learning rate is 0.00906155284684755.

[0036] Using Bayesian parameter tuning, a relatively good set of hyperparameters for the DNN model can be obtained, and the results of several sets of hyperparameters in the Bayesian iteration process can provide guidance for subsequent hyperparameter selection. Here, the DNN model is used. On the one hand, because the DNN model has relatively high accuracy, and during the experiment, the accuracy of the DNN model is generally higher than that of models such as GPR and LSTM. Secondly, the framework of the DNN model is relatively simple, and the parameters of the model are mainly the weights and biases of neurons in each layer, as well as the activation function. There is no need to save the current cell state of the model like the LSTM algorithm and pass this state to the next moment, which is too difficult to implement in Modelica; compared with GPR, when the amount of data is large, the number of parameters will not change due to the change of training data. GPR needs to save the covariance matrix of training parameters, and an increase in the amount of training data will greatly increase the number of parameters in the final encapsulated model. In comparison, the DNN model is easy to implement and can ensure high accuracy.

[0037] Step S5: Export the weights, biases of neurons in each layer of the tuned DNN model, and the activation function; Specifically, export the weight and bias matrices of each layer of the DNN after training is completed, thereby obtaining four weight matrices and four bias matrices, namely W1[69,6], W2[126,69], W3[126,126], W4[1,126], B1[69,1], B2[126,1], B3[126,1], B4[1,1]. In addition, export the type of activation function of the network.

[0038] Step S6: Write the weights and biases into the mechanism model, create a new function in the mechanism model, and set multiple parameters to store the weight and bias matrices, the mean and variance of 6-dimensional input features; Specifically, create a new function in the mo file, set six inputs input Real and one output output Real, and set eight parameter matrices to store the weight matrices and bias matrices exported by the DNN model, namely W1[69,6], W2[126,69], W3[126,126], W4[1,126], B1[69,1], B2[126,1], B3[126,1], B4[1,1], and a parameter matrix for storing the mean and variance of input features.

[0039] Step S7: Use the Modelica language in the newly created function to reconstruct the mapping relationship between neurons in each layer of the DNN model; Specifically, the reconstructed mapping relationship is as follows: It represents the result after the model input is normalized. It represents the mapping result of the neurons in the nth layer. It represents the weight matrix of the neurons in the nth layer; It is the bias matrix of the neurons in the nth layer. It represents the final mapping result, that is, the predicted result of the water supply flow resistance coefficient required, where It represents taking the larger value of each element in matrix A compared with 0, that is, the calculation method of ReLU. There is no ReLU in the last output layer, so there is only a linear mapping. ReLU is the activation function.

[0040] It is used for data normalization. After a set of input features are normalized in the function, it becomes x[6,1], and the output feature is y[1,1]. Subsequently, the following calculations are performed: This y is the water supply flow resistance coefficient for which parameter identification is required.

[0041] Writing this DNN model in the form of a function allows the function to be debugged separately in Modelica and also called in different models, improving the reusability of the function. At the same time, performing operations in the form of a function does not require writing all parameters into memory during model operation like a model, reducing the load on the computer during model operation and accelerating the model operation speed.

[0042] Step S8: Call this newly created function through the mechanism model to achieve the use of the DNN model in the mechanism model for parameter identification of the water supply flow resistance coefficient.

[0043] It solves the problem that the current mainstream hybrid models use various interfaces to achieve information transfer between the two models. However, the mechanism models and data-driven models on different platforms have great limitations in the actual application process, especially in the system modeling of Modelica unit device models, which will cause the system to become extremely redundant.

[0044] The final result is shown in Figure 2 , the abscissa represents the number of data groups, and the ordinate represents the pressure drop value in pascals. The figure shows the true value of the pressure drop, the prediction result of the mechanism model when using a fixed value as the water supply flow resistance coefficient, and the prediction result of the hybrid model after using DNN model parameter identification for the economizer pressure drop prediction. It can be seen from these three that the prediction accuracy after using parameter identification is greatly improved.

[0045] As can be seen from the above embodiments, the present application adopts the mechanism modeling method of Modelica, so the obtained final unit device model can be migrated to various similar scenarios, with strong reusability and expandability. The data-driven parameter identification method is adopted to overcome the problem of low model prediction accuracy existing in the mechanism model. The data-driven model is exported and written into the function of Modelica, realizing the unit device model of the mechanism-data hybrid model integrated on the Modelica platform, and solving the problem that the current hybrid model needs an interface to read the model in actual engineering applications.

[0046] Corresponding to the foregoing embodiment of the hybrid modeling method for economizer parameter identification based on Modelica, the present application also provides an embodiment of a hybrid modeling device for economizer parameter identification based on Modelica.

[0047] Figure 3 It is a block diagram of a hybrid modeling device for economizer parameter identification based on Modelica shown according to an exemplary embodiment. Referring to Figure 3 , the device includes: Modeling module 1, configured to use Modelica to model the heat transfer process between the flue gas and water in the economizer and the energy and momentum balance of the two fluid streams to obtain a mechanism model; First calculation module 2, configured to calculate the ideal feed water flow resistance coefficient using first on-site data, where the first on-site data includes the flue gas pressure drop at the inlet and outlet of the economizer, the feed water flow rate, the feed water temperature, the outlet water temperature, and the height difference between the inlet and outlet of the economizer; Second calculation module 3, configured to perform normalization processing on second on-site data, and then use the PCC and MI methods to weightedly calculate the DNN model input features; the second on-site data includes the furnace coal amount, the feed water flow rate, temperature, pressure, the inlet flue gas temperature, pressure, and flow rate; Training and optimization module 4, configured to use the DNN model input features and the ideal feed water flow resistance coefficient as output features to train the DNN model, and use Bayesian optimization to tune the structure and hyperparameters of the DNN model; Export module 5, configured to export the weights, biases, and activation functions of the neurons in each layer of the tuned DNN model; Writing module 6, configured to write the weights, biases, and activation functions into the mechanism model, create a new function in the mechanism model, and set multiple parameters to store the weight and bias matrices, and the mean and variance of the 6-dimensional input features; Reconstruction module 7, configured to reconstruct the mapping relationship between the neurons in each layer of the DNN model using the Modelica language in the newly created function; A calling module 8 is used to call the newly created function through the mechanism model, so as to realize parameter identification of the feed water flow resistance coefficient by using the DNN model in the mechanism model.

[0048] Regarding the device in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here in detail.

[0049] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can refer to the partial descriptions of the method embodiments. The device embodiments described above are only illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this application. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0050] Correspondingly, this application also provides an electronic device, including: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the Modelica-based economizer parameter identification hybrid modeling method as described above.

[0051] Correspondingly, this application also provides a computer-readable storage medium, on which computer instructions are stored, and when the instructions are executed by a processor, the Modelica-based economizer parameter identification hybrid modeling method as described above is implemented.

[0052] After considering the specification and practicing the content disclosed here, those skilled in the art will easily think of other implementation schemes of this application. This application aims to cover any variations, uses or adaptive changes of this application, and these variations, uses or adaptive changes follow the general principles of this application and include the common general knowledge or conventional technical means in the technical field not disclosed in this application. The specification and embodiments are only regarded as exemplary, and the true scope and spirit of this application are pointed out by the claims.

[0053] It should be understood that this application is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is only limited by the appended claims.

Claims

1. A hybrid modeling method for economizer parameter identification based on Modelica, characterized in that: include: Modelica was used to model the heat transfer process between flue gas and water in the economizer and the energy-momentum balance of the two streams to obtain a mechanism model. Calculating an ideal feedwater flow resistance coefficient using first field data, wherein the first field data includes an economizer inlet and outlet flue gas pressure drop, a feedwater flow rate, a feedwater temperature, an outlet water temperature, and an economizer inlet and outlet height difference; The second field data is standardized, and then the PCC and MI methods are used to weight the DNN model input features; The second field data includes furnace coal quantity, feed water flow, temperature, pressure, inlet flue gas temperature, pressure, flow; Using the DNN model input features and the ideal water flow resistance coefficient as output features to train the DNN model, and using Bayesian optimization to tune the DNN model structure and hyperparameters; Export the weights, biases and activation functions of neurons in each layer of the tuned DNN model; The weight, bias and activation function are written into the mechanism model, a new function is created in the mechanism model, and multiple parameters are set to store the weight and bias matrix, and the mean and variance of the 6-dimensional input features; Use Modelica language in the newly created function to rebuild the mapping relationship between neurons in each layer of the DNN model; The newly created function is called through the mechanism model to realize parameter identification of the water flow resistance coefficient using the DNN model in the mechanism model.

2. The method according to claim 1, characterized in that Modelica was used to construct equations for heat transfer and momentum balance of flue gas and water in economizers, including: The flue gas and water streams are obtained, represented by subscripts g and w respectively. The three parameters of the two streams, temperature, pressure, and flow rate, are represented by T, p, and qm respectively. The input and output streams are distinguished by subscripts i and o. The heat transfer equations of flue gas and water are constructed as follows: ； ； in, is the average temperature difference between the two streams, is the heat transfer power, the coefficient Fitting through training data; Construct the flue gas momentum balance equation: ； in, It is the fixed pressure drop caused by the height of the flue gas section. is the flow resistance coefficient of flue gas, is the smoke density; The momentum balance equation for water is: ； in, is the height difference between the economizer stream water inlet and outlet; Construct the energy balance equation: ； in, is the internal volume of the economizer, is the specific heat of flue gas.

3. The method according to claim 1, characterized in that The DNN model is used to identify the parameters of the water supply flow resistance coefficient in the mechanism model, and the standardized training data is input. The DNN model contains an input layer, a hidden layer and an output layer. The input feature dimension in the DNN model is 6, including the furnace coal amount, the economizer water supply flow rate and pressure, the inlet flue gas temperature, pressure and flow rate. The output feature dimension is 1, which is the water supply flow resistance coefficient. The number of intermediate hidden layers and the number of neurons in each layer are obtained through Bayesian optimization. In the DNN model, the data is transmitted to the hidden layer through the input layer, and then passes through all the hidden layers to finally reach the output layer, completing a forward transmission. Each neuron receives the output from the previous layer, and the output result is obtained after weighted summation through the activation function ReLU.

4. The method according to claim 1, characterized in that: The reconstructed mapping relationship is as follows: Represents the result after the model input is standardized. Represents the mapping result of the nth layer of neurons, Represents the weight matrix of the nth layer of neurons; is the bias matrix of the nth layer of neurons, represents the final mapping result, i.e. the required water flow resistance coefficient prediction result, where It means that each element in the matrix A is compared with 0 and the larger value is taken, which is the calculation method of ReLU. There is no ReLU in the last output layer, so there is only linear mapping. ReLU is the activation function.

5. A hybrid modeling device for economizer parameter identification based on Modelica, characterized in that: include: Modeling module, used to model the heat transfer process between flue gas and water in economizer and the energy-momentum balance of two streams using Modelica to obtain the mechanism model; A first calculation module is used to calculate an ideal feedwater flow resistance coefficient using first field data, wherein the first field data includes an economizer inlet and outlet flue gas pressure drop, a feedwater flow rate, a feedwater temperature, an outlet water temperature, and an economizer inlet and outlet height difference; The second calculation module is used to standardize the second field data and then use the PCC and MI methods to weightedly calculate the DNN model input features; The second field data includes furnace coal quantity, feed water flow, temperature, pressure, inlet flue gas temperature, pressure, flow; A training optimization module, used to train the DNN model using the DNN model input features and the ideal water flow resistance coefficient as output features, and to tune the DNN model structure and hyperparameters using Bayesian optimization; The export module is used to export the weights, biases and activation functions of neurons in each layer of the tuned DNN model; A writing module, used to write the weights, biases and activation functions into the mechanism model, create a new function in the mechanism model, and set multiple parameters to store weight and bias matrices, and the mean and variance of 6-dimensional input features; The reconstruction module is used to use the Modelica language in the newly created function to rebuild the mapping relationship between neurons in each layer of the DNN model; The calling module is used to call the newly created function through the mechanism model to realize parameter identification of the water flow resistance coefficient using the DNN model in the mechanism model.

6. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instruction is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images