A method and device for correcting helium concentration data based on mass spectrometry detection

Abstract

The present application belongs to the technical field of helium data correction, and provides a helium concentration data correction method and device based on mass spectrometric detection, which comprises: inputting an original mass spectrum signal and a weighted feature vector under a first preset value dimension into a double-channel convolutional neural network for processing, generating and outputting a fused feature map; calculating the distortion of the original mass spectrum signal through a nonlinear distortion model to obtain a distorted mass spectrum signal; compensating the feature map based on the distorted mass spectrum signal to obtain a compensated mass spectrum signal; and based on the compensated mass spectrum signal, using an inverse problem solving framework to dynamically invert the current helium concentration value to obtain a corrected helium concentration value. Through the correction method, the interference of different instrument state changes on the signal can be quantified and eliminated; the corrected helium concentration value obtained by dynamically inverting the current helium concentration value based on the compensated mass spectrum signal can reflect the actual helium concentration value.

Description

Technical Field

[0001] This invention belongs to the field of helium data correction technology, and particularly relates to a method and apparatus for correcting helium concentration data based on mass spectrometry detection. Background Technology

[0002] In the field of mass spectrometry, the measurement accuracy of mass spectrometers is crucial for detecting gas components (such as helium concentration). With the increasing demand for precise analysis in scientific research and industrial applications, efficiently and accurately resolving complex mass spectrometry signals and obtaining precise gas concentrations has become a significant technical challenge. Especially in dynamically changing environments and unstable operating conditions, the physical state of the instrument (such as vacuum level, temperature, and magnetic field) can significantly affect the mass spectrometry signal, leading to measurement errors. To overcome this challenge, scientists are constantly exploring new algorithms and methods to extract accurate concentration information from complex mass spectrometry data.

[0003] Currently, common methods in mass spectrometry analysis include signal denoising, peak extraction, DTW (Dynamic Time Warping) algorithms, and model fitting. Especially for the synchronous acquisition and processing of multi-dimensional data (such as mass spectrometry signals, environmental parameters, and magnetic field data), timestamp alignment using dynamic time warping algorithms has become a common technical solution. The DTW algorithm effectively solves the problem of sampling rate mismatch by minimizing the distance between data points (such as mutual information entropy), achieving accurate alignment across sampling rates and ensuring the synchronization of data from different dimensions. Furthermore, signal peak reconstruction methods based on physical models are often used in the final stage of concentration inversion, employing an inverse problem-solving framework to estimate gas concentrations from the compensated signal.

[0004] However, a significant problem remains in the existing technology: under complex environmental conditions, distortions in the mass spectrometry signal (such as changes in instrument status, vibration noise, etc.) can affect the signal quality, thereby affecting the accuracy of helium concentration calculation. Although the DTW algorithm is used to synchronize multidimensional data in time and a physical model is used for peak reconstruction, the signal distortion caused by changes in instrument status is still not fully considered.

[0005] Therefore, the technical problem to be solved is how to provide a correction method that can accurately reflect the actual helium concentration value based on quantifying and eliminating the interference of different instrument state changes on the signal. Summary of the Invention

[0006] Therefore, it is necessary to provide a correction method and device, computer-readable medium and electronic device for helium concentration data based on mass spectrometry detection, which addresses the shortcomings of existing technologies in that they cannot quantify and eliminate the impact of changes in different instrument states on the signal and cannot accurately reflect the actual helium concentration value.

[0007] In a first aspect, embodiments of the present invention provide a method for correcting helium concentration data based on mass spectrometry detection, the method comprising:

[0008] The features extracted from multiple dimensions are sequentially standardized, subjected to principal component analysis, and weighted feature processing to obtain a weighted feature vector under the first preset value dimension; the multiple dimensions include: mass spectrometry signal dimension, environmental parameter dimension, and magnetic field data dimension; the extraction methods used for multi-dimensional extraction include: data sampling extraction method and statistical analysis extraction method;

[0009] The original mass spectrometry signal and the weighted feature vector under the first preset dimension are input into a dual-channel convolutional neural network for processing to generate and output a fused feature map. The dual-channel convolutional neural network includes a first channel and a second channel. The fused feature map integrates the features of the original mass spectrometry signal and the environmental features.

[0010] The distortion of the original mass spectrometry signal is calculated by a nonlinear distortion model to obtain the distorted mass spectrometry signal. The nonlinear distortion model can quantify the influence of different instrument states on the mass spectrometry signal.

[0011] The feature map is compensated based on the distorted mass spectrum signal to obtain the compensated mass spectrum signal.

[0012] Based on the compensated mass spectrometry signal, an inverse problem-solving framework is used to dynamically invert the current helium concentration value to obtain the corrected helium concentration value.

[0013] Optionally, based on the compensated mass spectrometry signal, a reverse problem-solving framework is used to dynamically invert the initial helium concentration value to obtain the corrected helium concentration value, including:

[0014] An optimization problem is constructed based on the aforementioned inverse problem-solving framework, wherein the inverse problem-solving framework is a solution framework constructed based on a physical model and a regularization method;

[0015] Based on the optimization problem, the difference between the compensated mass spectrometry signal and the theoretical mass spectrometry signal is minimized; wherein, a regularization term based on prior knowledge is introduced in the process of minimizing the difference between the compensated mass spectrometry signal and the theoretical mass spectrometry signal.

[0016] The optimization problem that minimizes the difference between the compensated mass spectrometry signal and the theoretical mass spectrometry signal is taken as the optimal problem.

[0017] The optimal problem is solved to obtain the corrected helium concentration value.

[0018] Optionally, after obtaining the corrected helium concentration value, the method further includes:

[0019] The Monte Carlo Dropout method is used to quantify the uncertainty of helium concentration based on multiple sampled helium concentration values ​​and multiple predicted helium concentration values, and the quantification results are obtained.

[0020] Optionally, the Monte Carlo Dropout method is used to quantify the uncertainty of helium concentration based on multiple sampled helium concentration values ​​and multiple predicted helium concentration values, obtaining quantification results, including:

[0021] For a single sample, the following operations are performed:

[0022] The helium concentration value was obtained by sampling.

[0023] During the forward propagation of a dual-channel convolutional neural network, a predetermined proportion of neurons are randomly discarded to obtain a set of neural network parameters;

[0024] The helium concentration is predicted by using this set of neural network parameters, and the predicted helium concentration is obtained.

[0025] Repeat the single sampling operation multiple times to obtain multiple sampled helium concentration values ​​and corresponding multiple predicted helium concentration values;

[0026] Based on multiple sampled helium concentration values ​​and multiple predicted helium concentration values, the mean and variance of the helium concentration are calculated to obtain the mean and variance of the helium concentration.

[0027] The uncertainty of helium concentration is quantified based on the mean and variance of helium concentration, resulting in the quantification result.

[0028] Optionally, before calculating the distortion of the original mass spectrometry signal using a nonlinear distortion model, the method further includes:

[0029] Obtain multiple distortion coefficients of the nonlinear distortion model, including: a first distortion coefficient related to time decay, a second distortion coefficient related to the intensity of periodic distortion caused by vibration, and a third distortion coefficient related to the change in vacuum degree;

[0030] The first distortion coefficient, the second distortion coefficient, and the third distortion coefficient are updated in real time using the stochastic gradient descent method to capture changes in the instrument's state in real time.

[0031] Optionally, the features extracted from multiple dimensions are sequentially subjected to standardization, principal component analysis, and feature weighting to obtain a weighted feature vector under a first preset dimension, including:

[0032] Obtain multiple dimensions;

[0033] For any one of the multiple dimensions, extract multiple original features under the second preset value dimension in sequence;

[0034] All original features are standardized to obtain standardized features.

[0035] For the standardized features, principal component analysis is used to select a first preset number of principal components.

[0036] The principal components with the first preset number of values ​​are subjected to feature weighting to obtain the weighted feature vector under the first preset value dimension.

[0037] Optionally, the step of inputting the original mass spectrum signal and the weighted feature vector under the first preset dimension into a dual-channel convolutional neural network for processing, generating and outputting the fused feature map, includes:

[0038] For the first channel, feature extraction and information extraction are performed on the original mass spectrometry signal;

[0039] For the second channel, a weight adjustment matrix is ​​generated and output through a fully connected layer and a Softmax activation function;

[0040] The features of the first channel and the features of the second channel are fused using the fusion layer and the weight adjustment matrix.

[0041] The weights of the convolution kernel in the first channel are dynamically adjusted using the weight adjustment matrix to generate the final convolution kernel weights.

[0042] The fused feature map is generated and output based on the final convolutional kernel weights.

[0043] Optionally, the step of performing feature extraction and information extraction on the original mass spectrometry signal for the first channel includes:

[0044] For the first channel, local peak shape features of the original mass spectrum signal are extracted using a convolutional neural network; and

[0045] Extract time-domain information, including at least frequency and intensity variations, from the raw mass spectrometry signal. This time-domain information is related to the instrument noise and distortion of the raw mass spectrometry signal.

[0046] Optionally, before performing feature extraction and information extraction on the original mass spectrometry signal for the first channel, the method further includes:

[0047] For the first channel, the following operations are performed:

[0048] Set the kernel width of the first channel to a preset width to capture the local morphological features of the spectral peaks;

[0049] Set the activation function of the first channel to the ReLU activation function;

[0050] Set the pooling layer of the first channel to a max pooling layer to reduce the feature dimension.

[0051] Optionally, the step of generating and outputting a weight adjustment matrix for the second channel through a fully connected layer and a Softmax activation function includes:

[0052] For the second channel, the weighted feature vector under the first preset value dimension is input into the second channel;

[0053] The fully connected layer maps the weighted feature vectors under the first preset dimension into a set of weight matrices.

[0054] The weight adjustment matrix is ​​generated and output using the Softmax activation function.

[0055] Optionally, before performing standardization, principal component analysis, and feature weighting on the features extracted from the multi-dimensional sources in sequence, the method further includes:

[0056] Multiple preset methods are used to preprocess data from multiple dimensions to obtain preprocessed data; among them, the multiple preset methods include: sliding window partitioning method, cross sampling rate interpolation method, and dynamic time warping method.

[0057] Secondly, embodiments of the present invention provide a calibration device for helium concentration data based on mass spectrometry detection, the device comprising:

[0058] The first processing module is used to perform standardization, principal component analysis, and feature weighting on the features extracted from the multi-dimensional data in sequence to obtain a weighted feature vector under a first preset value dimension. The multi-dimensional data includes: mass spectrometry signal dimension, environmental parameter dimension, and magnetic field data dimension. The extraction methods used for multi-dimensional extraction include: data sampling extraction method and statistical analysis extraction method.

[0059] The second processing module is used to input the original mass spectrometry signal and the weighted feature vector under the first preset dimension into a dual-channel convolutional neural network for processing, and generate a fused feature map. The dual-channel convolutional neural network includes a first channel and a second channel, and the fused feature map integrates the features of the original mass spectrometry signal and the environmental features.

[0060] The output module is used to output the fused feature map;

[0061] The calculation module is used to calculate the distortion of the original mass spectrometry signal through a nonlinear distortion model to obtain the distorted mass spectrometry signal. The nonlinear distortion model can quantify the influence of different instrument states on the mass spectrometry signal.

[0062] The compensation module is used to compensate the feature map based on the distorted mass spectrum signal to obtain the compensated mass spectrum signal.

[0063] The dynamic inversion module is used to dynamically invert the current helium concentration value based on the compensated mass spectrometry signal and using an inverse problem-solving framework to obtain the corrected helium concentration value.

[0064] Thirdly, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method of the first aspect.

[0065] Fourthly, an electronic device is provided, including a memory and a processor, wherein the memory stores executable code, and the processor executes the executable code to implement the method of the first aspect.

[0066] In this embodiment of the invention, the features extracted from multiple dimensions are sequentially standardized, subjected to principal component analysis, and weighted feature processing to obtain a weighted feature vector under a first preset value dimension. The original mass spectrometry signal and the weighted feature vector under the first preset value dimension are input into a dual-channel convolutional neural network for processing to generate and output a fused feature map. The dual-channel convolutional neural network includes a first channel and a second channel, and the fused feature map integrates the features of the original mass spectrometry signal and environmental features. The distortion of the original mass spectrometry signal is calculated using a nonlinear distortion model to obtain a distorted mass spectrometry signal. The nonlinear distortion model can quantify the influence of different instrument states on the mass spectrometry signal. The feature map is compensated based on the distorted mass spectrometry signal to obtain a compensated mass spectrometry signal. Based on the compensated mass spectrometry signal, a reverse problem-solving framework is used to dynamically invert the current helium concentration value to obtain a corrected helium concentration value. The correction method provided in this embodiment of the invention can quantify and eliminate the interference of different instrument state changes on the signal. The corrected helium concentration value obtained by dynamically inverting the current helium concentration value based on the compensated mass spectrometry signal can reflect the actual helium concentration value. Attached Figure Description

[0067] Exemplary embodiments of the present invention can be more fully understood by referring to the accompanying drawings. The drawings are provided to further illustrate the embodiments of the present invention and form part of the specification. They are used together with the embodiments of the present invention to explain the invention and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0068] Figure 1 A flowchart illustrating a method for correcting helium concentration data based on mass spectrometry detection according to an exemplary embodiment of the present invention;

[0069] Figure 2 A schematic diagram of a calibration device for helium concentration data based on mass spectrometry detection according to an exemplary embodiment of the present invention. Detailed Implementation

[0070] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0071] It should be noted that, unless otherwise stated, the technical or scientific terms used in this invention should have the ordinary meaning as understood by one of ordinary skill in the art.

[0072] Furthermore, the terms "first" and "second," etc., are used to distinguish different objects, not to describe a specific order. Additionally, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to those processes, methods, products, or devices.

[0073] This invention provides a method and apparatus for correcting helium concentration data based on mass spectrometry detection, a computer-readable medium, and an electronic device, which are described below with reference to the accompanying drawings.

[0074] Please refer to Figure 1 The document illustrates flowcharts of a method for correcting helium concentration data based on mass spectrometry detection, provided by some embodiments of the present invention. Figure 1 As shown, the correction method for helium concentration data based on mass spectrometry detection may include the following steps:

[0075] Step S101: The features extracted from the multi-dimensional model are processed sequentially by standardization, principal component analysis, and feature weighting to obtain a weighted feature vector under the first preset value dimension; the multi-dimensional model includes: mass spectrometry signal dimension, environmental parameter dimension, and magnetic field data dimension; the extraction methods used for multi-dimensional extraction include: data sampling extraction method and statistical analysis extraction method.

[0076] In a specific application scenario, the first preset value is 128.

[0077] It should be noted that the features extracted from the above multi-dimensional sources may include: mass spectrometry signal dimension features, environmental parameter dimension features, and magnetic field data dimension features; the above are merely examples and are not limited to the features extracted from the above multi-dimensional sources.

[0078] The features extracted from the above multi-dimensional sources are explained as follows:

[0079] Dimensional characteristics of mass spectrometry signals: Helium characteristic peaks ( ) and adjacent mass number ( The time-domain signal characteristics of the signal include, but are not limited to, time-domain statistical characteristics such as mean, standard deviation, maximum value, minimum value, and fluctuation amplitude; frequency-domain characteristics such as frequency components and amplitude after Fourier transform; and peak characteristics such as peak number, peak area, peak curvature, and peak width.

[0080] Environmental parameter dimensions: ionization chamber vacuum level, ionization temperature ( Related characteristics, including but not limited to statistical characteristics such as mean, standard deviation, fluctuation, maximum and minimum values ​​of vacuum degree and temperature, trend characteristics such as local change rate calculated by sliding window, and coupling relationship characteristics between vacuum degree and temperature.

[0081] Magnetic field data dimensional characteristics: Magnetic field data characteristics of ion transport paths, including but not limited to magnetic field spatial gradient, periodic changes, fluctuation amplitude and other magnetic field variation characteristics, as well as the symmetry or uniformity of magnetic field distribution.

[0082] The extraction methods used in multi-dimensional extraction include data sampling extraction methods and statistical analysis extraction methods. The above are merely examples and are not limited to these methods. The extraction methods described above are conventional and will not be elaborated upon further.

[0083] In practical applications, for example, for the mass spectrometry signal dimension, the time-domain signals of helium characteristic peaks and adjacent mass numbers are sampled at preset time intervals to calculate the original 32-dimensional time-domain, frequency-domain, and peak characteristics.

[0084] The above preset time intervals can be configured according to the needs of different application scenarios, and no specific limitation is made on the preset time intervals here.

[0085] In one example, the features extracted from multiple dimensions are sequentially standardized, subjected to principal component analysis, and weighted processing to obtain a weighted feature vector under the first preset dimension, including the following steps:

[0086] Obtain multiple dimensions;

[0087] For any one of the multiple dimensions, extract multiple original features under the second preset value dimension in sequence;

[0088] All original features are standardized to obtain standardized features.

[0089] For the standardized features, principal component analysis is used to select the principal components with a first preset number of values.

[0090] The principal components with the first preset number of values ​​are subjected to feature weighting to obtain the weighted feature vector under the first preset value dimension.

[0091] In a specific application scenario, the second preset value is 32.

[0092] In practical applications, 32 original features are extracted for each dimension; the extracted features are standardized; principal components are selected using principal component analysis, and the first 128 principal components are retained.

[0093] The 128 selected principal components are weighted by feature importance to obtain a 128-dimensional weighted feature vector;

[0094] Formula (1);

[0095] In the formula: Indicates the weighting coefficients for feature importance; Represents the corresponding eigenvector eigenvalues; Represents a diagonal weight matrix; This represents the standardized feature vector; V represents the final feature vector after feature fusion and weighting, with a dimension of 128, which serves as the input to the subsequent model; V is the feature vector matrix of the first 128 principal components.

[0096] In practical applications, before performing standardization, principal component analysis, and feature weighting on the features extracted from multiple dimensions in sequence, the correction method provided in this embodiment of the invention further includes the following steps:

[0097] Simultaneous collection of multi-dimensional data.

[0098] The process of synchronously collecting multi-dimensional data is described in detail below:

[0099] Mass spectrometry dimension: Complete acquisition of the time-domain signal of the helium characteristic peak (m / z=4) and its adjacent mass numbers (m / z=3, 5);

[0100] Environmental dimension: Real-time monitoring of the vacuum level in the ionization chamber ( ), ionization temperature (150-300℃);

[0101] Time-series dimension: Record the continuous running time of the instrument (0-48 hours) and the sampling interval (1-60 seconds);

[0102] Spatial dimension: Install multiple probes to detect the magnetic field uniformity (±0.3% gradient) of the ion transport path.

[0103] It should be noted that by synchronously acquiring the above multi-dimensional data, accurate modeling of the instrument's status can be achieved, ensuring that mass spectrometry signals, environmental parameters, and magnetic field data remain consistent in terms of time and characteristics, thus providing a reliable data foundation for subsequent analysis.

[0104] In practical applications, before performing standardization, principal component analysis, and feature weighting on the features extracted from multiple dimensions in sequence, the correction method provided in this embodiment of the invention further includes the following steps:

[0105] Multimodal features are extracted.

[0106] The process of extracting multimodal features is described in detail below:

[0107] For example, feature extraction of a mass spectrometry signal includes the following steps:

[0108] Time-domain characteristics: mean, standard deviation, maximum value, minimum value, fluctuation range, etc.

[0109] Frequency domain features: Frequency components and their amplitudes are extracted using methods such as Fourier transform.

[0110] Peak characteristics: number of peaks, peak area, peak curvature, peak width, etc.

[0111] Feature extraction of environmental parameters includes:

[0112] Basic statistical characteristics: mean, standard deviation, fluctuation, maximum value, minimum value.

[0113] Trend analysis: Calculate the local rate of change using a sliding window.

[0114] Coupling analysis: such as the coupling relationship between vacuum degree and temperature.

[0115] Feature extraction of magnetic field data includes:

[0116] Spatial gradient: the rate of change of the magnetic field.

[0117] Volatility analysis: such as periodic changes, volatility amplitude, etc.

[0118] Symmetry: The symmetry or uniformity of the magnetic field distribution.

[0119] It should be noted that the purpose of feature extraction is to ensure the comprehensiveness of the data through a combination of different techniques, that is, to extract as much information as possible from each dimension to ensure the quality of subsequent data fusion and model training.

[0120] In the calibration method provided in this embodiment of the invention, time-series features are extracted from mass spectrometry signals, environmental parameters and magnetic field data through multimodal feature encoding to construct a multi-dimensional data stream containing the physical state of the instrument.

[0121] In one example, before performing standardization, principal component analysis, and feature weighting on the features extracted from multiple dimensions in sequence, the correction method for helium concentration data based on mass spectrometry detection provided in this embodiment of the invention may further include the following steps:

[0122] Multiple preset methods are used to preprocess data from multiple dimensions to obtain preprocessed data; among these preset methods are: sliding window partitioning, cross-sampling rate interpolation, and dynamic time warping.

[0123] If the current preset method is the sliding window partitioning method, the data preprocessing process is as follows:

[0124] Because different types of data (mass spectrometry signals, environmental parameters, magnetic field data) have different sampling frequencies, it is necessary to align these data with a unified time reference for subsequent analysis. Therefore, time windows are divided according to the sampling period of the mass spectrometry signal (i.e., data in the mass spectrometry dimension), and the window length is adaptively adjusted.

[0125] Formula (2);

[0126] in: This represents the range of the k-th time window; This represents the timestamp (μs precision) of the k-th mass spectrometry sampling point.

[0127] In a specific application scenario, the sampling period can be... The sampling period is set to 0.1ms. There is no specific limit to the duration of the sampling period. The sampling period can be adjusted according to the needs of different application scenarios.

[0128] If the current preset method is cross-sampling rate interpolation, the data preprocessing process is as follows:

[0129] Because the sampling frequency of mass spectrometry signals differs from that of other data, interpolation methods must be used to increase the sampling rate of low-frequency data to ensure it is perfectly aligned with the mass spectrometry signal in time. Cubic spline interpolation can smooth the data, avoid unnecessary overfitting, and ensure the true trend of data variation. Specifically, cubic spline interpolation is performed on low-frequency data (environmental parameters, magnetic fields):

[0130] Formula (3);

[0131] in: This represents the vacuum interpolation function, i.e., the predicted vacuum level at a given time t; , , , The coefficients represent the cubic spline interpolation coefficients, which are calculated using data between interpolation points.

[0132] If the current preset method is dynamic time warping, the data preprocessing process is as follows:

[0133] Calculate the alignment path between the mass spectrum signal and the interpolated environmental parameters:

[0134] Formula (4);

[0135] In the formula: This represents the cumulative distance between the i-th point of the mass spectrometer and the j-th point of the environmental parameter; This represents the mass spectrum signal at time i. Let represent the environmental parameters at time j. Due to the different sampling rates, direct linear alignment may lead to errors. The dynamic time warping algorithm is used to calculate the alignment path between the mass spectrometry signal and the interpolated environmental parameters, thereby making the alignment between the high-frequency mass spectrometry signal and the low-frequency environmental parameters more accurate and reducing the error to below 0.5 μs.

[0136] It should be noted that a sliding window dynamic time warping algorithm is used to align the timestamps of mass spectrometry signals, environmental parameters, and magnetic field data, achieving synchronization across sampling rates. By minimizing the mutual information entropy of the data within the window, sub-microsecond alignment between data from different dimensions is ensured.

[0137] Step S102: Input the original mass spectrum signal and the weighted feature vector under the first preset value dimension into a dual-channel convolutional neural network for processing, generate and output the fused feature map. The dual-channel convolutional neural network includes a first channel and a second channel. The fused feature map integrates the features of the original mass spectrum signal and the environmental features.

[0138] In a specific application scenario, the first and second channels of the constructed dual-channel convolutional neural network (represented by channel A and channel B in this scenario) are explained as follows:

[0139] Channel A: Input the raw mass spectrum signal and extract local peak shape features through a 1D convolutional neural network (e.g., CNN); the convolutional kernel width is set to 5 amu to capture the local morphological features of the spectral peaks.

[0140] It should be noted that in mass spectrometry signal processing, the convolution kernel width is used to capture signal features within a specific mass number range. The phrase "the convolution kernel width is set to 5 amu" means that the convolution kernel covers a mass number range of 5 atomic mass units (amu), which can effectively capture helium characteristic peaks. ) and adjacent mass number ( The local peak shape characteristics of mass spectrometry. amu is the atomic mass unit, the standard unit of measurement for mass number in mass spectrometry.

[0141] The goal of this process is to extract time-domain information such as frequency and intensity variations from the original signal. This information is related to instrument noise and distortion in the mass spectrometry signal. The ReLU activation function is used, and max pooling is employed to reduce feature dimensionality while preserving important features. The specific formula is as follows:

[0142] Formula (5);

[0143] In the formula: This represents the signal characteristics after convolution; Indicates a time step; Indicates the weights of the convolution kernel; This indicates the bias term.

[0144] Channel B: Input a 128-dimensional feature vector; the 128-dimensional feature vector is mapped to a set of weight matrices through a fully connected layer, and a weight adjustment matrix is ​​generated through the Softmax activation function to help with subsequent feature fusion and dynamic adjustment of convolution kernels.

[0145] Fusion layer: By fusing the features of channel A and channel B, and dynamically adjusting the convolution kernel weights of channel A, the model can adapt to changes in different instrument states;

[0146] Formula (6);

[0147] in: Represents the weight adjustment matrix; The basic convolutional kernel weights are represented by F; F represents the 128-dimensional feature vector. This indicates element-wise multiplication.

[0148] By dynamically adjusting the weights of the A-channel convolutional kernel using the output of the B channel, the dual-channel convolutional neural network can dynamically adapt to different instrument conditions, thereby improving its ability to extract the true signal from the mass spectrometry signal.

[0149] Specifically, the convolutional kernel weights of channel A are dynamically adjusted using the weight matrix output from channel B to generate the final convolutional kernel weights. :

[0150] Formula (7);

[0151] in: This represents the final convolutional kernel weights. Represents the weight adjustment matrix; This represents the weights of the basic convolution kernel.

[0152] Finally, based on the dynamically adjusted convolutional kernel weights, a fused feature map is output, which combines the features of the original mass spectrometry signal and the environmental features.

[0153] It should be noted that by using a dual-channel convolutional neural network (DC-CNN), the original mass spectrometry signal and multi-dimensional feature vectors are input separately, and the weights of the convolution kernel are dynamically adjusted to achieve adaptive feature extraction based on the instrument status.

[0154] In one example, the raw mass spectrum signal and the weighted feature vector under a first preset dimension are input into a dual-channel convolutional neural network for processing, including the following steps:

[0155] For the first channel, feature extraction and information extraction are performed on the raw mass spectrometry signal;

[0156] For the second channel, a weight adjustment matrix is ​​generated and output through a fully connected layer and a Softmax activation function;

[0157] The features of the first channel and the features of the second channel are fused using a fusion layer and a weight adjustment matrix.

[0158] The weights of the convolution kernel in the first channel are dynamically adjusted using a weight adjustment matrix to generate the final convolution kernel weights.

[0159] The final convolutional kernel weights are used to generate and output the fused feature map.

[0160] It should be noted that in the correction method provided in the embodiments of the present invention, by analyzing the dynamic spectral peak distortion, a dual-channel convolutional neural network (DC-CNN) combined with multi-dimensional features can adaptively adjust the weights of the convolutional kernel.

[0161] In one example, for the first channel, feature extraction and information extraction are performed on the raw mass spectrometry signal, including the following steps:

[0162] For the first channel, local peak shape features of the original mass spectrum signal are extracted using a convolutional neural network; and

[0163] Extract time-domain information, including at least frequency and intensity variations, from the raw mass spectrometry signal. This time-domain information is related to the instrument noise and distortion of the raw mass spectrometry signal.

[0164] In one example, before performing feature extraction and information extraction on the raw mass spectrometry signal for the first channel, the correction method for helium concentration data based on mass spectrometry detection may further include the following steps:

[0165] For the first channel, the following operations are performed:

[0166] Set the kernel width of the first channel to a preset width to capture the local morphological features of the spectral peaks;

[0167] Set the activation function of the first channel to the ReLU activation function;

[0168] Set the pooling layer of the first channel to a max pooling layer to reduce the feature dimension.

[0169] In one example, for the second channel, the generation and output of the fused feature map, through the generation and output of a weight adjustment matrix via a fully connected layer and a Softmax activation function, includes the following steps:

[0170] For the second channel, the weighted feature vector under the first preset value dimension is input into the second channel;

[0171] The weighted feature vectors under the first preset dimension are mapped to a set of weight matrices through a fully connected layer;

[0172] The weight adjustment matrix is ​​generated and output using the Softmax activation function.

[0173] Step S103: The distortion of the original mass spectrometry signal is calculated using a nonlinear distortion model to obtain the distorted mass spectrometry signal. The nonlinear distortion model can quantify the influence of different instrument states on the mass spectrometry signal.

[0174] In practical applications, before calculating the distortion of the original mass spectrum signal using a nonlinear distortion model, the correction provided in this embodiment of the invention may further include the following steps:

[0175] A nonlinear distortion model is established, which can quantify the influence of instrument status on the mass spectrometry signal:

[0176] Formula (8);

[0177] In the formula: This represents the distorted mass spectrum signal; This represents the distortion coefficient related to time decay; Represents a constant; Indicates the vibration frequency; The distortion coefficient represents the correlation between the intensity of periodic distortion caused by vibration. This represents the distortion coefficient related to changes in vacuum level; This represents the gradient of vacuum level.

[0178] It should be noted that by establishing a nonlinear distortion model, quantifying the distortion factor of the signal, and combining it with FFT analysis of magnetic field data, the distortion model parameters are updated in real time.

[0179] The distortion coefficients are updated in real time using the stochastic gradient descent method. , , It captures real-time changes in instrument status; it captures and quantifies the impact of factors such as magnetic field changes and equipment vibration on the signal, and provides a mathematical model for subsequent signal compensation.

[0180] In practical applications, different instrument states can be quantified, including different vibration states, different magnetic field changes, and different vacuum level changes. The above are merely examples; different instrument states can be introduced according to the needs of different application scenarios, which will not be elaborated further here.

[0181] In one example, before calculating the distortion of the original mass spectrometry signal using a nonlinear distortion model, the correction method for helium concentration data based on mass spectrometry detection provided in this embodiment of the invention may further include the following steps:

[0182] Multiple distortion coefficients of the nonlinear distortion model are obtained. These distortion coefficients include: a first distortion coefficient related to time decay, a second distortion coefficient related to the intensity of periodic distortion caused by vibration, and a third distortion coefficient related to the change in vacuum degree.

[0183] The first, second, and third distortion coefficients are updated in real time using the stochastic gradient descent method to capture changes in the instrument's state. The first distortion coefficient can be... This indicates that the second distortion coefficient can be expressed as... This indicates that the third distortion coefficient can be expressed as... express.

[0184] Step S104: Compensate the feature map based on the distorted mass spectrum signal to obtain the compensated mass spectrum signal.

[0185] In a specific application scenario, the distorted mass spectrum signal obtained through calculation is used to compensate for the feature map output by the dual-channel convolutional neural network:

[0186] Formula (9);

[0187] In the formula: This indicates the compensated signal; This represents the feature map output by the dual-channel convolutional neural network. It should be noted that the feature map output by the convolutional neural network cannot be directly used in the computation. Therefore, it can be restored to the feature signal through deconvolution.

[0188] It should be noted that, based on the distorted mass spectrum signal, the feature map output by the dual-channel convolutional neural network (DC-CNN) is compensated, and the feature signal is restored through deconvolution operation to achieve signal distortion compensation.

[0189] Step S105: Based on the compensated mass spectrometry signal, the current helium concentration value is dynamically inverted using an inverse problem-solving framework to obtain the corrected helium concentration value.

[0190] It should be noted that the reverse problem-solving framework includes the following modules:

[0191] Model module: Establishes a physical model of the relationship between mass spectrometry signals and helium concentration. For example, it describes how concentration affects the intensity and shape of the mass spectrometry signal; using physical laws or empirical models, it transforms concentration parameters into predictable mass spectrometry signals.

[0192] Optimization module: Finds the concentration parameters that best match the model output to the observed data. Optimization algorithms, such as gradient descent or conjugate gradient methods, are used to minimize the difference between the two. The concentration values ​​are iteratively adjusted until the model-predicted mass spectrometry signal matches the compensated signal.

[0193] Regularization module: Introduces prior knowledge, such as physical limitations or smoothness of concentration, to prevent overfitting and unstable solutions. Regularization terms are added during the optimization process to ensure the rationality of the solution, for example, to limit the drastic degree of concentration change.

[0194] Validation module: Evaluates the rationality and accuracy of the solution, ensures that the results meet expectations, checks whether the concentration solution is reasonable through cross-validation or comparison with known data, and adjusts the model or parameters to improve accuracy.

[0195] In one example, based on the compensated mass spectrometry signal, an inverse problem-solving framework is used to dynamically invert the initial helium concentration value to obtain the corrected helium concentration value, including the following steps:

[0196] An optimization problem is constructed based on the inverse problem-solving framework, which is a solution framework built on a physical model and a regularization method.

[0197] Based on the optimization problem, the difference between the compensated mass spectrometry signal and the theoretical mass spectrometry signal is minimized; in the process of minimizing the difference between the compensated mass spectrometry signal and the theoretical mass spectrometry signal, a regularization term based on prior knowledge is introduced.

[0198] The optimization problem that minimizes the difference between the compensated mass spectrometry signal and the theoretical mass spectrometry signal is taken as the optimal problem.

[0199] The optimal problem is solved to obtain the corrected helium concentration value.

[0200] In a specific application scenario, the regularization term based on prior knowledge introduced in the process of minimizing the difference between the compensated mass spectrometry signal and the theoretical mass spectrometry signal can be the Tikhonov regularization term.

[0201] It should be noted that an inverse problem-solving framework is adopted, which minimizes the difference between the physical model of the compensated signal and the mass spectrometry signal, and introduces a Tikhonov regularization term based on prior knowledge to achieve high-precision helium concentration calculation.

[0202] In a specific application scenario, based on the compensated signal The specific steps for dynamically inverting helium concentration using an inverse problem-solving framework are as follows:

[0203] Construct an optimization problem with the objective of minimizing the compensated mass spectrometry signal. Compared with theoretical mass spectrum signal The difference between them is addressed by adding a prior knowledge-based regularization using the following formula (10). :

[0204] Formula (10);

[0205] In the formula: This represents the compensated mass spectrometry signal; This is the theoretical mass spectrometry signal.

[0206] Formula (11);

[0207] Where: A represents the calibration factor, obtained from instrument calibration; Indicates helium concentration; represents the mass-to-charge ratio; k represents the Boltzmann constant; T represents the temperature.

[0208] Based on solving the above optimal problem, the corrected helium concentration value is obtained.

[0209] In one example, after obtaining the corrected helium concentration value, the correction method for helium concentration data based on mass spectrometry detection provided in this embodiment of the invention may further include the following steps:

[0210] The Monte Carlo Dropout method is used to quantify the uncertainty of helium concentration based on multiple sampled helium concentration values ​​and multiple predicted helium concentration values, and the quantification results are obtained.

[0211] It should be noted that Dropout is a regularization strategy that reduces the interdependence between feature detectors by randomly shutting down neurons during training, thereby improving the model's generalization ability.

[0212] In one example, the Monte Carlo Dropout method is used to quantify the uncertainty of helium concentration based on multiple sampled helium concentration values ​​and multiple predicted helium concentration values, obtaining the quantification result, including the following steps:

[0213] For a single sample, the following operations are performed:

[0214] The helium concentration value was obtained by sampling.

[0215] During the forward propagation of a dual-channel convolutional neural network, a predetermined proportion of neurons are randomly discarded to obtain a set of neural network parameters;

[0216] The helium concentration is predicted by using this set of neural network parameters, and the predicted helium concentration is obtained.

[0217] Repeat the single sampling operation multiple times to obtain multiple sampled helium concentration values ​​and corresponding multiple predicted helium concentration values;

[0218] Based on multiple sampled helium concentration values ​​and multiple predicted helium concentration values, the mean and variance of the helium concentration are calculated to obtain the mean and variance of the helium concentration.

[0219] The uncertainty of helium concentration is quantified based on the mean and variance of helium concentration, and the quantification results are obtained.

[0220] In a specific application scenario, to quantify the uncertainty in helium concentration calculation, the Monte Carlo Dropout method is used for error analysis. During the forward propagation of the dual-channel convolutional neural network, 20% of the neurons are randomly dropped, and multiple samplings are performed (e.g., M=100 times), calculating the helium concentration for each sampling. Specifically, as described below:

[0221] In each forward propagation, 20% of the neurons are randomly discarded, resulting in a set of neural network parameters. ;

[0222] Use the obtained neural network parameters Prediction was performed to obtain the helium concentration. ;

[0223] The mean and variance of the concentration are calculated by sampling multiple times, thereby quantifying the uncertainty.

[0224] It should be noted that, in order to quantify the calculation error, the Monte Carlo Dropout method was used to perform uncertainty propagation calculations. By sampling multiple times to quantify the error, the accuracy of concentration inversion was further improved.

[0225] In the correction method provided in this embodiment of the invention, during the dynamic concentration inversion process, a framework for solving the inverse problem is constructed by combining a physical model and a regularization method, and Monte Carlo Dropout uncertainty analysis is introduced, which not only makes the helium concentration calculation highly accurate, but also quantifies the error and improves the reliability of the results; it effectively overcomes the measurement error caused by changes in instrument status and improves the stability and accuracy of helium concentration calculation under complex environments.

[0226] The calibration method provided in this invention can quantify and eliminate the interference of different instrument state changes (e.g., changes in vacuum level, changes in magnetic field gradient, changes in vibration noise, etc.) on the signal; based on the compensated mass spectrometry signal, the calibrated helium concentration value obtained by dynamically inverting the current helium concentration value can reflect the actual helium concentration value.

[0227] In the above embodiments, a method for correcting helium concentration data based on mass spectrometry detection is provided. Correspondingly, the present invention also provides a device for correcting helium concentration data based on mass spectrometry detection. The device for correcting helium concentration data based on mass spectrometry detection provided in the embodiments of the present invention can implement the above-described method for correcting helium concentration data based on mass spectrometry detection. This device can be implemented through software, hardware, or a combination of both. For example, the device can include integrated or separate functional modules or units to perform the corresponding steps in the above methods.

[0228] Please refer to Figure 2 This illustration shows a schematic diagram of a helium concentration data correction device based on mass spectrometry detection, provided by some embodiments of the present invention. Since the device embodiments are substantially similar to the method embodiments, the description is relatively simple; relevant details can be found in the description of the method embodiments. The device embodiments described below are merely illustrative.

[0229] like Figure 2 As shown, the calibration device for helium concentration data based on mass spectrometry detection may include:

[0230] The first processing module 201 is used to perform standardization, principal component analysis, and feature weighting on the features extracted from the multi-dimensional data in sequence to obtain a weighted feature vector under a first preset value dimension. The multi-dimensional data includes: mass spectrometry signal dimension, environmental parameter dimension, and magnetic field data dimension. The extraction methods used for multi-dimensional extraction include: data sampling extraction method and statistical analysis extraction method.

[0231] The second processing module 202 is used to input the original mass spectrum signal and the weighted feature vector under the first preset value dimension into a dual-channel convolutional neural network for processing, and generate a fused feature map. The dual-channel convolutional neural network includes a first channel and a second channel. The fused feature map integrates the features of the original mass spectrum signal and the environmental features.

[0232] Output module 203 is used to output the fused feature map;

[0233] The calculation module 204 is used to calculate the distortion of the original mass spectrometry signal through a nonlinear distortion model to obtain the distorted mass spectrometry signal. The nonlinear distortion model can quantify the influence of different instrument states on the mass spectrometry signal.

[0234] The compensation module 205 is used to compensate the feature map based on the distorted mass spectrum signal to obtain the compensated mass spectrum signal.

[0235] The dynamic inversion module 206 is used to dynamically invert the current helium concentration value based on the compensated mass spectrometry signal and adopt an inverse problem-solving framework to obtain the corrected helium concentration value.

[0236] In some embodiments of the present invention, the dynamic inversion module 206 is specifically used for:

[0237] An optimization problem is constructed based on the inverse problem-solving framework, which is a solution framework built on a physical model and a regularization method.

[0238] Based on the optimization problem, the difference between the compensated mass spectrometry signal and the theoretical mass spectrometry signal is minimized; in the process of minimizing the difference between the compensated mass spectrometry signal and the theoretical mass spectrometry signal, a regularization term based on prior knowledge is introduced.

[0239] The optimization problem that minimizes the difference between the compensated mass spectrometry signal and the theoretical mass spectrometry signal is taken as the optimal problem.

[0240] The optimal problem is solved to obtain the corrected helium concentration value.

[0241] In some embodiments of the present invention, the correction device for helium concentration data based on mass spectrometry detection may further include:

[0242] Quantization module (in) Figure 2 (Not shown in the image) is used to quantify the uncertainty of helium concentration using the Monte Carlo Dropout method after obtaining the corrected helium concentration value. This is based on multiple sampled helium concentration values ​​and multiple predicted helium concentration values, resulting in a quantification result.

[0243] In some embodiments of the present invention, the quantization module is specifically used for:

[0244] For a single sample, the following operations are performed:

[0245] The helium concentration value was obtained by sampling.

[0246] During the forward propagation of a dual-channel convolutional neural network, a predetermined proportion of neurons are randomly discarded to obtain a set of neural network parameters;

[0247] The helium concentration is predicted by using this set of neural network parameters, and the predicted helium concentration is obtained.

[0248] Repeat the single sampling operation multiple times to obtain multiple sampled helium concentration values ​​and corresponding multiple predicted helium concentration values;

[0249] Based on multiple sampled helium concentration values ​​and multiple predicted helium concentration values, the mean and variance of the helium concentration are calculated to obtain the mean and variance of the helium concentration.

[0250] The uncertainty of helium concentration is quantified based on the mean and variance of helium concentration, and the quantification results are obtained.

[0251] In some embodiments of the present invention, the correction device for helium concentration data based on mass spectrometry detection may further include:

[0252] Get module (in) Figure 2 (not shown in the image) is used to: obtain multiple distortion coefficients of the nonlinear distortion model before calculating the distortion of the original mass spectrum signal using the nonlinear distortion model. The multiple distortion coefficients include: a first distortion coefficient related to time decay, a second distortion coefficient related to the intensity of periodic distortion caused by vibration, and a third distortion coefficient related to the change in vacuum degree.

[0253] Real-time update module (in) Figure 2 (Not shown in the image) is used to: update the first distortion coefficient, the second distortion coefficient, and the third distortion coefficient in real time using the stochastic gradient descent method to capture changes in the instrument's state in real time.

[0254] In some embodiments of the present invention, the first processing module 201 is specifically used for:

[0255] Obtain multiple dimensions;

[0256] For any one of the multiple dimensions, extract multiple original features under the second preset value dimension in sequence;

[0257] All original features are standardized to obtain standardized features.

[0258] For the standardized features, principal component analysis is used to select the principal components with a first preset number of values.

[0259] The principal components with the first preset number of values ​​are subjected to feature weighting to obtain the weighted feature vector under the first preset value dimension.

[0260] In some embodiments of the present invention, the second processing module 202 is specifically used for:

[0261] For the first channel, feature extraction and information extraction are performed on the raw mass spectrometry signal;

[0262] For the second channel, a weight adjustment matrix is ​​generated and output through a fully connected layer and a Softmax activation function;

[0263] The features of the first channel and the features of the second channel are fused using a fusion layer and a weight adjustment matrix.

[0264] The weights of the convolution kernel in the first channel are dynamically adjusted using a weight adjustment matrix to generate the final convolution kernel weights.

[0265] The final convolutional kernel weights are used to generate and output the fused feature map.

[0266] In some embodiments of the present invention, the second processing module 202 is specifically used for:

[0267] For the first channel, local peak shape features of the original mass spectrum signal are extracted using a convolutional neural network; and

[0268] Extract time-domain information, including at least frequency and intensity variations, from the raw mass spectrometry signal. This time-domain information is related to the instrument noise and distortion of the raw mass spectrometry signal.

[0269] In some embodiments of the present invention, the correction device for helium concentration data based on mass spectrometry detection may further include:

[0270] Settings module (in) Figure 2 (Not shown in the image), used for: performing the following operations on the first channel before performing feature extraction and information extraction on the raw mass spectrometry signal for the first channel:

[0271] Set the kernel width of the first channel to a preset width to capture the local morphological features of the spectral peaks;

[0272] Set the activation function of the first channel to the ReLU activation function;

[0273] Set the pooling layer of the first channel to a max pooling layer to reduce the feature dimension.

[0274] In some embodiments of the present invention, the second processing module 202 is specifically used for:

[0275] For the second channel, the weighted feature vector under the first preset value dimension is input into the second channel;

[0276] The weighted feature vectors under the first preset dimension are mapped to a set of weight matrices through a fully connected layer;

[0277] The weight adjustment matrix is ​​generated and output using the Softmax activation function.

[0278] In some embodiments of the present invention, the correction device for helium concentration data based on mass spectrometry detection may further include:

[0279] Preprocessing module (in) Figure 2 (Not shown in the image) is used to: preprocess the multi-dimensional data using various preset methods before performing standardization, principal component analysis, and feature weighting on the features extracted from multiple dimensions in sequence, so as to obtain preprocessed data; among them, the various preset methods include: sliding window partitioning method, cross-sampling rate interpolation method, and dynamic time warping method.

[0280] In some embodiments of the present invention, the helium concentration data correction device based on mass spectrometry detection provided in the present invention is based on the same inventive concept and has the same beneficial effects as the helium concentration data correction method based on mass spectrometry detection provided in the foregoing embodiments of the present invention.

[0281] According to another embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed in a computer, causes the computer to perform a combination Figure 1 The method described.

[0282] According to another embodiment, an electronic device is also provided, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, it implements a combination... Figure 1 The method described.

[0283] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using hardware, software, firmware, or any combination thereof. When implemented in software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium.

[0284] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for correcting helium concentration data based on mass spectrometry detection, characterized in that, The method includes: The features extracted from multiple dimensions are sequentially standardized, subjected to principal component analysis, and weighted feature processing to obtain a weighted feature vector under the first preset value dimension; the multiple dimensions include: mass spectrometry signal dimension, environmental parameter dimension, and magnetic field data dimension; the extraction methods used for multi-dimensional extraction include: data sampling extraction method and statistical analysis extraction method; The original mass spectrometry signal and the weighted feature vector under the first preset dimension are input into a dual-channel convolutional neural network for processing to generate and output a fused feature map. The dual-channel convolutional neural network includes a first channel and a second channel. The fused feature map integrates the features of the original mass spectrometry signal and the environmental features. Multiple distortion coefficients of a nonlinear distortion model are obtained, including: a first distortion coefficient related to time decay, a second distortion coefficient related to the intensity of periodic distortion caused by vibration, and a third distortion coefficient related to changes in vacuum level; this nonlinear distortion model can quantify the influence of instrument status on the mass spectrometry signal. In the formula: This represents the distorted mass spectrum signal; This represents the distortion coefficient related to time decay; Represents a constant; Indicates the vibration frequency; The distortion coefficient represents the correlation between the intensity of periodic distortion caused by vibration. This represents the distortion coefficient related to changes in vacuum level; The gradient representing the degree of vacuum; The first distortion coefficient, the second distortion coefficient, and the third distortion coefficient are updated in real time using the stochastic gradient descent method to capture changes in the instrument's state in real time. The distortion of the original mass spectrometry signal is calculated by a nonlinear distortion model to obtain the distorted mass spectrometry signal. The nonlinear distortion model can quantify the influence of different instrument states on the mass spectrometry signal. The feature map is compensated based on the distorted mass spectrum signal to obtain the compensated mass spectrum signal. Based on the compensated mass spectrometry signal, an inverse problem-solving framework is used to dynamically invert the current helium concentration value to obtain the corrected helium concentration value.

2. The correction method according to claim 1, characterized in that, Based on the compensated mass spectrometry signal, an inverse problem-solving framework is used to dynamically invert the current helium concentration value to obtain the corrected helium concentration value, including: An optimization problem is constructed based on the aforementioned inverse problem-solving framework, wherein the inverse problem-solving framework is a solution framework constructed based on a physical model and a regularization method; Based on the optimization problem, the difference between the compensated mass spectrometry signal and the theoretical mass spectrometry signal is minimized; wherein, a regularization term based on prior knowledge is introduced in the process of minimizing the difference between the compensated mass spectrometry signal and the theoretical mass spectrometry signal. The optimization problem that minimizes the difference between the compensated mass spectrometry signal and the theoretical mass spectrometry signal is taken as the optimal problem. The optimal problem is solved to obtain the corrected helium concentration value.

3. The correction method according to claim 2, characterized in that, After obtaining the corrected helium concentration value, the method further includes: The Monte Carlo Dropout method is used to quantify the uncertainty of helium concentration based on multiple sampled helium concentration values ​​and multiple predicted helium concentration values, and the quantification results are obtained.

4. The correction method according to claim 3, characterized in that, The Monte Carlo Dropout method is used to quantify the uncertainty of helium concentration based on multiple sampled helium concentration values ​​and multiple predicted helium concentration values, resulting in the following quantification results: For a single sample, the following operations are performed: The helium concentration value was obtained by sampling. During the forward propagation of a dual-channel convolutional neural network, a predetermined proportion of neurons are randomly discarded to obtain a set of neural network parameters; The helium concentration is predicted by using this set of neural network parameters, and the predicted helium concentration is obtained. Repeat the single sampling operation multiple times to obtain multiple sampled helium concentration values ​​and corresponding multiple predicted helium concentration values; Based on multiple sampled helium concentration values ​​and multiple predicted helium concentration values, the mean and variance of the helium concentration are calculated to obtain the mean and variance of the helium concentration. The uncertainty of helium concentration is quantified based on the mean and variance of helium concentration, resulting in the quantification result.

5. The correction method according to claim 1, characterized in that, The features extracted from multiple dimensions are sequentially standardized, subjected to principal component analysis, and weighted feature processing to obtain a weighted feature vector under a first preset dimension, including: Obtain the aforementioned multi-dimensional data; For any one of the multiple dimensions, extract multiple original features under the second preset value dimension in sequence; All original features are standardized to obtain standardized features. For the standardized features, principal component analysis is used to select a first preset number of principal components. The principal components with the first preset number of values ​​are subjected to feature weighting to obtain the weighted feature vector under the first preset value dimension.

6. The correction method according to claim 1, characterized in that, The process of inputting the original mass spectrum signal and the weighted feature vector under the first preset dimension into a dual-channel convolutional neural network for processing, generating and outputting the fused feature map includes: For the first channel, feature extraction and information extraction are performed on the original mass spectrometry signal; For the second channel, a weight adjustment matrix is ​​generated and output through a fully connected layer and a Softmax activation function; The features of the first channel and the features of the second channel are fused using the fusion layer and the weight adjustment matrix. The weights of the convolution kernel in the first channel are dynamically adjusted using the weight adjustment matrix to generate the final convolution kernel weights. The fused feature map is generated and output based on the final convolutional kernel weights.

7. The correction method according to claim 6, characterized in that, The step of extracting features and information from the original mass spectrometry signal for the first channel includes: For the first channel, local peak shape features of the original mass spectrum signal are extracted using a convolutional neural network; and Extract time-domain information, including at least frequency and intensity variations, from the raw mass spectrometry signal. This time-domain information is related to the instrument noise and distortion of the raw mass spectrometry signal.

8. The correction method according to claim 7, characterized in that, Before performing feature extraction and information extraction on the original mass spectrometry signal for the first channel, the method further includes: For the first channel, the following operations are performed: Set the kernel width of the first channel to a preset width to capture the local morphological features of the spectral peaks; Set the activation function of the first channel to the ReLU activation function; Set the pooling layer of the first channel to a max pooling layer to reduce the feature dimension.

9. The correction method according to claim 6, characterized in that, The process for generating and outputting a weight adjustment matrix for the second channel using a fully connected layer and a Softmax activation function includes: For the second channel, the weighted feature vector under the first preset value dimension is input into the second channel; The fully connected layer maps the weighted feature vectors under the first preset dimension into a set of weight matrices. The weight adjustment matrix is ​​generated and output using the Softmax activation function.

10. The correction method according to claim 1, characterized in that, Before performing standardization, principal component analysis, and feature weighting on the features extracted from the multi-dimensional sources, the method further includes: Multiple preset methods are used to preprocess data from multiple dimensions to obtain preprocessed data; among them, the multiple preset methods include: sliding window partitioning method, cross sampling rate interpolation method, and dynamic time warping method.

11. A calibration device for helium concentration data based on mass spectrometry detection, characterized in that, The device includes: The first processing module is used to perform standardization, principal component analysis, and feature weighting on the features extracted from the multi-dimensional data in sequence to obtain a weighted feature vector under a first preset value dimension. The multi-dimensional data includes: mass spectrometry signal dimension, environmental parameter dimension, and magnetic field data dimension. The extraction methods used for multi-dimensional extraction include: data sampling extraction method and statistical analysis extraction method. The second processing module is used to input the original mass spectrometry signal and the weighted feature vector under the first preset dimension into a dual-channel convolutional neural network for processing, and generate a fused feature map. The dual-channel convolutional neural network includes a first channel and a second channel, and the fused feature map integrates the features of the original mass spectrometry signal and the environmental features. The output module is used to output the fused feature map; The acquisition module is used to: acquire multiple distortion coefficients of the nonlinear distortion model before calculating the distortion of the original mass spectrometry signal using the nonlinear distortion model. These multiple distortion coefficients include: a first distortion coefficient related to time decay, a second distortion coefficient related to the intensity of periodic distortion caused by vibration, and a third distortion coefficient related to changes in vacuum level. This nonlinear distortion model can quantify the influence of instrument status on the mass spectrometry signal. In the formula: This represents the distorted mass spectrum signal; This represents the distortion coefficient related to time decay; Represents a constant; Indicates the vibration frequency; The distortion coefficient represents the correlation between the intensity of periodic distortion caused by vibration. This represents the distortion coefficient related to changes in vacuum level; The gradient representing the degree of vacuum; The real-time update module is used to: update the first distortion coefficient, the second distortion coefficient, and the third distortion coefficient in real time using the stochastic gradient descent method, so as to capture changes in the instrument status in real time; The calculation module is used to calculate the distortion of the original mass spectrometry signal through a nonlinear distortion model to obtain the distorted mass spectrometry signal. The nonlinear distortion model can quantify the influence of different instrument states on the mass spectrometry signal. The compensation module is used to compensate the feature map based on the distorted mass spectrum signal to obtain the compensated mass spectrum signal. The dynamic inversion module is used to dynamically invert the current helium concentration value based on the compensated mass spectrometry signal and using an inverse problem-solving framework to obtain the corrected helium concentration value.

12. A computer-readable storage medium, characterized in that, It contains a computer program that, when executed in a computer, causes the computer to perform the method described in any one of claims 1 to 10.

13. An electronic device, characterized in that, The method includes a memory and a processor, wherein the memory stores executable code, and the processor executes the executable code to implement the method according to any one of claims 1 to 10.

Images