Knowledge-embedded-learning-based electric induction furnace lining state evolution model construction method and device, storage medium and product

Abstract

The invention discloses an electric induction furnace lining state evolution model construction method based on knowledge embedding learning, and the method comprises the following steps: S1, constructing a furnace lining state evolution model, employing a hierarchical model architecture, and embedding multiple forms of knowledge in the induction heating field into different levels of the model; s2, constructing a knowledge embedded learning method, integrating knowledge in the induction heating field through a constraint optimization form, and converting a modeling problem based on knowledge embedded learning into a constraint optimization problem; and S3, constructing a comprehensive weighted loss function of data, physics and domain knowledge, and solving a constraint optimization problem to obtain a furnace lining state evolution model. And the furnace lining state characterization quantity output by the model is utilized to realize perception and quantitative evaluation of furnace lining state evolution. The furnace lining state evolution can be quantitatively represented according to the actual operation data of the system, so that a physically interpretable decision support is provided for furnace lining maintenance.

Description

Technical Field

[0001] This invention relates to a method for constructing a state evolution model of an induction furnace lining, specifically a method for constructing a state evolution model of an induction furnace lining based on knowledge embedding learning. Background Technology

[0002] Induction furnaces are widely used in steel mills and the foundry industry. Compared to traditional melting equipment, induction furnaces utilize electromagnetic induction heating, offering advantages such as rapid heating, high energy efficiency, and low pollution emissions. As the manufacturing industry accelerates its transformation towards digitalization and intelligence, it is necessary to establish intelligent monitoring and condition assessment methods for induction furnaces to improve production quality and operational efficiency. This ensures efficient and stable operation of the smelting process and supports the intelligent upgrading of manufacturing plants. The furnace lining, as a key structural component, is continuously consumed during long-term smelting, altering electromagnetic coupling conditions and causing system state evolution, thereby affecting energy efficiency and operational safety. Currently, industrial practice lacks unified quantitative indicators for furnace lining condition, and maintenance and operational decisions largely rely on experience, easily introducing uncertainty, reducing production efficiency, and increasing safety risks. Therefore, it is necessary to design a method that can automatically sense the evolution of the furnace lining condition.

[0003] Patent CN120068615A discloses a method for identifying electrical parameters of induction furnaces based on a hierarchical physical information learning model. By fusing an equivalent circuit model with a neural network, the method uses system sampling data to identify electrical parameters of induction furnaces. However, this method does not address the evolution of system parameters and furnace lining status caused by the gradual consumption of furnace lining during long-term smelting. Summary of the Invention

[0004] To address the problems and needs in the background technology, this invention provides a method for constructing a model of the state evolution of an induction furnace lining based on knowledge embedding learning. This method embeds various forms of induction heating domain knowledge into different levels of the model, constructing a hierarchical model architecture based on knowledge embedding learning. Based on the proposed model architecture, a modeling method based on knowledge embedding learning is constructed. This method integrates induction heating domain knowledge through constrained optimization, transforming the modeling problem based on knowledge embedding learning into a constrained optimization problem. By constructing a weighted loss function that integrates data, physics, and domain knowledge, the optimization problem is solved. Finally, using the unified-scale furnace lining state characterization quantity output by the model, the perception and quantitative evaluation of the furnace lining state evolution are achieved.

[0005] The technical solution of the present invention is as follows:

[0006] A method for constructing a state evolution model of an induction furnace lining based on knowledge embedding learning includes the following steps:

[0007] S1: Construct a furnace lining state evolution model. This model adopts a hierarchical model architecture to embed various forms of induction heating domain knowledge into different levels of the model.

[0008] S2: Construct a knowledge embedding learning method to integrate induction heating domain knowledge through constraint optimization, transforming the problem of constructing a furnace lining state evolution model based on knowledge embedding learning into a constraint optimization problem;

[0009] S3: Construct a weighted loss function integrating data, physics, and domain knowledge to solve the constrained optimization problem and obtain model parameters, thus deriving a furnace lining state evolution model. Utilize the unified-scale furnace lining state characterization quantity output by the model to achieve perception and quantitative assessment of the furnace lining state evolution.

[0010] In S1, the hierarchical model architecture consists of a furnace-level sharing layer, an intra-furnace sharing layer, and a frequency task layer. The furnace-level sharing layer adopts a fully connected neural network. To extract common features across multiple furnace batches, for a multi-furnace data input set The features extracted from the shared layer of each furnace are represented as follows:

[0011]

[0012] in, For the shared feature set of furnaces, These are the network parameters for the furnace-sharing layer. Furnace index based on samples. Through the furnace selector Will Routing to the corresponding furnace-shared layer:

[0013]

[0014]

[0015] in, It is a specific batch Furnace shared feature set, This is the furnace batch The frequency corresponding to the i-th operating point within the range. Indicates furnace number The shared characteristics of the furnace corresponding to the i-th working point. For furnace The set of frequencies corresponding to the effective operating points within the range. The number of effective working points. =1,2,3 , This is the furnace batch A dedicated set of shared features within the furnace. For furnace A dedicated shared layer within the furnace is used to extract common features within the same furnace cycle. These are the network parameters of the shared layer within the furnace. Based on the frequency of the operating point data, via a frequency selector... Will Routing to the corresponding frequency task layer:

[0016]

[0017]

[0018] in, This is the furnace batch The shared features within the furnace corresponding to the i-th working point. It is the model output. For furnace The frequency task layer corresponding to the i-th operating point is used to extract the physical features of the operating point. These are the network parameters for the task layer at this frequency. The final overall mapping relationship of the furnace lining state evolution model architecture is as follows:

[0019]

[0020] in, Output a set for the model. This represents the set of all furnace indices. Indicates the number of furnace cycles.

[0021] In S2, for furnace batch Sampling data of the i-th working point A furnace lining state evolution model is used to capture the complex nonlinear relationships between system variables:

[0022]

[0023] in, For model output, This represents a multi-layered nonlinear mapping implemented by the furnace lining state evolution model architecture.

[0024] For furnace For the i-th operating point, the differential equation of the control circuit is established as follows:

[0025]

[0026]

[0027]

[0028]

[0029] in, and These are capacitors connected in parallel and in series, respectively. For leakage sensation, For mutual intuition, For iron loss resistance, For copper loss resistors, For smelting resistors, and The values ​​measured after rectification and inversion of the medium-frequency induction power supply are respectively... and Voltage at both ends, It refers to the variable and its derivative term. It is an electrical parameter vector.

[0030] For furnace At the i-th operating point, the physical equivalent circuit model of the induction furnace involving the electrical parameters to be estimated is established using the output of the furnace lining state evolution model as follows:

[0031]

[0032] in, It is a physical operator used to constrain the electrical parameters and variable derivatives estimated by the model to satisfy the physical control equations. yes The estimate, It is the vector of electrical parameters to be identified in the parameterization of the network model. It is set as the frequency task layer. Learnable model parameters :

[0033]

[0034]

[0035] in, Indicates frequency task layer parameters, This represents the parameters for frequency-based task feature extraction.

[0036] Based on knowledge in the field of induction heating, for furnace batches The i-th working point, mutual inductance and smelting resistance and the frequency of the working point The following proportional relationship exists:

[0037]

[0038] in, It is a frequency index used to describe the common physical scaling laws followed by the system under different operating conditions, reflecting the overall frequency dependence characteristics of the induction heating process. It is a furnace lining characteristic index, and its value can quantitatively characterize the health status of the furnace lining.

[0039] For furnace Define the intra-furnace knowledge operator for the i-th working point. as follows:

[0040]

[0041] in, and These are smelting resistors and mutual induction The estimate, This is the furnace batch The average value of the furnace lining characteristic index for all frequency tasks.

[0042] For furnace Define operators as follows:

[0043]

[0044] in, It is the frequency index The estimate, and These represent the electrical parameter estimation term and the frequency term, respectively. This is the zero-mean projection operator. Further, the following global frequency exponential consistency operator can be obtained. :

[0045]

[0046] in, Indicates the number of furnace runs. Set as a shared layer for furnaces Learnable model parameters :

[0047]

[0048]

[0049] in, Indicates the parameters of the shared layer for each furnace. This indicates the shared feature extraction parameters for each furnace.

[0050] Specifically, S3 is:

[0051] First, for multi-batch furnace datasets The ideal form of the proposed knowledge embedding learning model can be expressed as follows:

[0052]

[0053] in, This is the furnace batch The actual measured value at the i-th working point.

[0054] Next, the formula for the loss function of the proposed model is as follows:

[0055]

[0056]

[0057]

[0058]

[0059] in, It is the data modeling loss. It is the loss of physical modeling. It is the domain knowledge modeling loss. It is the loss monitored by the melting resistance. This is the furnace batch The pseudo-physical reference value of the melting resistance at the i-th operating point. It is the monitoring frequency band of furnace batch b. The number of frequencies included.

[0060] Then, the formulas for the total loss function and the total optimization objective are as follows:

[0061]

[0062]

[0063] in, These are the weights of data loss, physical loss, domain knowledge loss, and resistance supervision loss, respectively. These are the network parameters of all shared layers within the furnace in the furnace lining state evolution model. These are the network parameters for all frequency task layers of the furnace lining state evolution model.

[0064] Finally, the equivalent parameters obtained from the optimized furnace lining state evolution model and and frequency index estimation Calculate using the following formula:

[0065]

[0066] in, For furnace The furnace lining characteristic index estimate at the i-th operating point within the furnace is obtained by weighted averaging the furnace lining characteristic index estimates at all operating points within the furnace. Based on the comparability of this indicator across furnace cycles and its evolutionary pattern as the furnace progresses, the ability to perceive and quantify the evolution of the furnace lining condition is realized.

[0067] Another technical solution of the present invention is:

[0068] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method.

[0069] Another technical solution of the present invention is:

[0070] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method.

[0071] Another technical solution of the present invention is:

[0072] A computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the method.

[0073] Compared with the prior art, the present invention has the following beneficial effects:

[0074] (1) The furnace lining state evolution model designed in this invention based on knowledge embedding learning constructs a hierarchical model architecture based on knowledge embedding learning, embedding various forms of induction heating domain knowledge into different levels of the model, capturing common features across furnace batches while retaining the differences in tasks at each working point, thereby ensuring the consistency and comparability of the furnace lining health state representation corresponding to different furnace batches.

[0075] (2) The furnace lining state evolution model designed in this invention constructs a modeling method based on knowledge embedding learning. It integrates knowledge in the field of induction heating through constraint optimization, transforming the modeling problem based on knowledge embedding learning into a constraint optimization problem, thereby enhancing the physical consistency between furnaces and within furnaces and ensuring the coordination and unity between local dynamic behavior and global state representation.

[0076] (3) The furnace lining state evolution model designed in this invention based on knowledge embedding learning constructs a comprehensive weighted loss function of data, physics and domain knowledge to solve the constrained optimization problem. It can automatically output a uniform scale furnace lining state characterization quantity based on smelting data, realize the perception and quantitative evaluation of furnace lining state evolution, and is suitable for practical engineering applications. Attached Figure Description

[0077] Figure 1 This is a diagram illustrating the furnace lining state evolution model proposed in this invention.

[0078] Figure 2 This is a schematic diagram of the knowledge embedding learning method for the furnace lining state evolution model proposed in this invention.

[0079] Figure 3 This is a modeling accuracy diagram of the furnace lining state evolution model proposed in the present invention in the embodiments;

[0080] Figure 4 This is a diagram illustrating the physical modeling accuracy of the furnace lining state evolution model proposed in this invention, as shown in the embodiments.

[0081] Figure 5 This is a diagram showing the global frequency index consistency results of the furnace lining state evolution model proposed in this invention, as illustrated in the embodiments.

[0082] Figure 6 This is a diagram showing the results of the furnace lining state evolution model proposed in this invention for modeling domain knowledge within a furnace cycle, as presented in the embodiments.

[0083] Figure 7 This is a graph showing the furnace lining state evolution perception results of the furnace lining state evolution model proposed in this embodiment. Detailed Implementation

[0084] To enable those skilled in the art to more clearly understand the purpose, technical solution and advantages of the present invention, the present invention will be further described below in conjunction with the accompanying drawings and embodiments, but the present invention is not limited to the following embodiments.

[0085] In this embodiment, a medium-frequency induction furnace with a melting rate of 35 t / h was selected for research.

[0086] This invention provides a method for constructing a state evolution model of an induction furnace lining based on knowledge embedding learning, comprising the following steps:

[0087] S1: Construct a furnace lining state evolution model. This model adopts a hierarchical model architecture, embedding various forms of induction heating domain knowledge into different levels of the model.

[0088] Specifically:

[0089] The hierarchical model architecture proposed in this invention is as follows: Figure 1 As shown, the system consists of a furnace-level sharing layer, an in-furnace sharing layer, and a frequency task layer. The furnace-level sharing layer employs a fully connected neural network. To extract common features across multiple furnace batches, for a multi-furnace data input set The features extracted from the shared layer of each furnace are represented as follows:

[0090]

[0091] in, For the shared feature set of furnaces, These are the network parameters for the furnace-sharing layer. Furnace index based on samples. Through the furnace selector Will Routing to the corresponding furnace-shared layer:

[0092]

[0093]

[0094] in, It is a specific batch Furnace shared feature set, This is the furnace batch The frequency corresponding to the i-th operating point within the range. Indicates furnace number The shared characteristics of the furnace corresponding to the i-th working point. For furnace The set of frequencies corresponding to the effective operating points within the range. The number of effective working points. This is the furnace batch A dedicated set of shared features within the furnace. For furnace A dedicated shared layer within the furnace is used to extract common features within the same furnace cycle. These are the network parameters of the shared layer within the furnace. Based on the operating point frequency, via a frequency selector. Will Routing to the corresponding frequency task layer:

[0095]

[0096]

[0097] in, This is the furnace batch The shared features within the furnace corresponding to the i-th working point. It is the model output. For furnace The frequency task layer corresponding to the i-th operating point is used to extract the physical features of the operating point. These are the network parameters for the task layer at that frequency.

[0098] The final overall mapping relationship of the furnace lining state evolution model architecture is as follows:

[0099]

[0100] in, Output a set for the model. This represents the set of all furnace indices. Indicates the number of furnace cycles.

[0101] S2: Construct a knowledge embedding learning method to integrate induction heating domain knowledge through constraint optimization, transforming the problem of constructing a furnace lining state evolution model based on knowledge embedding learning into a constraint optimization problem;

[0102] Specifically:

[0103] The knowledge embedding learning method for the furnace lining state evolution model proposed in this invention is as follows: Figure 2 As shown, for furnace batches The sampling data corresponding to the i-th working point A furnace lining state evolution model is used to capture the complex nonlinear relationships between system variables:

[0104]

[0105] in, For model output, This represents a multi-layered nonlinear mapping implemented by the furnace lining state evolution model architecture.

[0106] For furnace For the i-th operating point, the differential equation of the control circuit is established as follows:

[0107]

[0108]

[0109]

[0110]

[0111] in, and These are capacitors connected in parallel and in series, respectively. For leakage sensation, For mutual intuition, For iron loss resistance, For copper loss resistors, For smelting resistors, and The values ​​measured after rectification and inversion of the medium-frequency induction power supply are respectively... and Voltage at both ends, It is a variable ( and ) and its derivative terms, It is an electrical parameter vector, and T is the transpose of the matrix.

[0112] For furnace At the i-th operating point, the physical equivalent circuit model of the induction furnace involving the electrical parameters to be estimated is established using the output of the furnace lining state evolution model as follows:

[0113]

[0114] in, It is a physical operator used to constrain the electrical parameters and variable derivatives estimated by the model to satisfy the physical control equations. yes The estimate, It is the vector of electrical parameters to be identified in the parameterization of the network model. It is set as the frequency task layer. Learnable model parameters :

[0115]

[0116]

[0117] in, Indicates frequency task layer parameters, This represents the parameters for frequency-based task feature extraction.

[0118] Based on knowledge in the field of induction heating, for furnace batches The i-th working point, mutual inductance and smelting resistance and the frequency of the working point The following proportional relationship exists:

[0119]

[0120] in, It is a frequency index used to describe the common physical scaling laws followed by the system under different operating conditions, reflecting the overall frequency dependence characteristics of the induction heating process. It is a furnace lining characteristic index, and its value can quantitatively characterize the health status of the furnace lining.

[0121] For furnace Define the intra-furnace knowledge operator for the i-th working point. as follows:

[0122]

[0123] in, and These are smelting resistors and mutual induction The estimate, This is the furnace batch The average value of the furnace lining characteristic index for all frequency tasks.

[0124] For furnace Define operators as follows:

[0125]

[0126] in, It is the frequency index The estimate, and These represent the electrical parameter estimation term and the frequency term, respectively. This is the zero-mean projection operator. Further, the following global frequency exponential consistency operator can be obtained. :

[0127]

[0128] in, Indicates the number of furnace runs. Set as a shared layer for furnaces Learnable model parameters :

[0129]

[0130]

[0131] in, Indicates the parameters of the shared layer for each furnace. This indicates the shared feature extraction parameters for each furnace.

[0132] S3: Construct a weighted loss function integrating data, physics, and domain knowledge to solve the constrained optimization problem and obtain model parameters, thus deriving a furnace lining state evolution model. Utilize the unified-scale furnace lining state characterization quantity output by the model to achieve perception and quantitative assessment of the furnace lining state evolution.

[0133] S3 specifically refers to:

[0134] First, for multi-batch furnace datasets The ideal form of the proposed knowledge embedding learning model can be expressed as follows:

[0135]

[0136] in, This is the furnace batch The actual measured value corresponding to the i-th working point. It is the vector of electrical parameters estimated by the furnace lining state evolution model. It is the frequency index estimated by the furnace lining state evolution model.

[0137] Next, the formula for the loss function of the proposed model is as follows:

[0138]

[0139]

[0140]

[0141]

[0142] in, It is the data modeling loss. It is the loss of physical modeling. It is the domain knowledge modeling loss. It is the loss monitored by the melting resistance. This is the furnace batch The pseudo-physical reference value of the melting resistance at the i-th operating point. It is the monitoring frequency band of furnace batch b. The number of frequencies included.

[0143] Then, the formulas for the total loss function and the total optimization objective are as follows:

[0144]

[0145]

[0146] in, These are the weights of data loss, physical loss, domain knowledge loss, and resistance supervision loss, respectively. These are the network parameters of all shared layers within the furnace in the furnace lining state evolution model. These are the network parameters for all frequency task layers of the furnace lining state evolution model.

[0147] Finally, the equivalent parameters obtained from the optimized furnace lining state evolution model and and frequency index estimation Calculate using the following formula:

[0148]

[0149] in, For furnace The furnace lining characteristic index estimate at the i-th operating point within the furnace is obtained by weighted averaging the furnace lining characteristic index estimates at all operating points within the furnace. Based on the comparability of this indicator across furnace cycles and its evolutionary pattern as the furnace progresses, the ability to perceive and quantify the evolution of the furnace lining condition is realized.

[0150] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above method.

[0151] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of the above-described method.

[0152] The present invention also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0153] The technical solution of the present invention will be described in detail below with reference to specific embodiments.

[0154] The sampling data for the induction furnace operation used in this invention lasts approximately 90 to 100 minutes per complete furnace cycle, during which the system's operating frequency gradually increases from approximately 100 Hz to approximately 140 Hz. For each furnace cycle, the sampling frequency is 20000 Hz, the sampling interval is 1 minute, and the number of data points per sampling point is 1024. The specific configuration used for training the knowledge embedding learning model proposed in this invention is shown in Table 1. For this embodiment, , Total inductance , , .

[0155] Table 1 shows the specific configuration for training the furnace lining state evolution model.

[0156] category Configuration Furnace Set Furnace count 13 Frequency task range Resistance monitoring frequency band Optimizer Adam Activation function ReLU Learning rate Network structure (furnace sharing layer; furnace sharing layer; frequency task layer) [4,32,32, 32, 32]; [32, 32]; [32, 32, 2] Initial frequency index 0.5 Number of task sample points 800 Training data Top 400 Test data The last 400 Weight Maximum number of iterations 1000

[0157] This invention defines the mean absolute error (MAE) and root mean square error (RMSE) as follows to evaluate the model's performance, as shown in the formulas below:

[0158]

[0159]

[0160] in, These are observed values. It is an estimated value.

[0161] The modeling accuracy results of the furnace lining state evolution model proposed in this invention are as follows: Figure 3As shown in the figure. The results show that the method proposed in this invention maintains high fitting accuracy under multi-furnace conditions, significantly enhances the consistency and stability of the model, and provides reliable data estimation for subsequent physical modeling.

[0162] The physical modeling accuracy results of the furnace lining state evolution model proposed in this invention are as follows: Figure 4 As shown. The results indicate that the physical losses of the proposed method remain at a low level across all furnace cycles, and the model effectively satisfies the equivalent circuit constraints.

[0163] The global frequency exponential consistency results of the furnace lining state evolution model proposed in this invention are as follows: Figure 5 As shown in the figure. The results demonstrate that the method proposed in this invention can automatically learn the potential exponential relationship between frequency and electrical parameters, verifying the effectiveness of knowledge modeling.

[0164] The furnace lining state evolution model proposed in this invention provides the following in-furnace knowledge modeling results: Figure 6 As shown. The results indicate that, compared to methods based on physical models, the method proposed in this invention calculates... Maintaining consistency across all work points results in better physical consistency.

[0165] The furnace lining state evolution model proposed in this invention provides the following furnace lining state evolution perception results: Figure 7 As shown in the figure. The perception results demonstrate that the method proposed in this invention achieves a physically interpretable quantitative characterization of the furnace lining health state, establishes a unified physical scale for characterizing the impact of furnace lining degradation on the system state evolution, and verifies the effectiveness and engineering applicability of the method proposed in this invention.

[0166] Finally, it should be noted that the above embodiments and descriptions are only used to illustrate the technical solutions of the present invention and not to limit it. Those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the disclosure of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for constructing a state evolution model of an induction furnace lining based on knowledge embedding learning, characterized in that, Includes the following steps: S1: Construct a furnace lining state evolution model. This model adopts a hierarchical model architecture, embedding various forms of induction heating domain knowledge into different levels of the model. S2: Construct a knowledge embedding learning method to integrate induction heating domain knowledge through constraint optimization, transforming the problem of constructing a furnace lining state evolution model based on knowledge embedding learning into a constraint optimization problem; S3: Construct a weighted loss function that integrates data, physics, and domain knowledge to solve the constrained optimization problem and obtain model parameters, thereby obtaining a furnace lining state evolution model. Utilize the unified-scale furnace lining state characterization quantity output by the model to realize the perception and quantitative evaluation of the furnace lining state evolution.

2. The method for constructing a state evolution model of an induction furnace lining based on knowledge embedding learning according to claim 1, characterized in that, In S1, the hierarchical model architecture consists of a furnace-level sharing layer, an intra-furnace sharing layer, and a frequency task layer. The furnace-level sharing layer adopts a fully connected neural network. To extract common features across multiple furnace batches, for a multi-furnace data input set The features extracted from the shared layer of each furnace are represented as follows: in, For the shared feature set of furnaces, Network parameters for the shared layer of furnaces; Sample-based furnace index Through the furnace selector Will Routing to the corresponding furnace-shared layer: in, It is a specific batch Furnace shared feature set, This is the furnace batch The frequency corresponding to the i-th operating point within the range. Indicates furnace number The shared characteristics of the furnace corresponding to the i-th working point. For furnace The set of frequencies corresponding to the effective operating point within the range. The number of effective working points. =1,2,3 , This is the furnace batch A dedicated set of shared features within the furnace. For furnace A dedicated shared layer within the furnace is used to extract common features within the same furnace cycle. These are the network parameters of the shared layer within the furnace; Based on the operating point frequency, via a frequency selector Will Routing to the corresponding frequency task layer: in, This is the furnace batch The shared features within the furnace corresponding to the i-th working point. It is the model output. For furnace The frequency task layer corresponding to the i-th operating point is used to extract the physical features of the operating point. These are the network parameters of the task layer at that frequency; The final overall mapping relationship of the furnace lining state evolution model architecture is as follows: in, Output a set for the model. This represents the set of all furnace indices. Indicates the number of furnace cycles.

3. The method for constructing a state evolution model of an induction furnace lining based on knowledge embedding learning according to claim 2, characterized in that, In S2, for furnace batch Sampling data of the i-th working point A furnace lining state evolution model is used to capture the complex nonlinear relationships between system variables: in, For model output, This represents a multi-layered nonlinear mapping implemented by the furnace lining state evolution model architecture.

4. The method for constructing a state evolution model of an induction furnace lining based on knowledge embedding learning according to claim 3, characterized in that, In S2, for furnace batch For the i-th operating point, the physical equivalent circuit model of the induction furnace is established using the output of the furnace lining state evolution model as follows: in, It is a physical operator used to constrain the electrical parameters and variable derivatives estimated by the model to satisfy the physical control equations. These are the model output variables and their derivative terms. It is the vector of electrical parameters estimated by the model.

5. The method for constructing a state evolution model of an induction furnace lining based on knowledge embedding learning according to claim 4, characterized in that, In step S2, based on knowledge of the field of induction heating, for the furnace batch... The i-th working point, mutual inductance and smelting resistance and the frequency of the working point The following proportional relationship exists: in, It is a frequency index used to describe the common physical scaling laws followed by the system under different operating conditions, reflecting the overall frequency dependence characteristics of the induction heating process. It is a furnace lining characteristic index, and its value can quantitatively characterize the health status of the furnace lining. For furnace Define the intra-furnace knowledge operator for the i-th working point. as follows: in, and These are smelting resistors and mutual induction The estimate, This is the furnace batch The average value of the furnace lining characteristic index for all frequency tasks within the system; For furnace Define the frequency exponential consistency operator as follows: in, It is the frequency index The estimate, and These represent the electrical parameter estimation term and the frequency term, respectively. The zero-mean projection operator; Furthermore, the following global frequency exponential consistency operator can be obtained. : in, Indicates the number of furnace runs. Set as a shared layer for furnaces Learnable model parameters : in, Indicates the parameters of the shared layer for each furnace. This indicates the shared feature extraction parameters for each furnace.

6. The method for constructing a state evolution model of an induction furnace lining based on knowledge embedding learning according to claim 5, characterized in that, Specifically, S3 is: First, for multi-batch furnace datasets The ideal form of the proposed furnace lining state evolution model can be expressed as follows: in, This is the furnace batch The actual measured value at the i-th working point. It is the vector of electrical parameters estimated by the furnace lining state evolution model. It is the frequency index estimated by the furnace lining state evolution model; Next, the formula for the loss function of the proposed model is as follows: in, It is the data modeling loss. It is the loss of physical modeling. It is the domain knowledge modeling loss. It is the loss monitored by the melting resistance. This is the furnace batch The pseudo-physical reference value of the melting resistance at the i-th operating point. It is the monitoring frequency band of furnace batch b. The number of frequencies included; Then, the formulas for the total loss function and the total optimization objective are as follows: in, These are the weights of data loss, physical loss, domain knowledge loss, and resistance supervision loss, respectively. These are the network parameters of all shared layers within the furnace in the furnace lining state evolution model. These are the network parameters of all frequency task layers in the furnace lining state evolution model; Finally, the equivalent parameters obtained from the optimized furnace lining state evolution model and and frequency index estimation Calculate using the following formula: in, For furnace The furnace lining characteristic index estimate at the i-th operating point within the furnace is obtained by weighted averaging the furnace lining characteristic index estimates at all operating points within the furnace. Based on the comparability of this indicator across furnace cycles and its evolutionary pattern as the furnace progresses, the ability to perceive and quantify the evolution of the furnace lining condition is realized.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1 to 6.

Images