Excitation waveform identification method, system and device based on multi-algorithm fusion and storage medium

Abstract

The invention discloses an excitation waveform recognition method, system and device based on multi-algorithm fusion and a storage medium, and belongs to the technical field of excitation waveform recognition. The method comprises the following steps: acquiring a to-be-identified excitation wave set, and inputting the to-be-identified excitation wave set into a pre-trained excitation waveform identification model to obtain a waveform of the to-be-identified excitation wave set; the training process of the excitation waveform recognition model comprises the following steps: acquiring excitation waveform data sets of different magnetic elements under different working conditions; extracting high-dimensional characteristic variables representing magnetic flux density distribution and waveform shape characteristics; performing dimension reduction processing on the high-dimensional feature variables by adopting principal component analysis to obtain a dimension reduction matrix; and inputting the dimension reduction matrix into a pre-constructed excitation waveform recognition model for model training to obtain a final excitation waveform recognition result. According to the invention, through multi-algorithm fusion and an adaptive mechanism, the accuracy and robustness of excitation waveform identification are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of excitation waveform recognition, and particularly relates to an excitation waveform recognition method, system, device and storage medium based on multi-algorithm fusion. Background Art

[0002] In power electronics technology, magnetic components such as transformers and inductors play a crucial role in the efficiency and reliability of converters. The excitation waveform is one of the key factors affecting core loss. Different excitation waveforms will result in different rates of change of magnetic flux density over time, thereby affecting the loss characteristics of the core. Therefore, accurately identifying the excitation waveform is crucial for optimizing core design and improving the performance of power electronic systems.

[0003] Currently, the methods for identifying the excitation waveform of magnetic components mainly include methods based on traditional signal processing and methods based on machine learning. Traditional signal processing methods mainly rely on time-domain and frequency-domain analysis, such as Fourier transform (FFT) and wavelet transform, etc., and classify by extracting the characteristic parameters of the waveform. However, these methods have certain limitations in processing complex waveforms, especially when the waveform features are not obvious or there is noise, the classification accuracy is relatively low. The methods based on machine learning identify the excitation waveform by constructing a classification model. Commonly used classification algorithms include K-Nearest Neighbor algorithm (KNN), Random Forest, Support Vector Machine (SVM) and Logistic Regression, etc. These algorithms learn the mapping relationship between the features and labels in the training data to achieve the classification of unknown waveforms. Although these methods perform well in some cases, they each have some problems. For example, KNN is prone to overfitting when dealing with high-dimensional data, Random Forest is sensitive to noise, SVM is very sensitive to parameter selection, and Logistic Regression performs poorly when dealing with complex waveforms. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art, and provide an excitation waveform recognition method, system, device and storage medium based on multi-algorithm fusion, which improves the accuracy and robustness of excitation waveform recognition through multi-algorithm fusion and an adaptive mechanism.

[0005] The present invention provides the following technical solutions:

[0006] In the first aspect, an excitation waveform recognition method based on multi-algorithm fusion is provided, including: obtaining an excitation wave set to be recognized, and inputting it into a pre-trained excitation waveform recognition model to obtain the waveform of the excitation wave set to be recognized;

[0007] Wherein, the training process of the excitation waveform recognition model includes:

[0008] Obtain the excitation waveform data set of different magnetic components under different working conditions, and the excitation waveform data set includes a number of observation samples;

[0009] Based on the excitation waveform data set, extract high-dimensional feature variables characterizing the magnetic flux density distribution and waveform shape features;

[0010] Use principal component analysis to reduce the dimension of the high-dimensional feature variables to obtain a reduced-dimensional matrix;

[0011] Input the reduced-dimensional matrix into a pre-constructed excitation waveform recognition model for model training to obtain the final excitation waveform recognition result;

[0012] Among them, the excitation waveform recognition model includes multiple algorithms for predicting the excitation waveform of the reduced-dimensional matrix, and the prediction results of each algorithm are weighted and fused to obtain the final excitation waveform recognition result;

[0013] The excitation waveform recognition model further includes an adaptive mechanism for dynamically adjusting the weights of each algorithm according to the input reduced-dimensional matrix.

[0014] As an optional technical solution of the present invention, the excitation waveform data set includes the time series of magnetic flux density, magnetic core material, temperature, and classification label.

[0015] As an optional technical solution of the present invention, based on the excitation waveform data set, extracting high-dimensional feature variables characterizing the magnetic flux density distribution and waveform shape features includes:

[0016] The high-dimensional feature variables include rise time, fall time, peak time point, trough time point, peak value, trough value, and distribution uniformity;

[0017] The rise time represents the time interval from the start of the rise of the excitation waveform to reaching the maximum value, and is expressed as:

[0018] ;

[0019] Among them, represents the rise time, N represents the number of data points, represents the change in magnetic flux density at the th moment, and is expressed as , represents the magnetic flux density at the th moment;

[0020] The fall time represents the time interval from the start of the fall of the excitation waveform to reaching the minimum value, and is expressed as:

[0021] ;

[0022] Among them, represents the fall time;

[0023] The peak time point represents the time point when the excitation waveform reaches the maximum value, and is expressed as:

[0024] ;

[0025] Among them, represents the peak time point, represents the independent variable value when the objective function reaches the maximum value;

[0026] The trough time point represents the time point when the excitation waveform reaches the minimum value, and is expressed as:

[0027] ;

[0028] Among them, represents the trough time point, represents the independent variable value when the objective function reaches the minimum value;

[0029] The peak value represents the maximum value of the waveform, and is expressed as:

[0030] ;

[0031] Among them, represents the peak value, and max represents obtaining the maximum value of the objective function;

[0032] The trough value represents the minimum value of the waveform, and is expressed as:

[0033] ;

[0034] Among them, represents the trough value, and min represents obtaining the minimum value of the objective function;

[0035] The distribution uniformity represents the distribution of the magnetic flux density in the excitation waveform, and is expressed as:

[0036] ;

[0037] Among them, represents the distribution uniformity, represents the magnetic flux density histogram, represents the standard deviation of the magnetic flux density histogram, represents the average value of the magnetic flux density histogram.

[0038] As an optional technical solution of the present invention, principal component analysis is used to perform dimensionality reduction processing on the high-dimensional feature variables to obtain a dimensionality reduction matrix, including:

[0039] The high-dimensional feature variables of several observation samples constitute the original feature matrix , the original feature matrix whose rows represent the observation samples and columns represent the high-dimensional feature variables;

[0040] Decentralize the original feature matrix to obtain the decentralized matrix , expressed as:

[0041] ;

[0042] where represents the vector composed of the column averages of the original feature matrix ;

[0043] Based on the decentralized matrix , calculate the covariance matrix , expressed as:

[0044] ;

[0045] where represents the number of observation samples, represents the transpose of the decentralized matrix ;

[0046] Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues and the corresponding eigenvectors ;

[0047] Sort the eigenvalues from largest to smallest, and select the eigenvectors corresponding to the first eigenvalues to form the projection matrix , expressed as:

[0048] ;

[0049] where represents the eigenvector corresponding to the th eigenvalue;

[0050] Reduce the dimension of the centralized matrix through the projection matrix to obtain the reduced-dimension matrix , expressed as:

[0051] .

[0052] As an alternative technical solution of the present invention, the excitation waveform recognition model includes the K-nearest neighbor algorithm, the random forest algorithm, the SUM algorithm, and the logistic regression algorithm.

[0053] As an alternative technical solution of the present invention, the K-nearest neighbor algorithm includes:

[0054] For the observation sample to be classified , calculate its Euclidean distance or Manhattan distance from all the observation samples in the excitation waveform dataset, which are respectively expressed as:

[0055] ;

[0056] ;

[0057] Wherein, represents the Euclidean distance between the i-th observation sample and the observation sample to be classified , and respectively represent the values of the i-th observation sample and the observation sample to be classified on the th feature, D represents the dimension of the feature space, represents the Manhattan distance between the i-th observation sample and the observation sample to be classified ;

[0058] Sort the distances between the observation sample to be classified and all the observation samples in the excitation waveform dataset from small to large, and select the first K observation samples;

[0059] Obtain the classification labels of the first K observation samples, and count the frequencies of each category label. Take the classification label with the largest frequency as the prediction label of the observation sample to be classified .

[0060] As an alternative technical solution of the present invention, based on the dimensionality reduction matrix, confirm the initial weights of each algorithm in the excitation waveform recognition model;

[0061] After inputting the excitation wave set to be recognized into the trained excitation waveform recognition model, the adaptive mechanism calculates the classification accuracy of each algorithm, and dynamically allocates the weights of each algorithm from large to small according to the classification accuracy;

[0062] Based on the dynamically allocated weights of each algorithm, perform weighted fusion on the prediction results of each algorithm to obtain the final excitation waveform recognition result.

[0063] In a second aspect, a multi-algorithm fusion-based excitation waveform recognition system is provided, including: a recognition module, configured to obtain the excitation wave set to be recognized and input it into a pre-trained excitation waveform recognition model to obtain the waveform of the excitation wave set to be recognized;

[0064] Among them, the training process of the excitation waveform recognition model includes:

[0065] Obtain an excitation waveform data set of different magnetic components under different working conditions, where the excitation waveform data set includes a number of observation samples;

[0066] Based on the excitation waveform data set, extract high-dimensional feature variables characterizing the magnetic flux density distribution and waveform shape features;

[0067] Use principal component analysis to perform dimensionality reduction processing on the high-dimensional feature variables to obtain a dimensionality reduction matrix;

[0068] Input the dimensionality reduction matrix into a pre-constructed excitation waveform recognition model for model training to obtain the final excitation waveform recognition result;

[0069] Among them, the excitation waveform recognition model includes multiple algorithms for predicting the excitation waveform of the dimensionality reduction matrix, and the prediction results of each algorithm are weighted and fused to obtain the final excitation waveform recognition result;

[0070] The excitation waveform recognition model further includes an adaptive mechanism for dynamically adjusting the weights of each algorithm according to the input dimensionality reduction matrix.

[0071] In a third aspect, there is provided an excitation waveform recognition device based on multi-algorithm fusion, including a processor and a storage medium;

[0072] The storage medium is used to store instructions;

[0073] The processor is configured to operate according to the instructions to execute the steps of the excitation waveform recognition method based on multi-algorithm fusion described in the first aspect.

[0074] In a fourth aspect, there is provided a computer-readable storage medium, on which a computer program is stored, characterized in that when the program is executed by a processor, it implements the steps of the excitation waveform recognition method based on multi-algorithm fusion described in the first aspect.

[0075] Compared with the prior art, the beneficial effects of the present invention are:

[0076] The excitation waveform recognition method based on multi-algorithm fusion provided by the present invention integrates multiple algorithms for predicting the excitation waveform of the dimensionality reduction matrix, makes full use of the advantages of each algorithm, improves the accuracy and robustness of classification, can better capture the complex features of the waveform, and reduces the possibility of misclassification; the adaptive weight adjustment mechanism dynamically adjusts the weights of each algorithm according to the performance of each algorithm, ensures the optimality of the final classification result, effectively responds to waveform changes under different working conditions, and improves the flexibility and adaptability of classification. Description of the Drawings

[0077] Figure 1It is the training flow chart of the excitation waveform recognition model in the embodiment of the present invention;

[0078] Figure 2 It is the graph of the relationship between the magnetic flux density and the time series in the embodiment of the present invention;

[0079] Figure 3 It is the excitation waveform graph of the sine wave in the embodiment of the present invention;

[0080] Figure 4 It is the excitation waveform graph of the triangular wave in the embodiment of the present invention;

[0081] Figure 5 It is the excitation waveform graph of the trapezoidal wave in the embodiment of the present invention. Detailed implementation manners

[0082] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0083] Embodiment 1

[0084] This embodiment provides an excitation waveform recognition method based on multi-algorithm fusion. It includes:

[0085] Obtain the excitation wave set to be recognized, and input it into the pre-trained excitation waveform recognition model to obtain the waveform of the excitation wave set to be recognized.

[0086] In this embodiment, the excitation wave set to be recognized includes 20 waveforms to be predicted for each of the four materials, that is, 80 observation samples to be classified.

[0087] Among them, the training process of the excitation waveform recognition model includes the following steps:

[0088] Step 1: Obtain the excitation waveform data set of different magnetic components under different working conditions, and the excitation waveform data set includes several observation samples.

[0089] After obtaining the excitation waveform dataset, a preliminary inspection is carried out on the excitation waveform dataset to identify and process missing values, outliers, and duplicate records. For missing values, an interpolation method based on material properties is adopted, that is, the mean or median filling is performed using other sample data of the same material to reduce data bias. For outliers, they are identified through the box plot method. By showing the median, quartiles, and outliers, the distribution of indicators can be intuitively reflected. The upper and lower edges of the box represent the third quartile (Q3) and the first quartile (Q1) respectively, while the line inside the box represents the median (Median). The outliers are shown as marks of different colors to help identify the extreme values in frequency and core loss. At the same time, the extreme values are replaced with adjacent normal values or deleted to ensure the consistency and reliability of the data. In addition, the data is also de-duplicated to ensure the uniqueness of each sample and avoid model overfitting.

[0090] The excitation waveform dataset includes the time series of magnetic flux density, magnetic core material, temperature, and classification labels.

[0091] Step 2: Based on the excitation waveform dataset, extract high-dimensional feature variables that characterize the magnetic flux density distribution and waveform shape features.

[0092] As Figure 2 shown, the color gradient of the image represents the intensity of the magnetic flux density. The red area represents a higher magnetic flux density, while the blue area represents a lower magnetic flux density. Accordingly, feature variables reflecting the magnetic flux density distribution and waveform shape can be extracted: one is the rise time, which represents the time required for the magnetic flux density to rise from the lowest point to the highest point, reflecting the rapidity of the change in magnetic flux density; the other is the trough time point, which represents the time when the magnetic flux density reaches the lowest point, helping to identify the periodicity and stability of the waveform.

[0093] In this embodiment, the high-dimensional feature variables include rise time, fall time, peak time point, trough time point, peak value, trough value, and distribution uniformity;

[0094] The rise time represents the time interval from the start of the rise of the excitation waveform to reaching the maximum value, which is expressed as:

[0095] ;

[0096] where represents the rise time, N represents the number of data points, represents the change in magnetic flux density at the th moment, which is expressed as , represents the magnetic flux density at the th moment;

[0097] The fall time represents the time interval from the start of the fall of the excitation waveform to reaching the minimum value, and is expressed as:

[0098] ;

[0099] Wherein, represents the fall time;

[0100] The peak time point represents the time point when the excitation waveform reaches the maximum value, and is expressed as:

[0101] ;

[0102] Wherein, represents the peak time point, represents the value of the independent variable when the objective function reaches the maximum value;

[0103] The trough time point represents the time point when the excitation waveform reaches the minimum value, and is expressed as:

[0104] ;

[0105] Wherein, represents the trough time point, represents the value of the independent variable when the objective function reaches the minimum value;

[0106] The peak value represents the maximum value of the waveform, and is expressed as:

[0107] ;

[0108] Wherein, represents the peak value, and max represents obtaining the maximum value of the objective function;

[0109] The trough value represents the minimum value of the waveform, and is expressed as:

[0110] ;

[0111] Wherein, represents the trough value, and min represents obtaining the minimum value of the objective function;

[0112] The distribution uniformity represents the distribution of the magnetic flux density in the excitation waveform, and is expressed as:

[0113] ;

[0114] Wherein, represents the distribution uniformity, represents the magnetic flux density histogram, represents the standard deviation of the magnetic flux density histogram, represents the average value of the magnetic flux density histogram.

[0115] Step 3: Use principal component analysis to reduce the dimensionality of the high-dimensional feature variables and obtain a dimensionality reduction matrix.

[0116] To reduce the collinearity effect among high-dimensional feature variables and ensure the effectiveness and reliability of the recognition model, the principal component analysis (PCA) method is used to reduce the dimensionality of the high-dimensional feature variables. The PCA method can transform the high-dimensional feature variables into a new set of independent features, namely principal components. By selecting the principal components with the largest variances, the dimensionality of the data can be effectively reduced while retaining as much information as possible in the high-dimensional feature variables.

[0117] The high-dimensional feature variables of several observation samples form the original feature matrix , and the original feature matrix has rows representing observation samples and columns representing high-dimensional feature variables;

[0118] Center the original feature matrix to obtain the centered matrix , which is expressed as:

[0119] ;

[0120] where represents the vector composed of the column means of the original feature matrix ;

[0121] Based on the centered matrix , calculate the covariance matrix , which is expressed as:

[0122] ;

[0123] where represents the number of observation samples, represents the transpose of the centered matrix ;

[0124] Perform eigen decomposition on the covariance matrix to obtain the eigenvalues and the corresponding eigenvectors . The eigenvalues represent the importance of each principal component, while the eigenvectors define the directions of the principal components;

[0125] Sort the eigenvalues in descending order and select the eigenvectors corresponding to the top eigenvalues to form the projection matrix , which is expressed as:

[0126] ;

[0127] Among them, represents the eigenvector corresponding to the th eigenvalue;

[0128] By the projection matrix to perform dimensionality reduction on the centralized matrix to obtain the dimensionality reduction matrix , which is expressed as:

[0129] .

[0130] Step 4: Input the dimensionality reduction matrix into a pre-constructed excitation waveform recognition model for model training to obtain the final excitation waveform recognition result.

[0131] Among them, the excitation waveform recognition model includes multiple algorithms for predicting the excitation waveform of the dimensionality reduction matrix. The prediction results of each algorithm are weighted and fused to obtain the final excitation waveform recognition result. The excitation waveform recognition model also includes an adaptive mechanism for dynamically adjusting the weights of each algorithm according to the input dimensionality reduction matrix.

[0132] In this embodiment, the excitation waveform recognition model includes the K-nearest neighbor algorithm, the random forest algorithm, the SUM algorithm, and the logistic regression algorithm.

[0133] Specifically, the K-nearest neighbor algorithm identifies the most similar observation samples for classification by calculating the distances between observation samples, including:

[0134] For the observation sample to be classified , calculate its Euclidean distance or Manhattan distance from all the observation samples in the excitation wave set to be recognized, which are respectively expressed as:

[0135] ;

[0136] ;

[0137] Among them, represents the Euclidean distance between the th observation sample and the observation sample to be classified, and respectively represent the values of the th observation sample and the observation sample to be classified on the th feature, D represents the dimension of the feature space, and the observation sample to be classified is the Manhattan distance;

[0138] The observation samples to be classified Sort the distances between the observation samples to be classified and all the observation samples in the excitation wave set to be recognized from small to large, and select the top K observation samples;

[0139] Obtain the classification labels of the top K observation samples, count the frequencies of each category label, and use the category label with the largest frequency as the prediction label of the observation samples to be classified of.

[0140] The random forest algorithm classifies by constructing multiple decision trees and taking the majority vote. The SUM (support vector machine) algorithm is a classification method based on the maximum margin principle, which is suitable for non-linear classification problems. The logistic regression algorithm is a linear classification method, which is suitable for binary classification problems.

[0141] Based on the dimensionality reduction matrix, confirm the initial weights of each algorithm in the excitation waveform recognition model;

[0142] After inputting the dimensionality reduction matrix into the trained excitation waveform recognition model, the adaptive mechanism calculates the classification accuracy of each algorithm, and dynamically assigns the weights of each algorithm from large to small according to the classification accuracy. That is, for different materials and waveform types, the adaptive mechanism will automatically select the most suitable model combination to improve the accuracy and stability of classification.

[0143] Based on the weights of each algorithm dynamically assigned, perform weighted fusion on the prediction results of each algorithm to obtain the final excitation waveform recognition result.

[0144] Specifically, the dynamic assignment of the weights of each algorithm by the adaptive mechanism is expressed as:

[0145] ;

[0146] Among them, represents the weight of the m-th algorithm in the excitation waveform recognition model at time t, represents the classification accuracy of the m-th algorithm in the excitation waveform recognition model at time t.

[0147] In this embodiment, the excitation waveforms include sine waves, triangular waves, and trapezoidal waves. As Figure 3 - Figure 5 shown, a sine wave usually has smooth rising and falling curves, and the change between the peak and trough is relatively gentle, which is suitable for describing periodic changes; a triangular wave has the characteristics of linear rising and falling, and the waveform is relatively sharp, which is suitable for describing rapidly changing situations; a trapezoidal wave has gentle rising and falling segments, which is suitable for describing changes in a stable state.

[0148] Use the trained excitation waveform recognition model to predict the excitation wave set to be recognized, and the results are shown in Table 1.

[0149] Table 1 Number of Three Waveforms in the Excitation Wave Set to be Identified

[0150] Sine wave Triangle wave Trapezoidal wave Quantity / unit 20 44 16

[0151] Example 2

[0152] This example provides an excitation waveform recognition system based on multi - algorithm fusion, including:

[0153] An identification module, configured to obtain the excitation wave set to be identified, input it into a pre - trained excitation waveform recognition model, and obtain the waveform of the excitation wave set to be identified;

[0154] Among them, the training process of the excitation waveform recognition model includes:

[0155] Obtain the excitation waveform data set of different magnetic components under different working conditions, and the excitation waveform data set includes a number of observation samples;

[0156] Based on the excitation waveform data set, extract high - dimensional feature variables characterizing the magnetic flux density distribution and waveform shape features;

[0157] Use principal component analysis to perform dimensionality reduction processing on the high - dimensional feature variables to obtain a dimensionality reduction matrix;

[0158] Input the dimensionality reduction matrix into a pre - constructed excitation waveform recognition model for model training to obtain the final excitation waveform recognition result;

[0159] Among them, the excitation waveform recognition model includes multiple algorithms for predicting the excitation waveform of the dimensionality reduction matrix, and the prediction results of each algorithm are weighted and fused to obtain the final excitation waveform recognition result;

[0160] The excitation waveform recognition model further includes an adaptive mechanism for dynamically adjusting the weights of each algorithm according to the input dimensionality reduction matrix.

[0161] Example 3

[0162] This example provides an excitation waveform recognition device based on multi - algorithm fusion, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of the excitation waveform recognition method based on multi - algorithm fusion described in Example 1.

[0163] Example 4

[0164] This example provides a computer - readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the steps of the excitation waveform recognition method based on multi - algorithm fusion described in Example 1.

[0165] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0166] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices produce a device for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0167] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device that implements the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0168] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0169] The above is only the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. An excitation waveform recognition method based on multi-algorithm fusion, characterized in that, Including: Obtain the excitation wave set to be recognized, input it into the pre-trained excitation waveform recognition model, and obtain the waveform of the excitation wave set to be recognized; Among them, the training process of the excitation waveform recognition model includes: Obtain the excitation waveform data set of different magnetic components under different working conditions, and the excitation waveform data set includes a number of observation samples; Based on the excitation waveform data set, extract high-dimensional feature variables characterizing the magnetic flux density distribution and waveform shape features; Use principal component analysis to perform dimensionality reduction processing on the high-dimensional feature variables to obtain a dimensionality reduction matrix; Input the dimensionality reduction matrix into the pre-constructed excitation waveform recognition model for model training to obtain the final excitation waveform recognition result; Among them, the excitation waveform recognition model includes multiple algorithms for predicting the excitation waveform of the dimensionality reduction matrix, and the prediction results of each algorithm are weighted and fused to obtain the final excitation waveform recognition result; The excitation waveform recognition model also includes an adaptive mechanism for dynamically adjusting the weights of each algorithm according to the input dimensionality reduction matrix.

2. The excitation waveform recognition method based on multi-algorithm fusion according to claim 1, wherein The excitation waveform data set includes the time series of magnetic flux density, magnetic core material, temperature, and classification labels.

3. The excitation waveform recognition method based on multi-algorithm fusion according to claim 1, characterized in that Based on the excitation waveform data set, extracting high-dimensional feature variables characterizing the magnetic flux density distribution and waveform shape features includes: The high-dimensional feature variables include rise time, fall time, peak time point, trough time point, peak value, trough value, and distribution uniformity; The rise time represents the time interval from the start of the rise of the excitation waveform to reaching the maximum value, expressed as: ； Among them, represents the rise time, N represents the number of data points, represents the change amount of magnetic flux density at the moment, expressed as represents the magnetic flux density at the The fall time represents the time interval from the start of the fall of the excitation waveform to reaching the minimum value, expressed as: ； Among them, represents the fall time; The peak time point represents the time point when the excitation waveform reaches the maximum value, expressed as: ； Among them, represents the peak time point, represents the independent variable value when the objective function reaches the maximum value; The trough time point represents the time point when the excitation waveform reaches the minimum value, expressed as: ； Among them, represents the wave trough time point, represents the independent variable value when the objective function reaches the minimum value; The peak value represents the maximum value of the waveform, expressed as: ； Among them, represents the peak value, and max represents obtaining the maximum value of the objective function; The trough value represents the minimum value of the waveform, expressed as: ； wherein, represents the wave valley value, and min represents obtaining the minimum value of the objective function; The distribution uniformity represents the distribution of the magnetic flux density in the excitation waveform, expressed as: ； Among them, represents the distribution uniformity, represents the magnetic flux density histogram, represents the standard deviation of the magnetic flux density histogram, represents the average value of the magnetic flux density histogram.

4. The excitation waveform recognition method based on multi-algorithm fusion according to claim 1, wherein, Using principal component analysis to perform dimensionality reduction processing on the high-dimensional feature variables to obtain a dimensionality reduction matrix includes: The high-dimensional feature variables of several observation samples form an original feature matrix , and the original feature matrix has rows representing observation samples and columns representing high-dimensional feature variables; For the original feature matrix Decentralize it to obtain a decentralized matrix , which is expressed as: ； Among them, represents the vector composed of the column averages of the original feature matrix ; Based on the decentralized matrix , calculate the covariance matrix , which is expressed as: ； Among them, represents the number of observation samples, represents the transpose of the decentralized matrix; For the covariance matrix perform eigen decomposition to obtain the eigenvalues and the corresponding eigenvectors ; Sort the eigenvalues from largest to smallest, and select the eigenvectors corresponding to the first eigenvalues to form the projection matrix , denoted as: ； Among them, represents the eigenvector corresponding to the th eigenvalue; Through the projection matrix reduce the dimension of the centralized matrix to obtain the dimensionality-reduced matrix , expressed as: 。 5. The excitation waveform recognition method based on multi-algorithm fusion according to claim 1, characterized in that The excitation waveform recognition model includes the K-nearest neighbor algorithm, random forest algorithm, SUM algorithm, and logistic regression algorithm.

6. The excitation waveform recognition method based on multi-algorithm fusion according to claim 5, characterized in that The K-nearest neighbor algorithm includes: For the observation samples to be classified , calculate the Euclidean distance or Manhattan distance between it and all the observation samples in the excitation waveform dataset, which are respectively expressed as: ； ； Among them, represents the Euclidean distance between the i-th observed sample and the observed sample to be classified , and respectively represent the values of the i-th observed sample and the observed sample to be classified on the -th feature, D represents the dimension of the feature space, represents the Manhattan distance between the i-th observed sample and the observed sample to be classified ; The observation samples to be classified Sort the distances between the observation samples to be classified and all the observation samples in the excitation waveform dataset from small to large, and select the top K observation samples; Obtain the classification labels of the top K observed samples, count the frequencies of each category label, and use the classification label with the largest frequency as the predicted label of the observed sample to be classified. of the predicted label.

7. The excitation waveform recognition method based on multi-algorithm fusion according to claim 1, characterized in that Based on the dimensionality reduction matrix, confirm the initial weights of each algorithm in the excitation waveform recognition model; After inputting the excitation wave set to be recognized into the trained excitation waveform recognition model, the adaptive mechanism calculates the classification accuracy of each algorithm and dynamically assigns the weights of each algorithm from large to small according to the classification accuracy; Based on the dynamically assigned weights of each algorithm, the prediction results of each algorithm are weighted and fused to obtain the final excitation waveform recognition result.

8. An excitation waveform recognition system based on multi-algorithm fusion, characterized in that, Including: A recognition module for obtaining the excitation wave set to be recognized, inputting it into the pre-trained excitation waveform recognition model, and obtaining the waveform of the excitation wave set to be recognized; Among them, the training process of the excitation waveform recognition model includes: Obtain the excitation waveform data set of different magnetic components under different working conditions, and the excitation waveform data set includes a number of observation samples; Based on the excitation waveform data set, extract high-dimensional feature variables that characterize the magnetic flux density distribution and waveform shape features; Use principal component analysis to perform dimensionality reduction processing on the high-dimensional feature variables to obtain a dimensionality reduction matrix; Input the dimensionality reduction matrix into a pre-constructed excitation waveform recognition model for model training to obtain the final excitation waveform recognition result; Among them, the excitation waveform recognition model includes multiple algorithms for predicting the excitation waveform of the dimensionality reduction matrix, and the prediction results of each algorithm are weighted and fused to obtain the final excitation waveform recognition result; The excitation waveform recognition model also includes an adaptive mechanism for dynamically adjusting the weights of each algorithm according to the input dimensionality reduction matrix.

9. An excitation waveform recognition device based on multi-algorithm fusion, characterized in that, It includes a processor and a storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of the excitation waveform recognition method based on multi-algorithm fusion according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it realizes the steps of the excitation waveform recognition method based on multi-algorithm fusion according to any one of claims 1 to 7.

Images