Method and system for determining coal element content and computing equipment

Abstract

The invention discloses a method and a system for determining the content of coal elements and computing equipment, relates to the technical field of radioactive safety detection, and solves the technical problems of complicated standardized correction and low precision of coal element analysis and prediction results in existing coal element analysis. The method comprises the following steps: acquiring actually measured energy spectrum data of an initial coal sample through a gamma detector; performing data enhancement on the actually measured energy spectrum data to obtain a plurality of experimental coal samples; constructing an element recognition model through a self-attention mechanism and a multi-modal attention mechanism; inputting the plurality of experimental coal samples into an element recognition model to obtain types of elements in the experimental coal samples and predicted content distribution of the corresponding types; if it is evaluated that the element recognition model is in a preset convergence state through the rating indexes, marking that recognition of the elements in the experimental coal sample is completed; the prediction accuracy of the element content in the coal is improved.

Description

Technical Field

[0001] This invention belongs to the field of radioactive safety detection and relates to nuclide energy spectrum prediction technology, specifically a method, system and computing device for determining the elemental content of coal. Background Technology

[0002] Existing methods for coal elemental analysis mainly include chemical analysis, X-ray fluorescence spectrometry (XRF), and laser-induced breakdown spectroscopy. While existing chemical analysis methods, such as gravimetric and titration methods, offer high precision, their experimental procedures are cumbersome, analysis times are long, and complex pretreatment steps are required, making them unsuitable for real-time online measurement. XRF spectroscopy can rapidly detect some elements in coal, but its detection capability for light elements such as C, H, and O is weak, and it is also significantly affected by matrix effects. LIBS technology suffers from significant influence on measurement repeatability and stability by plasma state, and still exhibits certain errors when coal composition is complex and matrix effects are significant.

[0003] Existing technologies employ complex standardization corrections in coal elemental analysis methods, resulting in low accuracy of coal elemental analysis and identification results.

[0004] This invention provides a method for determining the elemental content of coal to solve the above-mentioned technical problems. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a method, system and computing device for determining the elemental content of coal, which solves the technical problem that the coal elemental analysis and prediction results are low due to the complex standardization corrections performed by the coal elemental analysis method in the prior art.

[0006] To achieve the above objectives, a first aspect of the present invention provides a method for determining the elemental content of coal, comprising:

[0007] Several compounds were combined in different predetermined proportions to obtain several initial coal samples;

[0008] Measured energy spectrum data of the initial coal sample were obtained using a gamma detector; data enhancement was performed on the measured energy spectrum data to obtain several experimental coal samples.

[0009] An element recognition model is constructed using self-attention and multimodal attention mechanisms. Several experimental coal samples are input into the element recognition model to obtain the types of elements in the experimental coal samples and the predicted content distribution of the corresponding types.

[0010] Evaluation indicators are constructed by identifying the types of elements in experimental coal samples and the predicted content distribution of the corresponding types. If the element identification model is in a preset convergence state as assessed by the rating indicators, it is marked as having completed the identification of elements in the experimental coal samples. The preset convergence state is defined as the evaluation indicator being less than a preset threshold.

[0011] Preferably, the step of acquiring the measured energy spectrum data of the initial coal sample using a gamma detector includes:

[0012] Several different types of detectors were retrieved, and initial energy spectrum data of the initial coal sample were collected simultaneously under the same test environment and within the same preset period. The initial energy spectrum data were then subtracted to obtain several measured energy spectrum data.

[0013] This invention constructs coal samples by precisely mixing known compounds in specific proportions, thereby ensuring that the types and contents of target elements in the samples are controllable.

[0014] Preferably, the step of subtracting the initial energy spectrum data to obtain several measured energy spectrum data includes:

[0015] Retrieve initial energy spectrum data, and use the i-th initial energy spectrum data x i Input to the detector; retrieve the detector's background count rate;

[0016] Will be through formula The initial sample count rate R was calculated. i In the formula, C(x) i ) represents the count of the i-th measured energy spectrum data, and T is the counting time;

[0017] The background count rate is removed from the initial sample count rate to obtain the measured sample count rate. The energy spectrum data corresponding to the measured sample count rate is recorded as the measured energy spectrum data and marked as the completion of the subtraction operation.

[0018] It should be noted that when the detector has no sample, the count rate calculated by the detector is the background count rate.

[0019] Preferably, the step of data augmentation of the measured energy spectrum data to obtain several experimental coal samples includes:

[0020] Retrieve the j-th measured energy spectrum data S j , will S j With S j+1 Mix according to the ratio λ:(1-λ), S j With S j+1 The corresponding element content labels Y1 and Y2 are also calculated according to the ratio of λ:(1-λ), and several new sample labels Y are generated. new For: Y new=Y1λ+(1-λ)Y2; where λ takes values ​​in the interval (0, 1);

[0021] Several energy spectrum data with different counts were obtained according to the new sample labels. The energy spectrum data were divided into a test set and a verification set according to a preset ratio. The energy spectrum data of the test set were synthesized to obtain new energy spectrum data. Several corresponding experimental coal samples were obtained based on the new energy spectrum data.

[0022] The preparation of samples in this invention requires a lot of consumables and effort. Therefore, data augmentation methods are used to expand the samples with different contents to make up for the lack of samples. In order to ensure the consistency of data labels, the labels of the synthetic samples, i.e. the element contents, are also adjusted according to the corresponding proportions. The diversity of data is enhanced by randomly taking values ​​of λ in the interval [0, 1].

[0023] Preferably, the step of inputting several experimental coal samples into the elemental identification model to obtain the types of elements in the experimental coal samples and the predicted content distribution of the corresponding types includes:

[0024] New energy spectrum data corresponding to several experimental coal samples are retrieved, and the new energy spectrum data is preprocessed to obtain the energy spectrum after fusion of different detectors; the Transformer backbone network is called, the fused energy spectrum is input into the Transformer backbone network, and the high-dimensional features of the fused energy spectrum are extracted based on the self-attention mechanism.

[0025] The cross-correlation information of data from different types of detectors is calculated using a multimodal attention mechanism;

[0026] The high-dimensional features and mutual information are mapped to the fully connected layer of the Transformer, and the output of the fully connected layer is the type of element and the predicted content distribution of the corresponding type.

[0027] This invention designs a multimodal attention mechanism that calculates the cross-correlation between the data from the two detectors, enabling the model to fully utilize the sensitivity of the NaI detector to low-energy gamma rays and the detection capability of the BGO detector to high-energy gamma rays. This allows for the rapid fusion of detection data from the NaI and BGO detectors and improves the accuracy of element identification.

[0028] Preferably, the step of preprocessing the new energy spectrum data to obtain the fused energy spectrum from different detectors includes:

[0029] Let the energy spectrum of detector A be S. NaI (E), the energy spectrum of detector B is S BGO (E);

[0030] Through linear transformation E cal=a·channel+b S is calculated as a·channel+b NaI (E) and S BGO (E) is the same mapped energy; where channel is S NaI (E) and S BGO (E) is the track address, where a and b are scale coefficients;

[0031] Through formula S fused(E) =ω(E)·S NaI (E)+[1-ω(E)]·S BGO (E) The energy spectrum after fusion of different detectors is calculated; where ω(E) is the preset weighting function related to energy.

[0032] It should be noted that the calibration coefficients are obtained through a standard radioactive source (such as...). 137 Cs、 60 Co) calibration is obtained; the preset weight function ω(E) is automatically learned through a neural network. The low-energy region (energy < 500 keV) tends to be NaI data, and the high-energy region (energy > 1 MeV) tends to be BGO data.

[0033] This invention addresses the need for energy scale alignment of the energy spectra of NaI and BGO detectors due to their different energy response characteristics; and assigns dynamic weights to the energy spectrum data of the two detectors to optimize signal contribution in different energy ranges.

[0034] Preferably, the calculation of the cross-correlation information of different types of detector data through the multimodal attention mechanism includes:

[0035] A multimodal attention mechanism is invoked to learn the energy spectrum embedding features of detectors A and B; where the energy spectrum embedding feature of detector A is denoted as X. NaI The energy spectrum embedding feature of detector B is denoted as X. BGO ;

[0036] The cross-modal attention mechanism is invoked to calculate the mutual information Attention(Q, K, V) between detectors A and B:

[0037] Where Q = X NaI ·W Q K = X BGO ·W K V = X BGO ·W V WQ, WK, and WV are learnable parameters.

[0038] This invention combines the advantages of spatial attention mechanism (SAM) to achieve the effect of joint optimization of global and local features in Transformer.

[0039] Preferably, the construction of evaluation indicators based on the types of elements in experimental coal samples and the predicted content distribution of the corresponding types includes:

[0040] The predicted content distribution of the nth element is denoted as C. n Through formula The mean absolute error is calculated; where X n Let n be the actual content of the nth element; n = 1, 2, 3, ..., k;

[0041] Mean absolute error is used as the evaluation metric for the element recognition model.

[0042] It should be noted that the closer the evaluation index is to 0, the better the element recognition model's element recognition performance.

[0043] This invention achieves accurate regression of element content by using mean absolute error as the model evaluation metric; and improves the training stability of the model by using the Adam optimizer in combination with a learning rate scheduling strategy, enabling the model to converge efficiently.

[0044] To achieve the above objectives, a second aspect of the present invention provides a system for determining the elemental content of coal, comprising: a data acquisition module, a model building module, and an analysis and evaluation module;

[0045] Data acquisition module: Several compounds are combined in different preset proportions to obtain several initial coal samples; measured energy spectrum data of the initial coal samples are acquired using a gamma detector; and data enhancement is performed on the measured energy spectrum data to obtain several experimental coal samples.

[0046] Model building module: Constructs an element recognition model through self-attention and multimodal attention mechanisms; inputs several experimental coal samples into the element recognition model to obtain the types of elements in the experimental coal samples and the predicted content distribution of the corresponding types;

[0047] Analysis and evaluation module: Evaluation indicators are constructed by the types of elements in the experimental coal sample and the predicted content distribution of the corresponding types. If the element identification model is in a preset convergence state through the evaluation indicators, it is marked as having completed the identification of elements in the experimental coal sample. The preset convergence state is when the evaluation indicator is less than a preset threshold.

[0048] An electronic device for determining the elemental content of coal, characterized in that it comprises: a memory and a processor, the memory storing executable instructions of the processor; wherein the processor is configured to execute a method for determining the elemental content of coal as described in the first aspect by executing the executable instructions.

[0049] Compared with the prior art, the beneficial effects of the present invention are:

[0050] 1. This invention constructs coal samples by precisely mixing known compounds to ensure that the types and contents of target elements in the samples are controllable. In real-world environments, it is difficult to directly obtain coal samples with all possible contents, and relying solely on experimentally collected data may not be sufficient to train a highly robust neural network model. Therefore, this invention employs a data augmentation strategy to expand the training dataset. During the data augmentation process, the collected 30 energy spectrum data are synthesized according to the ratio of different elements, that is, new energy spectrum data are generated based on the energy spectrum data of two or more experimental samples. At the same time, to ensure the consistency of data labels, the labels of the synthesized samples, i.e., the element contents, are also adjusted according to the corresponding ratios.

[0051] 2. This invention constructs coal samples by precisely mixing known compounds to ensure that the types and contents of target elements in the samples are controllable. The compounds used in this invention are 10 kinds: C, CH2, C3H6N6, Al2O3, SiO2, CaCO3, Fe2O3, TiO2, K2CO3, Na2CO3, and NH4SCN. Different compounds are mixed in specific mass ratios to construct coal samples with different types and concentrations of elements. By controlling the ratio, this invention obtains energy dispersive spectroscopy data at different concentration levels and ensures the representativeness and scalability of the data. In addition, this invention more closely approximates the compositional distribution of real coal by randomly adjusting the ratio of each element within a certain range. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a schematic diagram of the process for determining element content in this invention;

[0054] Figure 2 This is a schematic diagram illustrating the specific process of data acquisition in this invention;

[0055] Figure 3 This is a schematic diagram of the prediction and recognition process of the element recognition model of the present invention;

[0056] Figure 4 This is a schematic diagram illustrating the prediction error of the element recognition model of the present invention;

[0057] Figure 5 This is a schematic diagram illustrating the module relationships included in this invention. Detailed Implementation

[0058] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] Please see Figure 1 The first aspect of the present invention provides a method for determining the elemental content of coal, comprising:

[0060] Several compounds were combined in different predetermined proportions to obtain several initial coal samples;

[0061] Measured energy spectrum data of the initial coal sample were obtained using a gamma detector; data enhancement was performed on the measured energy spectrum data to obtain several experimental coal samples.

[0062] An element recognition model is constructed using self-attention and multimodal attention mechanisms. Several experimental coal samples are input into the element recognition model to obtain the types of elements in the experimental coal samples and the predicted content distribution of the corresponding types.

[0063] Evaluation indicators are constructed by identifying the types of elements in experimental coal samples and the predicted content distribution of the corresponding types. If the element identification model is in a preset convergence state as assessed by the rating indicators, it is marked as having completed the identification of elements in the experimental coal samples. The preset convergence state is defined as the evaluation indicator being less than a preset threshold.

[0064] Please see Figure 2 The specific data acquisition process is as follows: Several compounds are combined in different preset proportions to obtain several initial coal samples; several different types of detectors are used to simultaneously collect the initial energy spectrum data of the initial coal samples under the same test environment and within the same preset period; the i-th initial energy spectrum data x... i Input to the detector; retrieve the detector's background count rate, which will be calculated using the formula. The initial sample count rate R was calculated. i In the formula, C(x) i ) represents the count of the i-th measured energy spectrum data, and T is the counting time;

[0065] The background count rate is removed from the initial sample count rate to obtain the measured sample count rate. The energy spectrum data corresponding to the measured sample count rate is recorded as the measured energy spectrum data and marked as the completion of the subtraction operation.

[0066] Retrieve the j-th measured energy spectrum data S j , will Sj With S j+1 Mix according to the ratio λ:(1-λ), S j With S j+1 The corresponding element content labels Y1 and Y2 are also calculated according to the ratio of λ:(1-λ), and several new sample labels Y are generated. new For: Y new =Y1λ+(1-λ)Y2; where λ takes values ​​in the interval (0, 1);

[0067] Several energy spectrum data with different counts were obtained according to the new sample labels. The energy spectrum data were divided into a test set and a verification set according to a preset ratio. The energy spectrum data of the test set were synthesized to obtain new energy spectrum data. Several corresponding experimental coal samples were obtained based on the new energy spectrum data.

[0068] For example, sample preparation: Coal samples are prepared based on the elemental content range of real-world coal. Coal is composed of 12 elements: C, H, O, Si, Al, S, N, Fe, Ca, Na, K, and Ti. Since the elemental composition of actual coal samples is complex and it is difficult to comprehensively cover all possible content distributions, coal samples are constructed by precisely mixing known compounds to ensure that the types and contents of target elements in the samples are controllable. This paper uses 10 compounds: C, CH2, C3H6N6, Al2O3, SiO2, CaCO3, Fe2O3, TiO2, K2CO3, Na2CO3, and NH4SCN. Different compounds are mixed in specific mass ratios to construct coal samples with different elemental types and concentrations. By controlling the ratios, we can obtain energy dispersive spectroscopy (EDS) data at different concentration levels, ensuring the representativeness and scalability of the data. In addition, in order to simulate a compositional distribution closer to that of real coal, we randomly adjusted the ratio of each element within a certain range to cover a wider range of element concentrations. A total of 40 coal samples with different concentrations were prepared, of which 30 were used for training and 10 were used for prediction.

[0069] Acquiring measured energy spectrum data: NaI detector and BGO detector were used. NaI detector was digitally multi-channel, and BGO detector had 4096 channel addresses. Both detectors were used to collect data simultaneously under the same test environment, measuring the background and different types of coal samples. Each collection time was 5 minutes, and the collected experimental energy spectrum was subtracted according to the background count rate.

[0070] Since it is difficult to directly obtain coal samples with all possible content in real-world environments, relying solely on experimentally collected data may be insufficient for training a highly robust neural network model. Therefore, we further employ data augmentation strategies to expand the training dataset. During data augmentation, we synthesize the spectral data from 30 collected energy spectra according to the ratios of different elements; that is, we generate new energy spectra based on the energy spectra of two or more experimental samples. Simultaneously, to ensure the consistency of data labels, the labels of the synthesized samples, i.e., elemental content, are adjusted according to the corresponding proportions. For example, for the measured sample energy spectra, they are mixed according to a certain ratio, and their corresponding elemental content labels are calculated according to the same ratio, generating a new sample label: where the value is random between [0, 1], thereby enhancing the diversity of the data.

[0071] Please see Figure 3-4 The specific process for data acquisition is as follows: retrieve new energy spectrum data corresponding to several experimental coal samples, preprocess the new energy spectrum data to obtain the energy spectrum after fusion of different detectors; call the Transformer backbone network, input the fused energy spectrum into the Transformer backbone network, and extract the high-dimensional features of the fused energy spectrum based on the self-attention mechanism.

[0072] The cross-correlation information of data from different types of detectors is calculated using a multimodal attention mechanism;

[0073] High-dimensional features and mutual information are mapped to the fully connected layer of the Transformer, and the fully connected layer outputs the types of elements and the predicted content distribution of the corresponding types.

[0074] Let the energy spectrum of detector A be S. NaI (E), the energy spectrum of detector B is S BGO (E);

[0075] Through linear transformation E cal =a·channel+b S is calculated as a·channel+b NaI (E) and S BGO (E) is the same mapped energy; where channel is S NaI (E) and S BGO (E) is the track address, where a and b are scale coefficients;

[0076] Through formula S fused(E) =ω(E)·S NaI (E)+[1-ω(E)]·S BGO(E) The energy spectrum of the fused detectors is calculated; where ω(E) is the preset weight function related to energy; the multimodal attention mechanism is invoked to learn the energy spectrum embedding features of detectors A and B; where the energy spectrum embedding feature of detector A is denoted as X. NaI The energy spectrum embedding feature of detector B is denoted as X. BGO ;

[0077] The cross-modal attention mechanism is invoked to calculate the mutual information Attention(Q, K, V) between detectors A and B:

[0078] Where Q = X NaI ·W Q K = X BGO ·W K V = X BGO ·W V WQ, WK, and WV are learnable parameters.

[0079] For example, in this study, the NaI detector and the BGO detector provided gamma spectral data for two modes; the NaI detector has high sensitivity to low-energy gamma rays, while the BGO detector performs better in the higher energy range and has better detection capabilities; by inputting the spectral information from these two data sources into the neural network, the model can fully learn the characteristics of each detector in different energy regions, thereby enhancing the ability to identify elements in coal samples.

[0080] A multimodal Transformer-based γ-ray spectrum identification network model is constructed for nuclide training and prediction. The proposed multimodal Transformer-based γ-ray spectrum identification method is trained using a training set. Since the spectral data is a channel-ordered sequence, the spectral input is first embedded into a high-dimensional feature space. Under the self-attention mechanism of the Transformer, the model can capture the long-range dependencies between different energy channels and identify the characteristic peaks of different elements in the spectrum. To effectively fuse data from NaI and BGO detectors, a multimodal attention mechanism is designed. This mechanism calculates the cross-correlation between the two detector data, enabling the model to fully utilize the sensitivity of the NaI detector to low-energy γ-rays and the detection capability of the BGO detector to high-energy γ-rays, thereby improving the accuracy of element identification.

[0081] exist Figure 4In the diagram, the horizontal axis represents the actual element content, and the vertical axis represents the predicted element content. The blue dot on the horizontal axis represents the actual value, and on the vertical axis, it represents the predicted value. If the blue dot is exactly on the red line, it means the predicted value and the actual value are equal. If the blue dot is above and to the left of the red line, it means the actual value is less than the predicted value; if the blue dot is below and to the right of the red line, it means the actual value is greater than the predicted value. Figure 4 It can be used to measure the error between the actual value and the predicted value;

[0082] At the model output, a fully connected layer is used to map the high-dimensional features extracted by the Transformer to the probability distribution of element types. Simultaneously, to predict the specific content of each element in the coal sample, the Mean Absolute Error (MAE) is used as the model evaluation metric, thereby achieving accurate regression of element content. To improve the model's training stability, the Adam optimizer is used in conjunction with a learning rate scheduling strategy, enabling the model to converge efficiently. Furthermore, data augmentation techniques are introduced to enhance the model's generalization ability.

[0083] Validation set validation: The network model is validated using a validation set. The learning rate, number of iterations, and regularization parameters are adjusted, and the optimal network model is obtained based on the evaluation metric, with the mean absolute error (MAE) used as the model evaluation indicator. The closer these values ​​are to 0, the better the recognition performance. When the weights are saved after 500 training iterations, the average MAE value is below 0.73%, indicating that the element recognition model described in this invention has excellent performance in identifying coal element content.

[0084] Please see Figure 5 A second aspect of the present invention provides a system for determining the elemental content of coal, comprising: a data acquisition module, a model building module, and an analysis and prediction module;

[0085] Data acquisition module: Combines several compounds in different preset proportions to obtain several initial coal samples;

[0086] Measured energy spectrum data of the initial coal sample were obtained using a gamma detector; data enhancement was performed on the measured energy spectrum data to obtain several experimental coal samples.

[0087] Model building module: Builds an element recognition model for feature recognition;

[0088] Analysis and prediction module: By adjusting parameters and based on evaluation indicators, the elemental content of coal is identified using the elemental identification model.

[0089] Some of the data in the above formula are calculated by removing dimensions and taking their numerical values. The formula is the closest to the real situation obtained by software simulation of a large amount of collected data. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

[0090] The working principle of this invention is as follows: Several initial coal samples are obtained by combining several compounds in different preset proportions; measured energy spectrum data of the initial coal samples are obtained through a gamma detector; the measured energy spectrum data are augmented to obtain several experimental coal samples; an element identification model is constructed through a self-attention mechanism and a multimodal attention mechanism; the several experimental coal samples are input into the element identification model to obtain the types of elements in the experimental coal samples and the predicted content distribution of the corresponding types; an evaluation index is constructed based on the types of elements in the experimental coal samples and the predicted content distribution of the corresponding types. If the element identification model is in a preset convergence state as evaluated by the rating index, it is marked as having completed the identification of elements in the experimental coal samples.

[0091] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for determining the elemental content of coal, characterized in that, include: Several compounds were combined in different predetermined proportions to obtain several initial coal samples; Measured energy spectrum data of the initial coal sample were obtained using a gamma detector; Several experimental coal samples were obtained by data augmentation of the measured energy spectrum data. An element recognition model is constructed using self-attention and multimodal attention mechanisms. Several experimental coal samples are input into the element recognition model to obtain the types of elements in the experimental coal samples and the predicted content distribution of the corresponding types. Evaluation indicators are constructed by identifying the types of elements in experimental coal samples and the predicted content distribution of the corresponding types. If the element identification model is in a preset convergence state as assessed by the rating indicators, it is marked as having completed the identification of elements in the experimental coal samples. The preset convergence state is defined as the evaluation indicator being less than a preset threshold.

2. The method for determining the elemental content of coal according to claim 1, characterized in that, The acquisition of measured energy spectrum data of the initial coal sample via a gamma detector includes: Several different types of detectors were retrieved, and initial energy spectrum data of the initial coal sample were collected simultaneously under the same test environment and within the same preset period. The initial energy spectrum data were then subtracted to obtain several measured energy spectrum data.

3. The method for determining the elemental content of coal according to claim 2, characterized in that, The process of subtracting the initial energy spectrum data to obtain several measured energy spectrum data includes: Retrieve initial energy spectrum data, and use the i-th initial energy spectrum data x i Input to the detector; retrieve the detector's background count rate; Will be through formula The initial sample count rate R was calculated. i In the formula, C(x) i ) represents the count of the i-th measured energy spectrum data, and T is the counting time; The background count rate is removed from the initial sample count rate to obtain the measured sample count rate. The energy spectrum data corresponding to the measured sample count rate is recorded as the measured energy spectrum data and marked as the completion of the subtraction operation.

4. The method for determining the elemental content of coal according to claim 1, characterized in that, The process involves data augmentation of measured energy spectrum data to obtain several experimental coal samples, including: Retrieve the j-th measured energy spectrum data S j , will S j With S j+1 Mix according to the ratio λ:(1-λ), S j With S j+1 The corresponding element content labels Y1 and Y2 are also calculated according to the ratio of λ:(1-λ), and several new sample labels Y are generated. new For: Y new =Y1λ+(1-λ)Y2; where λ takes values ​​in the interval (0, 1); Several energy spectrum data with different counts were obtained according to the new sample labels. The energy spectrum data were divided into a test set and a verification set according to a preset ratio. The energy spectrum data of the test set were synthesized to obtain new energy spectrum data. Several corresponding experimental coal samples were obtained based on the new energy spectrum data.

5. The method for determining the elemental content of coal according to claim 4, characterized in that, The step of inputting several experimental coal samples into an elemental identification model to obtain the types of elements in the experimental coal samples and the predicted content distribution of the corresponding types includes: New energy spectrum data corresponding to several experimental coal samples are retrieved, and the new energy spectrum data is preprocessed to obtain the energy spectrum after fusion of different detectors; the Transformer backbone network is called, the fused energy spectrum is input into the Transformer backbone network, and the high-dimensional features of the fused energy spectrum are extracted based on the self-attention mechanism. The cross-correlation information of data from different types of detectors is calculated using a multimodal attention mechanism; The high-dimensional features and mutual information are mapped to the fully connected layer of the Transformer, and the output of the fully connected layer is the type of element and the predicted content distribution of the corresponding type.

6. A method for determining the elemental content of coal according to claim 5, characterized in that, The step of preprocessing the new energy spectrum data to obtain the fused energy spectrum from different detectors includes: Let the energy spectrum of detector A be S. NaI (E), the energy spectrum of detector B is S BGO (E); Through linear transformation E cal =a·channel+b to calculate S NaI (E) and S BGO (E) is the same mapped energy; where channel is S NaI (E) and S BGO (E) is the track address, where a and b are scale coefficients; Through formula S fused(E) =ω(E)·S NaI (E)+[1-ω(E)]·S BGO (E) The energy spectrum after fusion of different detectors is calculated; where ω(E) is the preset weighting function related to energy.

7. A method for determining the elemental content of coal according to claim 5, characterized in that, The calculation of cross-correlation information of different types of detector data through a multimodal attention mechanism includes: A multimodal attention mechanism is invoked to learn the energy spectrum embedding features of detectors A and B; where the energy spectrum embedding feature of detector A is denoted as X. NaI The energy spectrum embedding feature of detector B is denoted as X. BGO ; The cross-modal attention mechanism is invoked to calculate the mutual information Attention(Q, K, V) between detectors A and B: Where Q = X NaI ·W Q K = X BGO ·W K V = X BGO ·W V WQ, WK, and WV are learnable parameters.

8. A method for determining the elemental content of coal according to claim 1, characterized in that, The evaluation index is constructed by analyzing the types of elements in experimental coal samples and the predicted content distribution of those elements, including: The predicted content distribution of the nth element is denoted as C. n Through formula The mean absolute error is calculated; where X n Let n be the actual content of the nth element; n = 1, 2, 3, ..., k; Mean absolute error is used as the evaluation metric for the element recognition model.

9. A system for determining the elemental content of coal, applied to the method for determining the elemental content of coal as described in any one of claims 1-8, characterized in that, include: Data acquisition module, model building module, and analysis and evaluation module; Data acquisition module: Combines several compounds in different preset proportions to obtain several initial coal samples; Measured energy spectrum data of the initial coal sample were obtained using a gamma detector; data augmentation was then performed on the measured energy spectrum data to obtain several experimental coal samples. Model building module: Constructs an element recognition model through self-attention and multimodal attention mechanisms; inputs several experimental coal samples into the element recognition model to obtain the types of elements in the experimental coal samples and the predicted content distribution of the corresponding types; Analysis and evaluation module: Evaluation indicators are constructed by the types of elements in the experimental coal sample and the predicted content distribution of the corresponding types. If the element identification model is in a preset convergence state through the evaluation indicators, it is marked as having completed the identification of elements in the experimental coal sample. The preset convergence state is when the evaluation indicator is less than a preset threshold.

10. A system computing device for determining the elemental content of coal, comprising: A memory and a processor, wherein the memory stores executable instructions of the processor; wherein the processor is configured to perform a method for determining the elemental content of coal according to any one of claims 1-8 by executing the executable instructions.

Images