A method for predicting electromagnetic field distribution of radio frequency quadrupole accelerator using support vector machine

Abstract

The present invention relates to the technical field of radio frequency quadrupole accelerators, and in particular to a method for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator by using a support vector machine, including collecting electric field signals, magnetic field signals and electromagnetic field distribution in the radio frequency quadrupole accelerator by pre-setting electric field probes and Hall sensors; performing denoising, detrending, normalizing and smoothing on the collected electric field signals and magnetic field signals to improve signal quality and the accuracy of subsequent analysis; performing wavelet transform analysis on the collected electric field signals to extract electric field characteristic parameters, and performing wavelet transform analysis on the collected magnetic field signals to extract magnetic field characteristic parameters. The present invention adopts a method combining wavelet transform with support vector machine, which can give full play to the advantages of wavelet transform in multi-scale signal analysis and noise suppression, and at the same time utilizes the powerful classification and regression capabilities of SVM to significantly improve the accuracy, efficiency, robustness and adaptability of the prediction of the electromagnetic field distribution of the radio frequency quadrupole accelerator.

Description

Technical Field

[0001] The invention relates to the technical field of radio frequency quadrupole accelerators, and in particular to a method for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator by using a support vector machine. Background Art

[0002] A radio frequency quadrupole accelerator is a device that accelerates charged particles by alternating high-frequency electric and magnetic fields. Its main features are that it uses a quadrupole field to focus the particle beam and provides energy through a radio frequency field, so that the particles gradually gain higher energy along a curved trajectory inside the accelerator.

[0003] The Chinese patent application with publication number CN119044615A discloses a four-wing radio frequency quadrupole accelerator field distribution measurement device, including a support part, a perturbation body, a driving part and a detection part, wherein the support part is used to connect with the accelerator cavity to fix the accelerator cavity and set the axis of the accelerator cavity vertically, the perturbation body is set in the accelerator cavity, the perturbation body is connected to the driving part, the driving part is used to drive the perturbation body to move along the axial direction of the accelerator cavity in the accelerator cavity, and the detection part is connected to the accelerator cavity and the driving part respectively. By connecting the support part with the accelerator cavity and setting the axis of the accelerator cavity vertically, the position error introduced by the droop of the perturbation body due to the influence of gravity is avoided, the positioning accuracy of the perturbation body in the accelerator cavity is ensured, thereby eliminating the measurement error caused by gravity when measuring the accelerator cavity, and greatly improving the measurement accuracy of the electromagnetic field distribution of the four-wing radio frequency quadrupole accelerator cavity.

[0004] As in the above application, in a radio frequency quadrupole accelerator, the acquisition of electromagnetic field distribution usually relies on traditional experimental measurements, that is, experimental measurements usually use tools such as electric field probes, Hall sensors, and probe arrays to directly measure electric and magnetic field signals at different locations in the accelerator or in the accelerator cavity. Although experimental measurements are a commonly used method for obtaining electromagnetic field distribution, they also have some shortcomings. The results of experimental measurements can only cover the areas where the sensors are placed. For other areas inside the accelerator, the complete electromagnetic field distribution is often unable to be obtained. Installing multiple sensors, collecting data and analyzing them requires a high investment of time and money. Summary of the invention

[0005] In order to solve the above problems, the present invention provides a method for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator by using a support vector machine.

[0006] The present invention adopts the following technical scheme, and adopts a method for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator using a support vector machine, comprising:

[0007] The electric field signal, magnetic field signal and electromagnetic field distribution in the radio frequency quadrupole accelerator are collected by pre-setting electric field probes and Hall sensors;

[0008] De-noising, de-trending, normalizing and smoothing the collected electric and magnetic field signals to improve signal quality and the accuracy of subsequent analysis;

[0009] Perform wavelet transform analysis on the collected electric field signal to extract the electric field characteristic parameters;

[0010] Perform wavelet transform analysis on the collected magnetic field signal to extract magnetic field characteristic parameters;

[0011] By analyzing the correlation of all characteristic parameters in the electric field characteristic parameters and the magnetic field characteristic parameters, redundant characteristic parameters are removed, characteristic parameters closely related to the electromagnetic field distribution prediction are retained and marked as specific characteristic data;

[0012] A set of specific feature data and the corresponding electromagnetic field distribution are converted into a corresponding set of training data, m sets of training data are obtained, a machine learning model for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator is trained based on the training data, real-time specific feature data of the radio frequency quadrupole accelerator is collected, and the electromagnetic field distribution is predicted based on the trained machine learning model.

[0013] As a further description of the above technical solution: the method of performing wavelet transform analysis on the collected electric field signal to extract electric field characteristic parameters includes:

[0014] According to the characteristics of the electric field signal, select the appropriate wavelet basis;

[0015] The electric field signal is decomposed into multiple frequency range components by discrete wavelet transform. The first layer of decomposition obtains low-frequency components. and high frequency components , for low frequency components Further decomposition is performed to obtain the low-frequency components of the second layer of decomposition and high frequency components , and so on, until the preset decomposition level N is reached;

[0016] Extract the characteristic parameters of the low-frequency and high-frequency components of the electric field signal.

[0017] As a further description of the above technical solution: the characteristic parameters of the low-frequency component and the high-frequency component of the electric field signal include the average value, standard deviation and energy of the low-frequency component and the high-frequency component.

[0018] As a further description of the above technical solution: the method of selecting the wavelet basis according to the characteristics of the electric field signal includes:

[0019] According to the smoothness, symmetry and complexity of the electric field signal, the corresponding Haar wavelet, Symlets wavelet and Coiflets wavelet are selected, and the threshold is manually set through the threshold judgment method. The selection conditions include:

[0020] Condition 1: When the smoothness of the electric field signal is higher than the preset smoothness threshold, the Haar wavelet is selected as the wavelet basis;

[0021] Condition 2: When the symmetry of the electric field signal is higher than the preset symmetry threshold, the Symlets wavelet is selected as the wavelet basis;

[0022] Condition three: When the complexity of the electric field signal is higher than the preset complexity threshold, the Coiflets wavelet is selected as the wavelet basis.

[0023] As a further description of the above technical solution: when the electric field signal does not meet the first, second and third conditions, the Daubechies wavelet is selected;

[0024] When the electric field signal satisfies any two or three of the conditions one, two and three, the wavelet transforms corresponding to the two or three conditions are randomly selected as the wavelet basis.

[0025] As a further description of the above technical solution: the method of performing wavelet transform analysis on the collected magnetic field signal to extract magnetic field characteristic parameters includes:

[0026] According to the characteristics of the magnetic field signal, select the appropriate wavelet basis;

[0027] The magnetic field signal is decomposed into multiple frequency range components by discrete wavelet transform. The first layer of decomposition obtains low-frequency components. and high frequency components , for low frequency components Further decomposition is performed to obtain the low-frequency components of the second layer of decomposition and high frequency components , and so on, until the preset decomposition level M is reached;

[0028] The characteristic parameters of the low-frequency component and the high-frequency component of the magnetic field signal are extracted, wherein the characteristic parameters include the average value, standard deviation and energy of the low-frequency component and the high-frequency component.

[0029] As a further description of the above technical solution: the method of removing redundant characteristic parameters by analyzing the correlation of all characteristic parameters in the electric field characteristic parameters and the magnetic field characteristic parameters includes:

[0030] Sort the values ​​of each characteristic parameter in the electric field characteristic parameter and the magnetic field characteristic parameter, and assign a ranking to each observation value, wherein for observation values ​​with the same value, the average ranking needs to be used;

[0031] Any two feature parameters are paired together, recorded as a feature parameter group, and the ranking difference of each pair of feature parameters is calculated. The calculation formula is: , where , Characteristics and Features No. The ranking of observations;

[0032] Based on the ranking differences of each pair of feature parameters, the correlation coefficient of association between each pair of feature parameters was calculated;

[0033] A correlation level threshold Z is preset, and the correlation level correlation coefficient between each pair of characteristic parameters obtained is compared and analyzed with the preset correlation level threshold to generate a specific characteristic data marking instruction.

[0034] As a further description of the above technical solution: the method for obtaining the correlation coefficient of the association level includes:

[0035] ;

[0036] In the formula, is the correlation coefficient of the association rank, n is the number of characteristic parameter groups, is the difference in ranking for each pair of feature parameters.

[0037] As a further description of the above technical solution: the method for generating a specific feature data marking instruction includes:

[0038] when When ≥Z, one of the two characteristic parameters in the characteristic parameter group is randomly removed, and the other characteristic parameter is marked as specific characteristic data;

[0039] when When <Z, both feature parameters in the feature parameter group are marked as specific feature data.

[0040] As a further description of the above technical solution: the training method of the machine learning model for training and predicting the electromagnetic field distribution of the radio frequency quadrupole accelerator includes:

[0041] Converting a set of specific characteristic data and corresponding electromagnetic field distribution into a corresponding set of characteristic vectors;

[0042] Each group of specific feature data is used as the input of the machine learning model, and the machine learning model takes the electromagnetic field distribution corresponding to each group of specific feature data as the output, and takes the electromagnetic field distribution actually corresponding to each group of specific feature data as the prediction target, and takes minimizing the loss function value of the machine learning model as the training target. When the loss function value of the machine learning model is less than or equal to the preset target loss value, the training is stopped. The machine learning model is a support vector machine regression model.

[0043] Beneficial effects:

[0044] The method provided by the present invention adopts support vector machine to predict the electromagnetic field distribution of radio frequency quadrupole accelerator, collects electric field signals in the radio frequency quadrupole accelerator, performs wavelet transform on the electric field signals and the magnetic field signals, and decomposes the electromagnetic field signals, thereby being able to simultaneously capture the high-frequency and low-frequency information of the signals, wherein the high-frequency components reflect the rapid changes of the signals, and the low-frequency components reveal the overall trend of the signals, which enables the signal features at different scales to be effectively extracted, thereby comprehensively grasping the dynamic characteristics of the electromagnetic field, and the wavelet transform has good time-frequency localization characteristics, which can suppress the interference of high-frequency noise, highlight the effective components of the electromagnetic field signals, thereby improving the quality of the signals.

[0045] By further analyzing the correlation of all characteristic parameters in the electric field characteristic parameters and the magnetic field characteristic parameters and removing redundant characteristic parameters, the number of features is reduced, and the computing resources required for data processing and model training are also reduced accordingly, which can speed up the model training and prediction. Therefore, removing redundant characteristic parameters can not only simplify the model structure and improve computing efficiency, but also improve the prediction accuracy and robustness of the model.

[0046] Finally, based on the specific feature data and the corresponding electromagnetic field distribution as training data, a machine learning model for predicting the electromagnetic field distribution of the RF quadrupole accelerator is trained, the real-time specific feature data of the RF quadrupole accelerator is collected, and the electromagnetic field distribution is predicted based on the trained machine learning model. The machine learning model is used to directly predict the electromagnetic field distribution, thereby effectively combining the multi-resolution analysis advantages of wavelet transform and the classification and regression capabilities of the support vector machine model, improving the accuracy and efficiency of electromagnetic field distribution prediction, and overcoming the defects of the existing technology that the experimental measurement cost is high and the complete electromagnetic field distribution cannot be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The present invention will be further explained below in conjunction with the accompanying drawings and embodiments:

[0048] Figure 1 A process for a method for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator using a support vector machine provided in an embodiment of the present invention;

[0049] Figure 2A flow chart of a method for performing wavelet transform analysis on collected electric field signals and extracting electric field characteristic parameters provided by an embodiment of the present invention;

[0050] Figure 3 A flowchart of a method for removing redundant characteristic parameters by analyzing the correlation of all characteristic parameters in electric field characteristic parameters and magnetic field characteristic parameters provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the technical means, creative features, objectives and effects of the present invention easy to understand, the present invention is further described below with reference to specific diagrams. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0052] Example 1

[0053] See also Figure 1 The embodiment of the present invention provides a technical solution: a method for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator using a support vector machine, comprising the following steps:

[0054] The electric field signal, magnetic field signal and electromagnetic field distribution in the radio frequency quadrupole accelerator are collected by pre-setting electric field probes and Hall sensors.

[0055] The collected electric and magnetic field signals are denoised, detrended, normalized and smoothed to improve signal quality and the accuracy of subsequent analysis.

[0056] The collected electric field signals are analyzed by wavelet transform to extract the characteristic parameters of the electric field.

[0057] The collected magnetic field signals are analyzed by wavelet transform to extract the magnetic field characteristic parameters.

[0058] By analyzing the correlation between all characteristic parameters in the electric field characteristic parameters and the magnetic field characteristic parameters, redundant characteristic parameters are removed, and characteristic parameters that are closely related to the electromagnetic field distribution prediction are retained and marked as specific characteristic data.

[0059] A set of specific feature data and the corresponding electromagnetic field distribution are converted into a corresponding set of training data, m sets of training data are obtained, a machine learning model for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator is trained based on the training data, real-time specific feature data of the radio frequency quadrupole accelerator is collected, and the electromagnetic field distribution is predicted based on the trained machine learning model.

[0060] The training method of the machine learning model for training and predicting the electromagnetic field distribution of the radio frequency quadrupole accelerator includes:

[0061] A set of specific characteristic data and the corresponding electromagnetic field distribution are converted into a corresponding set of characteristic vectors.

[0062] Each group of specific feature data is used as the input of the machine learning model, and the machine learning model takes the electromagnetic field distribution corresponding to each group of specific feature data as the output, and takes the electromagnetic field distribution actually corresponding to each group of specific feature data as the prediction target, and takes minimizing the loss function value of the machine learning model as the training target. When the loss function value of the machine learning model is less than or equal to the preset target loss value, the training is stopped.

[0063] The machine learning model is a support vector machine regression model.

[0064] The loss function of the machine learning model is the mean square error, which is one of the commonly used loss functions. Minimization is used to train the model so that the machine learning model can better fit the data, thereby improving the performance and accuracy of the model.

[0065] In the loss function is the loss function value of the machine learning model, is the feature vector group number; is the number of eigenvector groups; For the predicted The electromagnetic field distribution corresponding to the group eigenvector, For the The electromagnetic field distribution actually corresponding to the group of eigenvectors.

[0066] In this embodiment, electric field signals, magnetic field signals and electromagnetic field distribution in the radio frequency quadrupole accelerator are collected, and the collected electric field signals and magnetic field signals are denoised, detrended, normalized and smoothed to improve signal quality and accuracy of subsequent analysis, and then wavelet transform analysis is performed on the collected electric field signals and magnetic field signals to extract electric field characteristic parameters and magnetic field characteristic parameters, and redundant characteristic parameters are removed by analyzing the correlation of all characteristic parameters in the electric field characteristic parameters and the magnetic field characteristic parameters, retaining characteristic parameters that are closely related to the prediction of the electromagnetic field distribution and marking them as specific characteristic data, converting a set of specific characteristic data and the corresponding electromagnetic field distribution into a corresponding set of training data, obtaining m sets of training data, and based on the training data According to the machine learning model trained to predict the electromagnetic field distribution of the radio frequency quadrupole accelerator, the specific feature data of the real-time radio frequency quadrupole accelerator is collected, and the electromagnetic field distribution is predicted based on the trained machine learning model. In summary, the method of combining wavelet transform with support vector machine (SVM) can give full play to the advantages of wavelet transform in multi-scale signal analysis and noise suppression, and at the same time use the powerful classification and regression capabilities of SVM to significantly improve the accuracy, efficiency, robustness and adaptability of the electromagnetic field distribution prediction of the radio frequency quadrupole accelerator. This method can not only improve the accuracy of the prediction model, but also effectively reduce the risk of overfitting and enhance the interpretability of the model. It is an innovative and effective solution in the field of electromagnetic field distribution prediction.

[0067] Example 2

[0068] Reference Figure 1 and Figure 2 Based on the above embodiments, this embodiment further discloses a method for performing wavelet transform analysis on the collected electric field signal to extract electric field characteristic parameters, including:

[0069] According to the characteristics of the electric field signal, a suitable wavelet basis is selected; the method of selecting the wavelet basis according to the characteristics of the electric field signal includes:

[0070] According to the smoothness, symmetry and complexity of the electric field signal, the corresponding Haar wavelet, Symlets wavelet and Coiflets wavelet are selected, and the threshold is manually set through the threshold judgment method. The selection conditions include:

[0071] Condition 1: When the smoothness of the electric field signal is higher than the preset smoothness threshold, the Haar wavelet is selected as the wavelet basis;

[0072] Condition 2: When the symmetry of the electric field signal is higher than the preset symmetry threshold, the Symlets wavelet is selected as the wavelet basis;

[0073] Condition 3: When the complexity of the electric field signal is higher than the preset complexity threshold, the Coiflets wavelet is selected as the wavelet basis;

[0074] When the electric field signal does not meet the first, second and third conditions, the Daubechies wavelet is selected;

[0075] When the electric field signal satisfies any two or three of the conditions one, two and three, the wavelet transforms corresponding to the two or three conditions are randomly selected as the wavelet basis.

[0076] The wavelet transform is Haar wavelet, Daubechies wavelet, Symlets wavelet and Coiflets wavelet.

[0077] The electric field signal is decomposed into multiple frequency range components by discrete wavelet transform. The first layer of decomposition obtains low-frequency components. and high frequency components , for low frequency components Further decomposition is performed to obtain the low-frequency components of the second layer of decomposition and high frequency components , and so on, until the preset decomposition level N is reached; the low-frequency component usually represents the overall trend of the electric field signal, while the high-frequency component represents the instantaneous change or the details.

[0078] The characteristic parameters of the low-frequency component and the high-frequency component of the electric field signal are extracted, wherein the characteristic parameters include the average value, standard deviation and energy of the low-frequency component and the high-frequency component.

[0079] The method for obtaining the average value of the low-frequency component and the high-frequency component of the electric field signal includes:

[0080] The calculation formula for the average value of the low-frequency component of the electric field signal is: .

[0081] The calculation formula for the average value of the high-frequency component of the electric field signal is: .

[0082] in, is the average value of the low-frequency component of the electric field signal, is the average value of the high frequency component of the electric field signal, N is the number of decomposition layers, For the The low-frequency components of the layer decomposition, For the The high frequency components of the layer decomposition.

[0083] The method for obtaining the standard deviation of the low-frequency component and the high-frequency component of the electric field signal comprises:

[0084] The calculation formula of the standard deviation of the low-frequency component of the electric field signal is: .

[0085] The calculation formula of the standard deviation of the high-frequency component of the electric field signal is: .

[0086] In the formula, is the standard deviation of the low-frequency component of the electric field signal, is the standard deviation of the high-frequency component of the electric field signal;

[0087] The method for acquiring the energy of the low-frequency component and the high-frequency component of the electric field signal comprises:

[0088] The calculation formula for the energy of the low-frequency component of the electric field signal is: .

[0089] The calculation formula for the energy of the high-frequency component of the electric field signal is: .

[0090] In the formula, is the energy of the low-frequency component of the electric field signal, is the energy of the high-frequency component of the electric field signal.

[0091] The method of performing wavelet transform analysis on the collected magnetic field signal and extracting the magnetic field characteristic parameters includes:

[0092] According to the characteristics of the magnetic field signal, a suitable wavelet basis is selected; the selection logic is the same as that of the electric field signal, where the wavelet basis includes:

[0093] Haar wavelet, Daubechies wavelet, Symlets wavelet, Coiflets wavelet.

[0094] The magnetic field signal is decomposed into multiple frequency range components by discrete wavelet transform. The first layer of decomposition obtains low-frequency components. and high frequency components , for low frequency components Further decomposition is performed to obtain the low-frequency components of the second layer of decomposition and high frequency components , and so on, until the preset decomposition level M is reached; the low-frequency component usually represents the overall trend of the magnetic field signal, while the high-frequency component represents the instantaneous change or the detail.

[0095] The characteristic parameters of the low-frequency component and the high-frequency component of the magnetic field signal are extracted, wherein the characteristic parameters include the average value, standard deviation and energy of the low-frequency component and the high-frequency component.

[0096] The method for obtaining the average value of the low-frequency component and the high-frequency component of the magnetic field signal includes:

[0097] The calculation formula for the average value of the low-frequency component of the magnetic field signal is: ;

[0098] The calculation formula for the average value of the high-frequency component of the magnetic field signal is: .

[0099] in, is the average value of the low-frequency component of the magnetic field signal, is the average value of the high frequency component of the magnetic field signal, is the number of decomposition layers, For the The low-frequency components of the layer decomposition, For the The high frequency components of the layer decomposition.

[0100] The method for obtaining the standard deviation of the low-frequency component and the high-frequency component of the magnetic field signal comprises:

[0101] The calculation formula of the standard deviation of the low-frequency component of the magnetic field signal is: .

[0102] The calculation formula of the standard deviation of the high-frequency component of the magnetic field signal is: .

[0103] In the formula, is the standard deviation of the low-frequency component of the magnetic field signal, is the standard deviation of the high frequency component of the magnetic field signal.

[0104] The method for acquiring the energy of the low-frequency component and the high-frequency component of the magnetic field signal comprises:

[0105] The calculation formula for the energy of the low-frequency component of the magnetic field signal is: .

[0106] The calculation formula for the energy of the high-frequency component of the magnetic field signal is: .

[0107] In the formula, is the energy of the low-frequency component of the magnetic field signal, is the energy of the high-frequency component of the magnetic field signal.

[0108] In this implementation, by performing wavelet transform on the electric field signal and the magnetic field signal and decomposing the electromagnetic field signal, the high-frequency and low-frequency information of the signal can be captured at the same time. The high-frequency component reflects the rapid changes of the signal, and the low-frequency component reveals the overall trend of the signal. This allows the signal characteristics at different scales to be effectively extracted, thereby fully grasping the dynamic characteristics of the electromagnetic field. In addition, the wavelet transform has good time-frequency localization characteristics, which can suppress the interference of high-frequency noise and highlight the effective components of the electromagnetic field signal, thereby improving the quality of the signal.

[0109] Example 3

[0110] Reference Figure 1 and Figure 3 Based on the above embodiments, this embodiment further discloses that a method for removing redundant characteristic parameters by analyzing the correlation of all characteristic parameters in the electric field characteristic parameters and the magnetic field characteristic parameters includes:

[0111] The numerical values ​​of each characteristic parameter in the electric field characteristic parameter and the magnetic field characteristic parameter are sorted, and a ranking is assigned to each observation value.

[0112] The characteristic parameters are: the average value of the low-frequency component and the high-frequency component of the electric field signal, the standard deviation of the low-frequency component and the high-frequency component of the electric field signal, the energy of the low-frequency component and the high-frequency component of the electric field signal, the average value of the low-frequency component and the high-frequency component of the magnetic field signal, the standard deviation of the low-frequency component and the high-frequency component of the magnetic field signal, and the energy of the low-frequency component and the high-frequency component of the magnetic field signal.

[0113] Example: For the observed values ​​of feature X {5, 3, 8, 7, 6}, their ranking is {2, 1, 5, 4, 3};

[0114] It should be noted that for observations with the same value, the average ranking needs to be used. For example, for the observations of feature Y {5, 5, 8, 7, 6}, the ranking is {2.5, 2.5, 5, 4, 3}.

[0115] Any two feature parameters are paired together, recorded as a feature parameter group, and the ranking difference of each pair of feature parameters is calculated. The calculation formula is: , where , Characteristics and Features No. Rank of observations.

[0116] Based on the ranking differences of each pair of feature parameters, the correlation coefficient of association between each pair of feature parameters was calculated.

[0117] The method for obtaining the correlation coefficient of the association level includes:

[0118] ;

[0119] In the formula, is the correlation coefficient of the association rank, n is the number of characteristic parameter groups, is the difference in ranking for each pair of feature parameters.

[0120] It should be noted that if the correlation coefficient of the association level of two feature parameters is close to 1 or -1, it means that there is a strong monotonic relationship between them, and you can choose to retain one of the feature parameters and remove the redundant feature parameters. If the correlation coefficient of the association level of two feature parameters is close to 0, it means that the correlation between them is weak, and these features can be retained to avoid information loss.

[0121] A correlation level threshold Z is preset, and the correlation level correlation coefficient between each pair of characteristic parameters obtained is compared and analyzed with the preset correlation level threshold to generate a specific characteristic data marking instruction.

[0122] The method of generating a specific characteristic data marking instruction includes:

[0123] when When ≥Z, one of the two feature parameters in the feature parameter group is randomly removed, and the other feature parameter is marked as specific feature data.

[0124] when When <Z, both feature parameters in the feature parameter group are marked as specific feature data.

[0125] In this embodiment, by analyzing the correlation of all characteristic parameters in the electric field characteristic parameters and the magnetic field characteristic parameters, redundant characteristic parameters are removed. After the number of features is reduced, the computing resources required for data processing and model training are also reduced accordingly, which can speed up the model training and prediction. Therefore, removing redundant characteristic parameters can not only simplify the model structure and improve computing efficiency, but also improve the prediction accuracy and robustness of the model.

[0126] The basic principles, main features and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and the description in the specification are only to illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator using a support vector machine, characterized in that: include: The electric field signal, magnetic field signal and electromagnetic field distribution in the radio frequency quadrupole accelerator are collected by pre-setting electric field probes and Hall sensors; De-noising, de-trending, normalizing and smoothing the collected electric and magnetic field signals; Perform wavelet transform analysis on the collected electric field signal to extract the electric field characteristic parameters, and perform wavelet transform analysis on the collected magnetic field signal to extract the magnetic field characteristic parameters; By analyzing the correlation of all characteristic parameters in the electric field characteristic parameters and the magnetic field characteristic parameters, redundant characteristic parameters are removed, characteristic parameters closely related to the electromagnetic field distribution prediction are retained and marked as specific characteristic data; A set of specific feature data and the corresponding electromagnetic field distribution are converted into a corresponding set of training data, m sets of training data are obtained, a machine learning model for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator is trained based on the training data, real-time specific feature data of the radio frequency quadrupole accelerator is collected, and the electromagnetic field distribution is predicted based on the trained machine learning model, wherein the machine learning model is a support vector machine regression model.

2. The method for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator using a support vector machine according to claim 1, characterized in that: The method of performing wavelet transform analysis on the collected electric field signal to extract electric field characteristic parameters comprises: According to the characteristics of the electric field signal, the wavelet basis is selected; The electric field signal is decomposed into multiple frequency range components by discrete wavelet transform. The first layer of decomposition obtains low-frequency components. and high frequency components , for low frequency components Further decomposition is performed to obtain the low-frequency components of the second layer of decomposition and high frequency components , and so on, until the preset decomposition level N is reached; Extract the characteristic parameters of the low-frequency and high-frequency components of the electric field signal.

3. The method for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator using a support vector machine according to claim 2, characterized in that: The characteristic parameters of the low-frequency component and the high-frequency component of the electric field signal include the average value, standard deviation and energy of the low-frequency component and the high-frequency component.

4. The method for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator using a support vector machine according to claim 3, characterized in that: The method of selecting a wavelet basis according to the characteristics of the electric field signal includes: According to the smoothness, symmetry and complexity of the electric field signal, the corresponding Haar wavelet, Symlets wavelet and Coiflets wavelet are selected, and the threshold is manually set through the threshold judgment method. The selection conditions include: Condition 1: When the smoothness of the electric field signal is higher than the preset smoothness threshold, the Haar wavelet is selected as the wavelet basis; Condition 2: When the symmetry of the electric field signal is higher than the preset symmetry threshold, the Symlets wavelet is selected as the wavelet basis; Condition three: When the complexity of the electric field signal is higher than the preset complexity threshold, the Coiflets wavelet is selected as the wavelet basis.

5. The method for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator using a support vector machine according to claim 4, characterized in that: When the electric field signal does not meet the first, second and third conditions, the Daubechies wavelet is selected; When the electric field signal satisfies any two or three of the first condition, the second condition and the third condition, the wavelet transforms corresponding to the two or three conditions are randomly selected as the wavelet basis.

6. The method for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator using a support vector machine according to claim 1, characterized in that: The method of performing wavelet transform analysis on the collected magnetic field signal to extract magnetic field characteristic parameters includes: According to the characteristics of the magnetic field signal, the wavelet basis is selected; The magnetic field signal is decomposed into multiple frequency range components by discrete wavelet transform. The first layer of decomposition obtains low-frequency components. and high frequency components , for low frequency components Further decomposition is performed to obtain the low-frequency components of the second layer of decomposition and high frequency components , and so on, until the preset decomposition level M is reached; The characteristic parameters of the low-frequency component and the high-frequency component of the magnetic field signal are extracted, wherein the characteristic parameters include the average value, standard deviation and energy of the low-frequency component and the high-frequency component.

7. The method for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator using a support vector machine according to claim 1, characterized in that: The method of removing redundant characteristic parameters by analyzing the correlation of all characteristic parameters in the electric field characteristic parameters and the magnetic field characteristic parameters comprises: Sort the values ​​of each characteristic parameter in the electric field characteristic parameter and the magnetic field characteristic parameter, and assign a ranking to each observation value, wherein for observation values ​​with the same value, the average ranking needs to be used; Any two feature parameters are paired together, recorded as a feature parameter group, and the ranking difference of each pair of feature parameters is calculated. The calculation formula is: , where , Characteristics and Features No. The ranking of observations; Based on the ranking differences of each pair of feature parameters, the correlation coefficient of association between each pair of feature parameters was calculated; A correlation level threshold Z is preset, and the correlation level correlation coefficient between each pair of characteristic parameters obtained is compared and analyzed with the preset correlation level threshold to generate a specific characteristic data marking instruction.

8. The method for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator using a support vector machine according to claim 7, characterized in that: The method for obtaining the correlation coefficient of the association level includes: ； In the formula, is the correlation coefficient of the association rank, n is the number of characteristic parameter groups, is the difference in ranking for each pair of feature parameters.

9. The method for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator using a support vector machine according to claim 8, characterized in that: The method for generating a specific feature data marking instruction comprises: when ≥Z, one of the two characteristic parameters in the characteristic parameter group is randomly removed, and the other characteristic parameter is marked as specific characteristic data; when When <Z, both feature parameters in the feature parameter group are marked as specific feature data.

10. The method for predicting the electromagnetic field distribution of a radio frequency quadrupole accelerator using a support vector machine according to claim 7, characterized in that: The training method of the machine learning model for training and predicting the electromagnetic field distribution of the radio frequency quadrupole accelerator includes: Converting a set of specific characteristic data and corresponding electromagnetic field distribution into a corresponding set of characteristic vectors; Each group of specific feature data is used as the input of the machine learning model, and the machine learning model takes the electromagnetic field distribution corresponding to each group of specific feature data as the output, and takes the electromagnetic field distribution actually corresponding to each group of specific feature data as the prediction target. Minimizing the loss function value of the machine learning model is used as the training target, and training is stopped when the loss function value of the machine learning model is less than or equal to the preset target loss value.

Images