Method for measuring dynamic characteristics of a reactor

Abstract

The application provides a method for measuring dynamic characteristics of a reactor, comprising: dividing each side surface of a plurality of first reactors of the same type into a plurality of detection areas, arranging a signal acquisition point in the center of each detection area to arrange a signal acquisition sensor; driving all signal acquisition sensors to perform signal acquisition according to a set signal acquisition control strategy to generate a first electric reactance dynamic characteristic signal sequence; training a first pre-trained neural network model and a second pre-trained neural network model based on the first electric reactance dynamic characteristic signal sequence to obtain a single-point electric reactance dynamic characteristic analysis model and a whole electric reactance dynamic characteristic analysis model; and obtaining a second electric reactance dynamic characteristic signal sequence obtained by performing signal acquisition on a second electric reactor and inputting the second electric reactance dynamic characteristic signal sequence into the single-point electric reactance dynamic characteristic analysis model and the whole electric reactance dynamic characteristic analysis model to predict the electric reactance dynamic characteristics of the second electric reactor.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and specifically to a method for measuring the dynamic characteristics of a reactor. Background Technology

[0002] As crucial reactive power compensation and voltage regulation devices in power systems, reactors' dynamic characteristics have a vital impact on the stability and security of power systems. However, current methods for measuring the dynamic characteristics of reactors still have many shortcomings, mainly in the following aspects: First, traditional reactor dynamic characteristic measurements often rely on a single sensor or a limited number of sensors for signal acquisition. This method struggles to comprehensively capture the dynamic changes of the reactor from all aspects, resulting in inaccurate and incomplete measurement results. Since the dynamic characteristics of a reactor involve multiple aspects such as electromagnetic fields, stress distribution, and vibration, a single or limited sensor layout can hardly cover all these factors, thus affecting the accuracy and reliability of the measurement.

[0003] Secondly, existing measurement methods often lack intelligent analysis and prediction capabilities. Traditional measurement methods typically involve simply collecting data and then relying on manual analysis and processing. This is not only inefficient but also makes it difficult to predict and evaluate the dynamic characteristics of reactors in real time and accurately. During the operation of a power system, the state of reactors may change at any time. Therefore, the lack of intelligent analysis and prediction capabilities will seriously affect the stability and security of the power system.

[0004] Finally, existing measurement methods are insufficient for measuring the dynamic characteristics of reactors under different fault conditions. In actual operation, reactors may encounter various fault conditions, such as short circuits and overloads, which significantly affect their dynamic characteristics. However, existing measurement methods often fail to accurately measure and predict the dynamic characteristics under these fault conditions, thus posing difficulties for fault diagnosis and troubleshooting in power systems. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a method for measuring the dynamic characteristics of a reactor, thereby at least solving or mitigating the problems existing in the prior art.

[0006] This application provides a method for measuring the dynamic characteristics of a reactor, comprising: Divide each side of multiple first reactors of the same model into several detection areas, set a signal acquisition point at the center of each detection area, and set a signal acquisition sensor at each signal acquisition point. Drive all signal acquisition sensors to acquire signals according to the set signal acquisition control strategy to generate the first reactance dynamic characteristic characterization signal sequence; Based on the first reactor dynamic characteristic characterization signal sequence, the first pre-trained neural network model and the second pre-trained neural network model are trained to obtain the single-point reactor dynamic characteristic analysis model and the overall reactor dynamic characteristic analysis model, respectively. Obtain the dynamic characteristic characterization signal sequence of the second reactor obtained by signal acquisition of the second reactor; The second reactor dynamic characteristic characterization signal sequence is input into the single-point reactor dynamic characteristic analysis model and the overall reactor dynamic characteristic analysis model to predict the reactor dynamic characteristics of the second reactor.

[0007] This invention proposes a novel method for measuring the dynamic characteristics of reactors. This method divides each side of multiple reactors of the same model into several detection zones, and sets a signal acquisition point at the center of each detection zone to comprehensively capture the dynamic changes of the reactors on all sides. Simultaneously, this method utilizes a first pre-trained neural network model and a second pre-trained neural network model to analyze and predict the dynamic characteristics of single-point reactors and the overall reactor dynamic characteristics, respectively, thereby improving the accuracy and intelligence of the measurement. Furthermore, this method can also measure and predict the dynamic characteristics of reactors under different fault conditions, providing strong support for fault diagnosis and troubleshooting in power systems. Attached Figure Description

[0008] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of the accompanying drawings are used to explain the application and do not constitute an undue limitation of the application. In the drawings: Figure 1 This is a flowchart illustrating a method for measuring the dynamic characteristics of a reactor according to this application. Detailed Implementation

[0009] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present application.

[0010] Figure 1 This is a flowchart illustrating a method for measuring the dynamic characteristics of a reactor according to this application. Figure 1 As shown, it includes: Divide each side of multiple first reactors of the same model into several detection areas, set a signal acquisition point at the center of each detection area, and set a signal acquisition sensor at each signal acquisition point. Drive all signal acquisition sensors to acquire signals according to the set signal acquisition control strategy to generate the first reactance dynamic characteristic characterization signal sequence; Based on the first reactor dynamic characteristic characterization signal sequence, the first pre-trained neural network model and the second pre-trained neural network model are trained to obtain the single-point reactor dynamic characteristic analysis model and the overall reactor dynamic characteristic analysis model, respectively. Obtain the dynamic characteristic characterization signal sequence of the second reactor obtained by signal acquisition of the second reactor; The second reactor dynamic characteristic characterization signal sequence is input into the single-point reactor dynamic characteristic analysis model and the overall reactor dynamic characteristic analysis model to predict the reactor dynamic characteristics of the second reactor.

[0011] Optionally, the step of dividing each side of multiple first reactors of the same model into several detection areas, and setting a signal acquisition point at the center of each detection area, before setting a signal acquisition sensor at each signal acquisition point includes: The geometric dimensions of the first reactor are acquired and digitized to obtain the reactor geometric model. Obtain the material property data of the first reactor and map it onto the geometric model of the reactor. The material property data includes electrical conductivity, magnetic permeability, density, elastic modulus, and Poisson's ratio. The operating condition description data of the first reactor is obtained to determine the boundary condition characteristics of its electromagnetic field and structural field. The boundary condition characteristics of the electromagnetic field include electric field boundary condition characteristics and magnetic field boundary condition characteristics. The boundary condition characteristics of the structural field include displacement boundary condition characteristics, force boundary condition characteristics, and thermal boundary condition characteristics. Based on the boundary condition characteristics of the electromagnetic field and structural field, the geometric model of the reactor is meshed to obtain several computational domain meshes, which are then labeled to generate a computational domain mesh set. Based on the mesh set of the computational domain, a finite element solution model corresponding to the first reactor is constructed, and the finite element solution model is solved to obtain the electromagnetic field distribution, stress distribution and vibration characteristic data of the first reactor; Based on the electromagnetic field distribution, stress distribution, and vibration characteristic data of the first reactor, the side surface on the first reactor that divides the detection area is determined.

[0012] The characteristics of electromagnetic field boundary conditions are explained as follows: 1. Electric field boundary conditions: Potential boundary conditions: specify the potential values ​​at certain points or surfaces in an electric field. For example, the potential on certain electrodes or ground surfaces of a reactor may be fixed at a known value.

[0013] Charge boundary conditions: specify the charge distribution at certain boundaries in an electric field. This is typically used to simulate charge accumulation or dissipation on the surface of a reactor.

[0014] Dielectric interface condition: If there is a dielectric interface with different dielectric constants inside or outside the reactor, the continuity condition of the electromagnetic field at the dielectric interface needs to be applied, that is, the tangential component of the electric field intensity E and the normal component of the electric displacement vector D are continuous on both sides of the interface.

[0015] 2. Magnetic field boundary conditions: Magnetic flux density boundary conditions: Specifies the magnetic flux density values ​​at certain boundaries in a magnetic field. This is typically used to simulate magnetic flux paths or magnetic shielding effects in reactors.

[0016] Magnetic vector potential boundary conditions: In magnetic field analysis, magnetic vector potential A is sometimes used to describe the magnetic field. In this case, it is necessary to specify the value or variation law of the magnetic vector potential on the boundary.

[0017] Dielectric interface conditions: Similar to electric fields, magnetic fields also need to satisfy the continuity condition at dielectric interfaces, that is, the tangential component of magnetic field strength H and the normal component of magnetic induction intensity B are continuous on both sides of the interface.

[0018] The characteristics of the structural field boundary conditions are explained as follows: 1. Displacement boundary conditions: Fixed constraints: Specify that the displacement of certain points or surfaces on the reactor is zero to simulate the connection between the reactor and a fixed structure.

[0019] Given displacement: In some cases, it may be necessary to specify known displacement values ​​for certain points or surfaces on the reactor.

[0020] 2. Force boundary conditions: External loads: These include various external forces experienced by the reactor during operation, such as gravity, wind force, and electromagnetic force. These loads need to be applied to the finite element model according to the actual situation.

[0021] Contact forces: If the reactor is in contact with other structures, it is necessary to simulate the contact forces on the contact surfaces. This usually involves complex contact mechanics analysis.

[0022] 3. Thermal boundary conditions: Temperature boundary conditions: specify the known temperature values ​​of certain points or surfaces on the reactor.

[0023] Heat flow boundary condition: Specifies the heat flow exchange relationship between the reactor surface and the external environment.

[0024] Through the above-mentioned optional technical solutions, the reactor dynamic characteristic measurement method of the present invention has the following technical advantages: First, by acquiring the geometric dimensions of the first reactor and constructing a digital geometric model of the reactor, a solid foundation is laid for subsequent accurate measurements and analysis. This step ensures that the measurement process accurately reflects the actual structure and shape of the reactor, avoiding measurement errors caused by inaccurate geometric dimensions.

[0025] Secondly, the material property data of the first reactor (such as conductivity, permeability, density, elastic modulus, Poisson's ratio, etc.) are mapped onto the reactor's geometric model, allowing the measurement process to fully consider the influence of material properties on the reactor's dynamic characteristics. This step improves the accuracy and realism of the measurement, making the measurement results closer to actual conditions.

[0026] Next, by acquiring the operating condition description data of the first reactor and determining the boundary condition characteristics of its electromagnetic and structural fields, the necessary inputs were provided for the subsequent finite element solution. This step ensures that the finite element solution model can accurately reflect the dynamic characteristics of the reactor under actual operating conditions, providing a reliable foundation for subsequent measurements and analysis.

[0027] Then, based on the boundary condition characteristics of the electromagnetic and structural fields, the reactor's geometric model is meshed, and a computational domain mesh set is generated. This step, through fine mesh generation, improves the accuracy and efficiency of the finite element solution, making the measurement results more accurate and reliable.

[0028] Furthermore, by constructing and solving the finite element model corresponding to the first reactor, the electromagnetic field distribution, stress distribution, and vibration characteristic data of the first reactor were obtained. These data provided crucial information for determining the side profile of the detection area, ensuring the rationality and effectiveness of the detection area division.

[0029] Finally, based on the electromagnetic field distribution, stress distribution, and vibration characteristic data of the first reactor, the sides of the detection area are determined. This step ensures that the detection area can fully cover the key parts where the reactor's dynamic characteristics change, improving the comprehensiveness and accuracy of the measurement.

[0030] Optionally, based on the boundary condition characteristics of the electromagnetic field and structural field, the reactor geometric model is meshed to obtain several computational domain meshes, which are then labeled to generate a computational domain mesh set, including: Based on the boundary condition characteristics of the electromagnetic field and structural field, the boundary clarity of the reactor geometric model is calculated; Based on boundary clarity, the mapping region for which the reactor geometric model can be divided into mapping meshes is determined; Based on the analysis requirements of the electromagnetic field and structural field, a mapping mesh is generated for the mapping region as a computational domain mesh, and the mapping mesh is labeled. The label includes the mesh type, size, and density. The mesh type represents the geometric regularity of the first reactor, the mesh size represents the geometric characteristics and physical behavior of the first reactor, and the mesh density represents the electromagnetic field strength and stress concentration intensity of the first reactor.

[0031] In the aforementioned technical solution, the process of meshing the reactor geometric model based on the boundary condition characteristics of the electromagnetic field and structural field, and generating a computational domain mesh set, involves several key steps. These steps collectively ensure the accuracy and efficiency of the finite element analysis. The detailed technical benefits are as follows: 1. Determining boundary clarity is a crucial preparatory step before mesh generation. It helps identify key regions in the reactor's geometric model, such as areas of drastic electromagnetic field changes and stress concentration points. Clear boundary information guides subsequent mesh generation, ensuring that the mesh accurately reflects the model's geometric characteristics and physical behavior.

[0032] 2. Based on boundary clarity, determine the mapping regions suitable for mapping mesh generation. This step avoids indiscriminate mesh generation across the entire geometric model, instead focusing on regions that significantly impact the analysis results. This not only improves computational efficiency but also ensures computational accuracy in critical areas.

[0033] 3. The type of mesh (e.g., structured mesh, unstructured mesh, etc.) characterizes the regularity of the reactor's geometry. For regions with regular shapes, structured meshes can improve computational efficiency and accuracy; for regions with complex shapes, unstructured meshes can better adapt to geometric boundaries. The mesh size directly reflects the reactor's geometric characteristics and physical behavior. In areas with drastic electromagnetic field changes or stress concentration, using a smaller mesh can improve computational accuracy; while in relatively flat areas, a larger mesh can save computational resources. The mesh density is closely related to physical quantities such as the reactor's electromagnetic field strength and stress concentration intensity. By reasonably controlling the mesh density, sufficient meshes can be ensured in regions with strong electromagnetic fields or stress concentration to capture changes in physical quantities, thereby improving the accuracy of the calculation results. Labeling the mapped mesh, including information such as mesh type, size, and density, provides clear guidance for subsequent finite element solutions. This not only helps improve computational efficiency but also ensures the reliability and repeatability of the calculation results. Simultaneously, the labeling information also facilitates subsequent analysis and optimization.

[0034] Optionally, constructing the finite element solution model corresponding to the first reactor based on the mesh set partitioned from the computational domain includes: Based on the computational domain mesh set, vector potential cells and scalar potential cells are generated. Based on the vector potential unit and scalar potential unit, construct electromagnetic field unit and structural field unit; Electromagnetic field boundary conditions and structural field boundary conditions are added to the electromagnetic field unit and structural field unit, so that local element equations are formed for each electromagnetic field unit and structural field unit respectively. The local element equations corresponding to all electromagnetic field elements and structural field elements are assembled into a global matrix equation, which serves as the constructed finite element solution model.

[0035] When constructing the finite element solution model corresponding to the first reactor, the technical solution ensures the accuracy and reliability of the model through a series of steps. The following is an explanation of the technical benefits of these steps: 1. Vector potential elements and scalar potential elements are the basic elements used in finite element analysis to describe electromagnetic fields. Vector potential elements accurately reflect the vector characteristics of electromagnetic fields, such as the direction and intensity of the magnetic field; while scalar potential elements are used to describe the scalar characteristics of electric fields, such as the distribution of electric potential. By generating these two types of elements, the electromagnetic field behavior in reactors can be captured comprehensively and accurately.

[0036] 2. In finite element analysis, electromagnetic field elements and structural field elements are used to describe the behavior of electromagnetic and structural fields, respectively. Electromagnetic field elements, constructed based on vector potential elements and scalar potential elements, can accurately simulate the electromagnetic field distribution and interactions within a reactor; while structural field elements are used to describe the structural response of the reactor under the influence of electromagnetic fields, such as stress and deformation. By constructing these two types of elements, a comprehensive analysis of the electromagnetic and structural field behavior of the reactor can be achieved.

[0037] 3. Boundary conditions are a crucial factor in finite element analysis, determining the behavior of electromagnetic and structural fields at the boundaries. Adding accurate boundary conditions to electromagnetic and structural field elements ensures that the finite element solution model accurately reflects the behavior of the reactor under actual operating conditions. This includes boundary conditions for electromagnetic fields (such as electric and magnetic field boundary conditions) and boundary conditions for structural fields (such as displacement, force, and thermal boundary conditions).

[0038] 4. In finite element analysis, each electromagnetic field element and structural field element forms a local element equation, which describes the physical behavior within the element. By assembling all these local element equations into a global matrix equation, a complete finite element solution model can be obtained. This model accurately reflects the overall electromagnetic field distribution and structural response of the reactor, providing a solid foundation for subsequent solutions and analyses.

[0039] Optionally, solving the finite element model to obtain the electromagnetic field distribution, stress distribution, and vibration characteristic data of the first reactor includes: Based on the constructed electromagnetic field and structural field coupled solver, the finite element solution model is solved according to the set solution type and iteration step size to obtain the electric potential intensity distribution, magnetic field intensity distribution, stress tensor distribution, strain distribution, displacement, velocity, and acceleration distribution. The electromagnetic field distribution is obtained by superimposing the electric potential intensity distribution and the magnetic field intensity distribution. Based on the stress-strain relationship, the stress tensor distribution and strain distribution are fused to obtain the stress distribution; The displacement, velocity, and acceleration distributions are subjected to Fourier transform to fuse them in the frequency domain to obtain vibration characteristic data.

[0040] In the aforementioned technical solution, the process of solving the finite element model of the first reactor to obtain its electromagnetic field distribution, stress distribution, and vibration characteristic data demonstrates the following technical advantages: 1. By constructing a coupled solver for electromagnetic and structural fields, a comprehensive simulation of the complex physical phenomena inside the reactor was achieved. The interaction between the electromagnetic and structural fields has a significant impact on the reactor's performance, and the coupled solver can accurately capture this interaction, thereby improving the accuracy of the solution. This coupled solution method is more accurate than solving the electromagnetic or structural fields separately, and can more comprehensively reflect the actual operating state of the reactor.

[0041] 2. By setting appropriate solution types and iteration step sizes based on the nature and requirements of the problem, the stability and convergence of the solution process can be ensured. Different solution types are suitable for different types of physical problems, while the choice of iteration step size directly affects the accuracy and efficiency of the solution. By setting these parameters reasonably, the solution efficiency can be improved while ensuring the accuracy of the solution.

[0042] 3. By superimposing the electric potential and magnetic field intensity distributions, the electromagnetic field distribution of the reactor can be obtained. This field strength superposition method can accurately reflect the distribution of the electromagnetic field inside the reactor, providing an important basis for subsequent electromagnetic field analysis and optimization.

[0043] 4. Based on the stress-strain relationship, the stress tensor distribution and strain distribution can be fused to obtain the stress distribution of the reactor. Stress distribution is an important indicator for evaluating the structural strength and stability of a reactor. The stress distribution data obtained through the finite element method (FEM) model has high accuracy and reliability, providing strong support for the structural design and optimization of reactors.

[0044] 5. Performing Fourier transforms on the displacement, velocity, and acceleration distributions allows for the fusion of these data in the frequency domain to obtain the reactor's vibration characteristic data. Fourier transform is a commonly used signal processing method that converts time-domain signals into frequency-domain signals, thereby revealing the signal's frequency components and vibration characteristics. Obtaining the reactor's vibration characteristic data provides crucial information for its dynamic performance analysis and optimization.

[0045] Optionally, determining the side surface on the first reactor that divides the detection area based on the electromagnetic field distribution, stress distribution, and vibration characteristic data of the first reactor includes: The electromagnetic field distribution is mapped onto the geometric model of the reactor to generate an electromagnetic field intensity distribution map. Based on the electromagnetic field intensity distribution map, the electromagnetic field intensity distribution value and the electromagnetic field change gradient value are statistically analyzed to determine the points corresponding to the electromagnetic field intensity distribution value being greater than a set electromagnetic field intensity distribution threshold and the electromagnetic field change gradient value being greater than the electromagnetic field change gradient threshold. These points are used as seeds for region growth to obtain the first region. The stress distribution is mapped onto the reactor geometric model to generate a stress distribution map. Based on the stress distribution map, the stress concentration degree is evaluated to obtain a stress concentration degree evaluation value. The point corresponding to the stress concentration degree evaluation value being greater than the set stress concentration degree evaluation threshold is determined. The point is used as a seed for region growth to obtain a second region. The vibration characteristic data is mapped onto the geometric model of the reactor to generate a vibration characteristic distribution map. Based on the vibration characteristic distribution map, the vibration peak value and resonant frequency are statistically analyzed to determine the points corresponding to the vibration peak value being greater than a set vibration peak value threshold and the resonant frequency being greater than a resonant frequency threshold. These points are used as seeds for region growth to obtain a third region. The overlapping area between the first region, the second region, and the third region is determined, and the overlapping area is used as the side of the first reactor that divides the detection area.

[0046] In the aforementioned technical solution, the side surface of the detection area is determined based on the electromagnetic field distribution, stress distribution, and vibration characteristic data of the first reactor. This process embodies the idea of ​​data fusion and comprehensive analysis, and has significant technical advantages. 1. By mapping the electromagnetic field distribution onto the geometric model of the reactor to generate an electromagnetic field intensity distribution map, the distribution of the electromagnetic field inside the reactor can be visually observed. Based on this map, statistical analysis of the electromagnetic field intensity distribution values ​​and electromagnetic field variation gradient values ​​can accurately identify regions with high electromagnetic field intensity and large variation gradients. Using these points as seeds for region growing, the resulting first region is the area where the electromagnetic field is most active and likely to cause problems. This data-driven method improves the accuracy and specificity of detection region segmentation.

[0047] 2. Stress distribution is a crucial indicator for evaluating the structural strength of reactors. Mapping the stress distribution onto the reactor's geometric model generates a stress distribution map, clearly showing the stress distribution within the reactor. By assessing the degree of stress concentration, areas with high stress concentration can be identified. Using these points as seeds for region growth, the resulting secondary regions represent areas with weaker structural strength and potential for failure. This assessment method helps to promptly identify and address structural safety hazards.

[0048] 3. Vibration characteristic data reflects the performance of the reactor under dynamic operating conditions. Mapping the vibration characteristic data onto the reactor's geometric model to generate a vibration characteristic distribution map allows for a visual observation of the vibration distribution within the reactor. By statistically analyzing the vibration peak value and resonant frequency, regions with large vibration peak values ​​and high resonant frequencies can be identified. Using these points as seeds for region growth, the resulting third region represents the area with the most severe vibration and the potential to induce resonance. This analytical method helps optimize the reactor's dynamic performance and reduce vibration-induced damage.

[0049] 4. By comprehensively considering data from electromagnetic fields, stress, and vibration, and mapping them onto the reactor's geometric model for integrated analysis, a more comprehensive understanding of the reactor's operating state and performance can be achieved. Determining the overlapping area between the first, second, and third regions as the side of the detection area allows for precise location of the most critical and attention-required areas within the reactor. This multi-field data fusion method improves the scientific rigor and effectiveness of the detection area division, facilitating the timely detection and resolution of potential problems within the reactor.

[0050] Optionally, before driving all signal acquisition sensors to acquire signals according to a set signal acquisition control strategy to generate a first reactance dynamic characteristic characterization signal sequence, the following steps are included: Determine the operating voltage of the first reactor under different fault conditions, and establish a mapping relationship feature between the operating voltage and the fault condition; Clustering is performed on the mapping relationship features to construct fault state clusters; Principal component analysis of the fault state cluster yields the key variables in the mapping relationship features; Determine the trigger threshold for signal acquisition events, and associate the trigger threshold with the key variable through logical operations to construct a signal acquisition control strategy.

[0051] The process of designing the signal acquisition and control strategy as described above in the technical solution has the following technical advantages: 1. By determining the operating voltage of the first reactor under different fault conditions and establishing the mapping relationship between them, basic data is provided for subsequent clustering and analysis. This mapping relationship directly reflects the operating characteristics of the reactor under fault conditions, which helps to accurately locate the fault type and severity.

[0052] 2. Clustering the mapping relationship features allows states with similar fault characteristics to be grouped into the same cluster, thus constructing fault state clusters. This process helps simplify the complexity of fault states, facilitating subsequent analysis and processing. Simultaneously, clustering can reveal the inherent connections and patterns between different fault states, providing strong support for reactor fault diagnosis and maintenance.

[0053] 3. Principal component analysis (PCA) of the fault state clusters reveals key variables in the mapping relationship characteristics. These key variables reflect the main characteristics and information of the reactor fault states, serving as an important basis for constructing signal acquisition and control strategies. Extracting these key variables simplifies the complexity of signal acquisition and improves the targeting and efficiency of data acquisition.

[0054] 4. Trigger thresholds for signal acquisition events are determined based on key variables, and these thresholds are logically correlated with the key variables to construct a signal acquisition control strategy. This process enables refined management of signal acquisition, ensuring timely data collection when reactors malfunction or experience abnormal conditions, providing strong support for subsequent analysis and processing. Furthermore, by logically correlating trigger thresholds and key variables, differentiated responses and processing for different fault states can be achieved, improving the accuracy and efficiency of fault diagnosis.

[0055] Optionally, the step of training the first pre-trained neural network model and the second pre-trained neural network model based on the first reactor dynamic characteristic characterization signal sequence to obtain the single-point reactor dynamic characteristic analysis model and the overall reactor dynamic characteristic analysis model includes: Time-domain features, frequency-domain features, and time-frequency-domain features are extracted from the first reactive dynamic characteristic characterization signal sequence. The extracted time-domain features, frequency-domain features, and time-frequency-domain features are embedded to obtain time-domain feature vectors, frequency-domain feature vectors, and time-frequency-domain feature vectors, respectively. The time-domain feature vector, frequency-domain feature vector, and time-frequency-domain feature vector are fused to obtain the model input feature vector; The model input feature vector is input into the first pre-trained neural network model and the second pre-trained neural network model to perform single-point reactance dynamic characteristic prediction and overall reactor dynamic characteristic prediction, respectively. Based on the predicted dynamic characteristics of the single-point reactor and the dynamic characteristics of the overall reactor, and the corresponding single-point prediction labels and overall prediction labels, the loss values ​​of the first pre-trained neural network model and the second pre-trained neural network model are calculated. The network parameters of the first pre-trained neural network model and the second pre-trained neural network model are adjusted based on the loss values ​​until the training is completed and the dynamic characteristic analysis model of the single-point reactor and the dynamic characteristic analysis model of the overall reactor are obtained.

[0056] Therefore, through comprehensive feature extraction, feature embedding and fusion, dual-model training and prediction, and accurate loss calculation and parameter adjustment, a dynamic characteristic analysis model for a single-point reactor and a dynamic characteristic analysis model for the overall reactor were obtained, which has the following technical advantages: 1. Comprehensive feature extraction By extracting features from the signal sequence representing the dynamic characteristics of the first reactor in the time domain, frequency domain, and time-frequency domain, multi-dimensional information about the reactor's dynamic characteristics can be comprehensively captured. Time-domain features reflect how the signal changes over time, frequency-domain features reveal the frequency components of the signal, while time-frequency-domain features combine the advantages of both, providing richer information. This comprehensive feature extraction lays a solid foundation for subsequent model training and prediction.

[0057] 2. Feature Embedding and Fusion The extracted time-domain, frequency-domain, and time-frequency-domain features are embedded and transformed into vector forms that are easy for the model to process. This not only reduces the dimensionality of the data but also improves the model's learning efficiency. Furthermore, feature fusion combines feature vectors from different domains to form a more comprehensive and accurate model input feature vector, further enhancing the model's predictive capabilities.

[0058] 3. Dual-model training and prediction A first pre-trained neural network model and a second pre-trained neural network model were used to predict the dynamic characteristics of single-point reactors and analyze the dynamic characteristics of the overall reactor, respectively, thus achieving a comprehensive understanding of the reactor's dynamic characteristics. The single-point model can focus on the detailed changes inside the reactor, while the overall model can reflect the overall state of the reactor from a macroscopic perspective. The two complement each other and provide strong support for the fault diagnosis and performance evaluation of the reactor.

[0059] 4. Precise loss calculation and parameter adjustment By calculating the loss value between the predicted result and the true label, the predictive performance of the model can be accurately evaluated. Based on the loss value, the network parameters of the model are adjusted, allowing the model to be continuously optimized during training, ultimately resulting in a single-point reactor dynamic characteristic analysis model and an overall reactor dynamic characteristic analysis model. This precise loss calculation and parameter adjustment mechanism ensures the predictive accuracy and stability of the model.

[0060] 5. Improve prediction accuracy and generalization ability The model trained through the above steps, by fully utilizing the multi-dimensional information of the reactor's dynamic characteristics and undergoing precise loss calculation and parameter adjustment, exhibits higher prediction accuracy and stronger generalization ability. This not only improves the accuracy of reactor fault diagnosis but also provides more reliable data support for reactor performance evaluation and optimization.

[0061] 6. Promote intelligent management of reactors The application of this technical solution helps promote the intelligent management of reactors. By collecting dynamic characteristic data of the reactor in real time and inputting it into a trained model, real-time monitoring and prediction of the reactor's status can be achieved. This not only enables the timely detection of potential fault risks but also provides a scientific basis for reactor maintenance and management, reducing maintenance costs and improving operational efficiency.

[0062] Optionally, the extraction of time-domain features, frequency-domain features, and time-frequency-domain features from the first reactance dynamic characteristic characterization signal sequence includes: The first reactive dynamic characteristic characterization signal sequence is orthogonally decomposed to obtain orthogonal components, and Hilbert transform is performed based on the orthogonal components to extract the time-domain features; The first reactance dynamic characteristic characterization signal sequence is subjected to empirical multimodal decomposition to obtain several intrinsic mode functions. For each intrinsic mode function, a first candidate frequency domain feature is decomposed in the frequency domain, and the correlation estimate between each first candidate frequency domain feature and the fault category is calculated. The first candidate frequency domain feature with a correlation estimate greater than a set correlation estimate threshold is selected. The first candidate frequency domain feature is recursively eliminated to obtain a second candidate frequency domain feature, which is used as the frequency domain feature. Variational mode decomposition is performed on the first reactance dynamic characteristic characterization signal sequence to obtain several intrinsic mode functions. The weighted peak ratio of each intrinsic mode function is calculated, and the time-frequency-energy spectrum is calculated based on the weighted peak ratio. The time-frequency-energy spectrum is envelope demodulated to obtain the fault characteristic frequency and its harmonics, which are used as the time-frequency domain features.

[0063] Therefore, using a combination of advanced signal processing techniques for feature extraction offers the following technical advantages: 1. Temporal Feature Extraction Orthogonal decomposition and Hilbert transform: Orthogonal decomposition breaks down the original signal into independent orthogonal components, simplifying the signal analysis process. Subsequent application of the Hilbert transform allows for the accurate extraction of the signal's time-domain features, such as instantaneous amplitude and phase. This method provides detailed information about how the signal changes over time, offering accurate data support for subsequent model training and prediction.

[0064] 2. Frequency Domain Feature Extraction Empirical Multimodal Decomposition (EMD) and Correlation Estimation: Empirical multimodal decomposition (EMD) can decompose the original signal into several intrinsic mode functions (IMFs), each IMF representing a fundamental mode of the signal. By performing frequency domain decomposition on each IMF, candidate frequency domain features can be obtained. Furthermore, by calculating the correlation estimates between these features and fault categories, and selecting features with high correlation for recursive feature elimination, optimized frequency domain features are finally obtained. This method can effectively screen out the frequency domain features most relevant to the fault, improving the model's predictive accuracy.

[0065] Recursive feature elimination: The application of recursive feature elimination methods helps to further simplify the feature set, remove redundant information, and improve the model's generalization ability. By iteratively training the model and removing the least important features, the final feature set is more compact and effective.

[0066] 3. Benefits of Time-Frequency Domain Feature Extraction Variational Mode Decomposition and Weighted Peak Ratio Calculation: Variational mode decomposition is an efficient signal decomposition method that can decompose a signal into several intrinsic mode functions (EMFs). Calculating the weighted peak ratio of each EMF highlights key frequency components in the signal. Furthermore, by calculating the time-frequency-energy spectrum and performing envelope demodulation, fault characteristic frequencies and their harmonics can be accurately extracted. This method can capture the complex changes in the signal in the time and frequency domains, providing rich feature information for model training and prediction.

[0067] 4. By comprehensively extracting features from the time domain, frequency domain, and time-frequency domain, multi-dimensional information about the dynamic characteristics of the reactor can be fully captured, improving the prediction accuracy of the model. This multi-dimensional feature extraction method helps to reveal hidden patterns in the signal and improves the model's ability to diagnose complex faults.

[0068] The application of methods such as recursive feature elimination helps remove redundant features and improve the model's generalization ability. This allows the model to maintain high prediction accuracy even when faced with new data or unknown faults.

[0069] Promoting Intelligent Management of Reactors: The application of this technical solution helps to promote the intelligent management of reactors. By collecting and analyzing the dynamic characteristic data of reactors in real time, potential fault risks can be detected in a timely manner, providing a scientific basis for reactor maintenance and management. This helps to reduce maintenance costs, improve operating efficiency, and ensure the safe and stable operation of the power system.

[0070] Optionally, the first pre-trained neural network model is a convolutional neural network model, and the second pre-trained neural network model is a recurrent neural network model. The convolutional neural network model includes multiple convolutional layers, pooling layers, and a first classifier, and the recurrent neural network model includes a serialization module, a gated recurrent unit, and a second classifier. The model input feature vector is input into a first pre-trained neural network model and a second pre-trained neural network model to predict the dynamic characteristics of single-point reactance and the dynamic characteristics of the overall reactor, respectively, including: The model input feature vector is convolved based on the multi-layer convolution to extract local spatial features in the first reactive dynamic characteristic characterization signal sequence. Based on the pooling layer, the local spatial features are pooled to obtain pooled features; Based on the first classifier, the probability prediction relationship of the predefined single-point reactance dynamic characteristic label of the pooling feature threshold is determined to obtain the predicted single-point reactance dynamic characteristic. Based on the serialization module, feature extraction is performed on the input feature vector of the model to obtain a time series feature vector; Based on the gated loop unit, feature extraction is performed on the time series feature vector to capture the time series feature map in the first reactance dynamic characteristic characterization signal sequence; Based on the second classifier, the probability prediction relationship between the time-series feature map and the predefined overall reactance dynamic characteristic label is determined to obtain the predicted overall reactance dynamic characteristic.

[0071] Therefore, a convolutional neural network (CNN) was selected as the first pre-trained neural network model, and a gated recurrent unit (such as LSTM or GRU) in a recurrent neural network (RNN) was selected as the second pre-trained neural network model to perform single-point reactance dynamic characteristic prediction and overall reactor dynamic characteristic analysis, respectively. This has the following technical advantages: 1. Convolutional Neural Network (CNN) Local spatial feature extraction: CNNs, through multi-layer convolutional operations, can effectively extract local spatial features from the input feature vector. This is crucial for capturing subtle changes and patterns in the dynamic characteristics of reactors, helping to improve the accuracy of single-point reactor dynamic characteristic prediction.

[0072] Pooling layers reduce the dimensionality of feature maps, reducing computational cost while preserving key information. This helps speed up model training and prevents overfitting.

[0073] Efficient prediction by the first classifier: By classifying pooled features using a first classifier (such as a fully connected layer with softmax), a probabilistic prediction relationship with predefined single-point reactance dynamic characteristic labels can be obtained. This design allows the model to directly output prediction results, facilitating subsequent applications and decision-making.

[0074] 2. Recurrent Neural Networks (RNNs) and Gated Recurrent Units Time series feature extraction: RNNs, through their serialization module, can process time series information in the input feature vector, capturing temporal dependencies in the data. This is crucial for analyzing the dynamic characteristics of the overall reactor, as the reactor's state may change continuously over time.

[0075] Gated recurrent units capture long-term dependencies: Traditional RNNs may encounter gradient vanishing or exploding problems when processing long sequence data. However, gated recurrent units (such as LSTM or GRU) can effectively capture long-term dependencies by introducing a gating mechanism, thereby improving the model's ability to predict the overall dynamic characteristics of the reactor.

[0076] Accurate Classification by the Second Classifier: By classifying the time-series feature map using a second classifier, a probabilistic prediction relationship with a predefined overall reactance dynamic characteristic label can be obtained. This design enables the model to accurately distinguish different overall states, providing strong support for reactor maintenance and management.

[0077] 3. Overall benefits CNNs and RNNs are complementary in feature extraction and prediction. CNNs excel at extracting local spatial features, while RNNs are adept at processing time-series data. By combining these two models, multi-dimensional information about the dynamic characteristics of reactors can be comprehensively captured, improving the accuracy and robustness of predictions.

[0078] This technical solution enhances the model's generalization ability through the design of multi-layer convolution, pooling, and gated recurrent units. This allows the model to maintain high prediction accuracy even when faced with new data or unknown faults.

[0079] Optionally, based on the following loss function, according to the predicted single-point reactor dynamic characteristics and overall reactor dynamic characteristics and the corresponding single-point prediction labels and overall prediction labels, the loss values ​​of the first pre-trained neural network model and the second pre-trained neural network model are calculated, so as to adjust the network parameters of the first pre-trained neural network model and the second pre-trained neural network model based on the loss values, until the training is completed and the single-point reactor dynamic characteristic analysis model and the overall reactor dynamic characteristic analysis model are obtained:

[0080] in, These are weighting coefficients that balance the contributions of different loss terms to the total loss.

[0081] Cross-entropy loss represents the difference between the probability distribution of the model's output and the probability distribution of the true labels. The model predicts the c-th type of probability of the dynamic characteristics of single-point reactance. : The c-th true label for single-point reactance dynamic characteristics : The model's c-th type prediction probability of the overall reactor dynamic characteristics. : The c-th true label for the overall reactor dynamic characteristics;

[0082] Mean squared error loss, used in regression tasks, calculates the average of the squared differences between the model's predicted values ​​and the actual values. The model's predicted value for the single-point reactance dynamic characteristics of the i-th sample. The actual value of the single-point reactance dynamic characteristic of the i-th sample. The model predicts the overall reactor dynamic characteristics for the i-th sample. : The actual value of the overall reactor dynamic characteristics of the i-th sample;

[0083] : Dynamic characteristic difference loss, specifically targeting changes in the dynamic characteristics of reactors. The model's predicted value for the dynamic characteristics of the i-th sample. : The actual dynamic characteristic value of the i-th sample;

[0084] Boundary condition sensitive loss ensures that the model is sensitive to boundary conditions. : The actual value of the j-th boundary condition : The predicted value of the j-th boundary condition;

[0085] Regularization loss is used to prevent the model from overfitting. The weights of the model;

[0086] Time smoothing loss is used to ensure the smoothness of time series data. : The model's predicted value for the time series.

[0087] Therefore, using a composite loss function composed of the above-mentioned multiple loss functions has the following technical advantages in the technical solution: 1. The mean squared error loss (MSE) and dynamic characteristic difference loss directly optimize the difference between the predicted and actual values. This helps the model learn a more accurate mapping relationship, thereby improving the prediction accuracy of the dynamic characteristics of single points and the whole reactor.

[0088] 2. Regularization loss reduces model complexity by penalizing excessively large weight values, prevents overfitting, and enhances the model's generalization ability on unseen data.

[0089] 3. Boundary condition sensitive loss ensures that the model is sensitive to changes in the input boundary conditions, which helps the model maintain stable performance under different operating conditions and improves the robustness of the model.

[0090] 4. Time smoothing loss encourages the model output to change more smoothly over time, which is especially important for time series prediction of reactor dynamic characteristics, since physical processes are usually continuous and smoothly changing.

[0091] 5. Cross-entropy loss can handle classification tasks simultaneously. If the prediction of dynamic characteristics of single-point and overall reactors includes classification problems (such as fault type classification), they can be optimized in a unified loss function.

[0092] 6. Combined loss functions can simultaneously optimize the model's performance on different aspects of the reactor's dynamic characteristics, such as amplitude, phase, and frequency response, by combining different types of losses.

[0093] 7. Different weighting coefficients ( This allows for adjusting the importance of different loss terms according to the needs of specific tasks, providing flexibility and adjustability. The reactor geometry data, material property data, and operating condition description data mentioned in the technical solution can be combined with the boundary condition sensitive loss in the loss function, enabling the model to not only learn from the data but also consider the actual constraints of the physical process.

[0094] Optionally, the first pre-trained neural network model is a convolutional neural network model, and the second pre-trained neural network model is a recurrent neural network model. The convolutional neural network model includes multiple convolutional layers, pooling layers, and a first classifier, and the recurrent neural network model includes a serialization module, a gated recurrent unit, and a second classifier. The step of inputting the second reactor dynamic characteristic characterization signal sequence into the single-point reactor dynamic characteristic analysis model and the overall reactor dynamic characteristic analysis model to predict the reactor dynamic characteristics of the second reactor includes: The model input feature vector is convolved based on the multi-layer convolution to extract local spatial features in the second reactive dynamic characteristic characterization signal sequence. Based on the pooling layer, the local spatial features are pooled to obtain pooled features; Based on the first classifier, the probability prediction relationship of the predefined single-point reactance dynamic characteristic label of the pooling feature threshold is determined to obtain the predicted single-point reactance dynamic characteristic. Based on the serialization module, feature extraction is performed on the input feature vector of the model to obtain a time series feature vector; Based on the gated loop unit, feature extraction is performed on the time series feature vector to capture the time series feature map in the second reactance dynamic characteristic characterization signal sequence; Based on the second classifier, the probability prediction relationship between the time-series feature map and the predefined overall reactance dynamic characteristic label is determined to obtain the predicted overall reactance dynamic characteristic.

[0095] Therefore, a convolutional neural network (CNN) is used as the dynamic characteristic analysis model for a single-point reactor, and a gated recurrent unit (such as LSTM or GRU) in a recurrent neural network (RNN) is used as the overall dynamic characteristic analysis model for the reactor to predict the dynamic characteristics of the second reactor. This has the following technical advantages: 1. Advantages of Convolutional Neural Networks (CNNs) in the dynamic characteristics analysis of single-point reactance Local feature extraction capability: Through multi-layer convolutional operations, CNNs can efficiently extract local spatial features from the signal sequence representing the dynamic characteristics of the second reactance. These features are crucial for understanding the state of the reactor at a specific time point and help improve the accuracy of single-point prediction.

[0096] Dimensionality reduction and feature selection: The use of pooling layers not only reduces the dimensionality of feature maps and the amount of computation, but also enhances the model's generalization ability by selecting key features. This helps prevent overfitting and improves the model's robustness.

[0097] Highly efficient classification performance: The first classifier (such as a fully connected layer with softmax) can quickly and accurately determine the probabilistic prediction relationship with predefined single-point reactance dynamic characteristic labels based on pooling features. This design allows the model to directly output prediction results, facilitating subsequent decision-making and applications.

[0098] 2. Advantages of Recurrent Neural Networks (RNNs) and Gated Recurrent Units in the Analysis of Overall Reactivity Dynamic Characteristics Temporal Feature Capture: RNNs, through their serialization module, can process time-series information in the input feature vector, capturing temporal dependencies in the data. This is crucial for analyzing the dynamic characteristics of reactors over time.

[0099] Long-term dependency handling: Gated recurrent units (such as LSTM or GRU) effectively solve the gradient vanishing or exploding problems that traditional RNNs may encounter when processing long-sequence data by introducing a gating mechanism. This enables the model to capture dependencies over a longer time range and improves the overall prediction accuracy.

[0100] Complex Pattern Recognition: The second classifier can identify complex overall reactance dynamic characteristic patterns based on time-series feature maps and establish probabilistic prediction relationships with predefined overall reactance dynamic characteristic labels. This capability enables the model to handle more complex reactor dynamic behaviors, providing more accurate guidance for reactor maintenance and management.

[0101] 3. Comprehensive technological benefits Enhanced Model Complementarity: CNNs and RNNs are complementary in feature extraction and prediction. CNNs excel at extracting local spatial features, while RNNs are adept at processing time-series data. By combining these two models, multi-dimensional information about the dynamic characteristics of reactors can be comprehensively captured, improving the accuracy and robustness of predictions.

[0102] Improved Prediction Accuracy: Through a carefully designed network structure and feature extraction method, this technical solution can significantly improve the accuracy of predicting the dynamic characteristics of single-point reactors and analyzing the dynamic characteristics of the overall reactor. This helps to detect potential fault risks in a timely manner, extend the service life of the reactor, and reduce maintenance costs.

[0103] Enhanced Model Generalization Ability: This technical solution enhances the model's generalization ability through the design of multi-layer convolution, pooling, and gated recurrent units. This enables the model to maintain high prediction accuracy when facing new data or unknown faults, providing strong support for the intelligent management of reactors.

[0104] In the above scheme, the predicted overall reactance dynamic characteristics can cover the overall behavior and state of the reactor under various operating conditions. These characteristics are generally closely related to the reactor's performance, health status, and possible failure modes. The following are some possible overall reactance dynamic characteristics: Reactance variation: The change in reactance value of a reactor under different operating conditions. This includes the range of reactance fluctuation during normal operation, as well as the trend of reactance value change under specific conditions (such as temperature changes, voltage fluctuations, etc.).

[0105] Loss characteristics: The energy loss of a reactor during operation, including iron losses and copper losses. These loss characteristics are crucial for evaluating the reactor's energy efficiency and heat generation.

[0106] Temperature characteristics: The temperature variation of the reactor under different load and environmental conditions. Excessive temperature may affect the insulation performance and lifespan of the reactor; therefore, temperature characteristics are an important factor to consider when predicting the overall dynamic characteristics of the reactor.

[0107] Stability: The stability of a reactor during long-term operation or under external disturbances (such as voltage fluctuations, short-circuit faults, etc.). This includes the stability of the reactance value, the stability of the temperature, and the overall operational stability.

[0108] Fault mode identification: By predicting the overall dynamic characteristics of the reactor, potential fault modes can be identified, such as localized overheating, insulation aging, and abnormal mechanical vibration. This helps to take preventative measures to prevent the occurrence or escalation of faults.

[0109] It is important to note that the predicted overall dynamic characteristics of the reactance may vary depending on factors such as the type and specifications of the reactor, the operating environment, and the accuracy of the prediction model. Therefore, in practical applications, it is necessary to train and validate the prediction model according to specific circumstances to ensure the accuracy and reliability of the prediction results.

[0110] Furthermore, the predicted overall dynamic characteristics of the reactor can be comprehensively evaluated by combining other monitoring data and analysis results. For example, multi-source data such as the reactor's electrical parameters, mechanical parameters, and environmental factors can be fused and analyzed using machine learning or deep learning algorithms to gain a more comprehensive understanding of the reactor's overall dynamic characteristics.

[0111] In summary, the predicted overall reactance dynamic characteristics are a comprehensive concept that encompasses the overall behavior and state of the reactor under various operating conditions. These characteristics are crucial for evaluating reactor performance, health status, and predicting potential failure modes.

[0112] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for measuring the dynamic characteristics of a reactor, characterized in that, include: Divide each side of multiple first reactors of the same model into several detection areas, set a signal acquisition point at the center of each detection area, and set a signal acquisition sensor at each signal acquisition point. Drive all signal acquisition sensors to acquire signals according to the set signal acquisition control strategy to generate the first reactance dynamic characteristic characterization signal sequence; Based on the first reactor dynamic characteristic characterization signal sequence, the first pre-trained neural network model and the second pre-trained neural network model are trained to obtain the single-point reactor dynamic characteristic analysis model and the overall reactor dynamic characteristic analysis model, respectively. Obtain the dynamic characteristic characterization signal sequence of the second reactor obtained by signal acquisition of the second reactor; The second reactor dynamic characteristic characterization signal sequence is input into the single-point reactor dynamic characteristic analysis model and the overall reactor dynamic characteristic analysis model to predict the reactor dynamic characteristics of the second reactor.

2. The method for measuring the dynamic characteristics of a reactor according to claim 1, characterized in that, The process of dividing each side of multiple first reactors of the same model into several detection areas, and setting a signal acquisition point at the center of each detection area, before setting a signal acquisition sensor at each signal acquisition point includes: The geometric dimensions of the first reactor are acquired and digitized to obtain the reactor geometric model. Obtain the material property data of the first reactor and map it onto the geometric model of the reactor. The material property data includes electrical conductivity, magnetic permeability, density, elastic modulus, and Poisson's ratio. The operating condition description data of the first reactor is obtained to determine the boundary condition characteristics of its electromagnetic field and structural field. The boundary condition characteristics of the electromagnetic field include electric field boundary condition characteristics and magnetic field boundary condition characteristics. The boundary condition characteristics of the structural field include displacement boundary condition characteristics, force boundary condition characteristics, and thermal boundary condition characteristics. Based on the boundary condition characteristics of the electromagnetic field and structural field, the geometric model of the reactor is meshed to obtain several computational domain meshes, which are then labeled to generate a computational domain mesh set. Based on the mesh set of the computational domain, a finite element solution model corresponding to the first reactor is constructed, and the finite element solution model is solved to obtain the electromagnetic field distribution, stress distribution and vibration characteristic data of the first reactor; Based on the electromagnetic field distribution, stress distribution, and vibration characteristic data of the first reactor, the side surface on the first reactor that divides the detection area is determined.

3. The method for measuring the dynamic characteristics of a reactor according to claim 2, characterized in that, Based on the boundary condition characteristics of the electromagnetic field and structural field, the geometric model of the reactor is meshed to obtain several computational domain meshes, which are then labeled to generate a computational domain mesh set, including: Based on the boundary condition characteristics of the electromagnetic field and structural field, the boundary clarity of the reactor geometric model is calculated; Based on boundary clarity, the mapping region for which the reactor geometric model can be divided into mapping meshes is determined; Based on the analysis requirements of the electromagnetic field and structural field, a mapping mesh is generated for the mapping region as a computational domain mesh, and the mapping mesh is labeled. The label includes the mesh type, size, and density. The mesh type represents the geometric regularity of the first reactor, the mesh size represents the geometric characteristics and physical behavior of the first reactor, and the mesh density represents the electromagnetic field strength and stress concentration intensity of the first reactor.

4. The method for measuring the dynamic characteristics of a reactor according to claim 2, characterized in that, The step of constructing the finite element solution model corresponding to the first reactor based on the mesh set of the computational domain includes: Based on the computational domain mesh set, vector potential cells and scalar potential cells are generated. Based on the vector potential unit and scalar potential unit, construct electromagnetic field unit and structural field unit; Electromagnetic field boundary conditions and structural field boundary conditions are added to the electromagnetic field unit and structural field unit, so that local element equations are formed for each electromagnetic field unit and structural field unit respectively. The local element equations corresponding to all electromagnetic field elements and structural field elements are assembled into a global matrix equation, which serves as the constructed finite element solution model.

5. The method for measuring the dynamic characteristics of a reactor according to claim 4, characterized in that, Solving the finite element model yields the electromagnetic field distribution, stress distribution, and vibration characteristic data of the first reactor, including: Based on the constructed electromagnetic field and structural field coupled solver, the finite element solution model is solved according to the set solution type and iteration step size to obtain the electric potential intensity distribution, magnetic field intensity distribution, stress tensor distribution, strain distribution, displacement, velocity, and acceleration distribution. The electromagnetic field distribution is obtained by superimposing the electric potential intensity distribution and the magnetic field intensity distribution and fusing them together. Based on the stress-strain relationship, the stress tensor distribution and strain distribution are fused to obtain the stress distribution; The displacement, velocity, and acceleration distributions are subjected to Fourier transform to fuse them in the frequency domain to obtain vibration characteristic data.

6. The method for measuring the dynamic characteristics of a reactor according to claim 5, characterized in that, The step of determining the side surface on the first reactor that divides the detection area based on the electromagnetic field distribution, stress distribution, and vibration characteristic data of the first reactor includes: The electromagnetic field distribution is mapped onto the geometric model of the reactor to generate an electromagnetic field intensity distribution map. Based on the electromagnetic field intensity distribution map, the electromagnetic field intensity distribution value and the electromagnetic field change gradient value are statistically analyzed to determine the points corresponding to the electromagnetic field intensity distribution value being greater than a set electromagnetic field intensity distribution threshold and the electromagnetic field change gradient value being greater than the electromagnetic field change gradient threshold. These points are used as seeds for region growth to obtain the first region. The stress distribution is mapped onto the reactor geometric model to generate a stress distribution map. Based on the stress distribution map, the stress concentration degree is evaluated to obtain a stress concentration degree evaluation value. The point corresponding to the stress concentration degree evaluation value being greater than the set stress concentration degree evaluation threshold is determined. The point is used as a seed for region growth to obtain a second region. The vibration characteristic data is mapped onto the geometric model of the reactor to generate a vibration characteristic distribution map. Based on the vibration characteristic distribution map, the vibration peak value and resonant frequency are statistically analyzed to determine the points corresponding to the vibration peak value being greater than a set vibration peak value threshold and the resonant frequency being greater than a resonant frequency threshold. These points are used as seeds for region growth to obtain a third region. The overlapping area between the first region, the second region, and the third region is determined, and the overlapping area is used as the side of the first reactor that divides the detection area.

7. The method for measuring the dynamic characteristics of a reactor according to claim 1, characterized in that, Before driving all signal acquisition sensors to acquire signals according to the set signal acquisition control strategy to generate the first reactive dynamic characteristic characterization signal sequence, the process includes: Determine the operating voltage of the first reactor under different fault conditions, and establish a mapping relationship feature between the operating voltage and the fault condition; Clustering is performed on the mapping relationship features to construct fault state clusters; Principal component analysis of the fault state cluster yields the key variables in the mapping relationship features; Determine the trigger threshold for signal acquisition events, and associate the trigger threshold with the key variable through logical operations to construct a signal acquisition control strategy.

8. The method for measuring the dynamic characteristics of a reactor according to claim 1, characterized in that, The step of training a single-point reactor dynamic characteristic analysis model and an overall reactor dynamic characteristic analysis model based on the first reactor dynamic characteristic characterization signal sequence includes: Time-domain features, frequency-domain features, and time-frequency-domain features are extracted from the first reactive dynamic characteristic characterization signal sequence. The extracted time-domain features, frequency-domain features, and time-frequency-domain features are embedded to obtain time-domain feature vectors, frequency-domain feature vectors, and time-frequency-domain feature vectors, respectively. The time-domain feature vector, frequency-domain feature vector, and time-frequency-domain feature vector are fused to obtain the model input feature vector; The model input feature vector is input into the first pre-trained neural network model and the second pre-trained neural network model to perform single-point reactance dynamic characteristic prediction and overall reactor dynamic characteristic prediction, respectively. Based on the predicted dynamic characteristics of the single-point reactor and the dynamic characteristics of the overall reactor, and the corresponding single-point prediction labels and overall prediction labels, the loss values ​​of the first pre-trained neural network model and the second pre-trained neural network model are calculated. The network parameters of the first pre-trained neural network model and the second pre-trained neural network model are adjusted based on the loss values ​​until the training is completed and the dynamic characteristic analysis model of the single-point reactor and the dynamic characteristic analysis model of the overall reactor are obtained.

9. The method for measuring the dynamic characteristics of a reactor according to claim 8, characterized in that, The extraction of time-domain features, frequency-domain features, and time-frequency-domain features from the first reactive dynamic characteristic characterization signal sequence includes: The first reactive dynamic characteristic characterization signal sequence is orthogonally decomposed to obtain orthogonal components, and Hilbert transform is performed based on the orthogonal components to extract the time-domain features; The first reactance dynamic characteristic characterization signal sequence is subjected to empirical multimodal decomposition to obtain several intrinsic mode functions. For each intrinsic mode function, a first candidate frequency domain feature is decomposed in the frequency domain, and the correlation estimate between each first candidate frequency domain feature and the fault category is calculated. The first candidate frequency domain feature with a correlation estimate greater than a set correlation estimate threshold is selected. The first candidate frequency domain feature is recursively eliminated to obtain a second candidate frequency domain feature, which is used as the frequency domain feature. Variational mode decomposition is performed on the first reactance dynamic characteristic characterization signal sequence to obtain several intrinsic mode functions. The weighted peak ratio of each intrinsic mode function is calculated, and the time-frequency-energy spectrum is calculated based on the weighted peak ratio. The time-frequency-energy spectrum is envelope demodulated to obtain the fault characteristic frequency and its harmonics, which are used as the time-frequency domain features.

10. The method for measuring the dynamic characteristics of a reactor according to claim 8, characterized in that, The first pre-trained neural network model is a convolutional neural network model, and the second pre-trained neural network model is a recurrent neural network model. The convolutional neural network model includes multiple convolutional layers, pooling layers, and a first classifier. The recurrent neural network model includes a serialization module, a gated recurrent unit, and a second classifier. The model input feature vector is input into a first pre-trained neural network model and a second pre-trained neural network model to predict the dynamic characteristics of single-point reactance and the dynamic characteristics of the overall reactor, respectively, including: The model input feature vector is convolved based on the multi-layer convolution to extract local spatial features in the first reactive dynamic characteristic characterization signal sequence. Based on the pooling layer, the local spatial features are pooled to obtain pooled features; Based on the first classifier, the probability prediction relationship of the predefined single-point reactance dynamic characteristic label of the pooling feature threshold is determined to obtain the predicted single-point reactance dynamic characteristic. Based on the serialization module, feature extraction is performed on the input feature vector of the model to obtain a time series feature vector; Based on the gated loop unit, feature extraction is performed on the time series feature vector to capture the time series feature map in the first reactance dynamic characteristic characterization signal sequence; Based on the second classifier, the probability prediction relationship between the time-series feature map and the predefined overall reactance dynamic characteristic label is determined to obtain the predicted overall reactance dynamic characteristic.

Images