Multi-channel coal quality on-line detection system and method based on signal enhancement

Abstract

The application discloses a multi-channel coal quality online detection system and method based on signal enhancement, and belongs to the technical field of coal quality detection, and solves the problem that the wavelet threshold denoising of the existing method adopts a fixed threshold rule and cannot adapt to the noise characteristic change of different temperature intervals, thereby affecting the accuracy and stability of the detection result, the method comprises the following steps: performing signal enhancement preprocessing on a real-time information set, performing initial feature selection on the signal enhancement set based on an index identification model, and performing index collaborative correction based on the drift risk quantity of the initial index set and the initial index set; in the application, the real-time acquisition of the multi-channel spectral information of the coal quality and the associated environmental information is subjected to signal enhancement preprocessing, the interference of the environmental factors on the spectral signal is considered during the signal enhancement preprocessing, the on-site three-dimensional temperature field is coupled with the adaptive time period, the interference of the environmental factors on the signal can be effectively compensated, and therefore the stability and accuracy of the signal are improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of coal quality detection, and particularly relates to a multi-channel coal quality online detection system and method based on signal enhancement. BACKGROUND

[0002] As a key link of quality control in the coal industry, coal quality detection technology has experienced a development process from traditional laboratory analysis to on-site rapid detection. Traditional coal quality analysis methods mainly rely on laboratory chemical analysis, such as industrial analysis and elemental analysis. Although these methods have high accuracy, they have significant shortcomings such as complex sample preparation, long analysis period, and inability to meet the real-time requirements of industrial production. With the development of spectral analysis technology, near-infrared spectroscopy and laser-induced breakdown spectroscopy have been introduced into the field of coal quality detection, realizing non-contact and rapid detection.

[0003] Existing coal quality online detection systems face many technical bottlenecks in signal acquisition, especially in remote detection scenarios. The optical signal is easily disturbed by the environment during transmission, resulting in low signal-to-noise ratio and poor stability of the collected spectral data. Moreover, traditional optical acquisition systems often use a separate design, with spatial deviations between the excitation light source and the signal collection light path, leading to unstable light radiation energy reception and seriously affecting the repeatability and accuracy of the detection results.

[0004] Chinese patent CN119290810B discloses a self-adaptive calibration method and device for coal quality detection environmental influence correction. The self-adaptive calibration method includes spectral preprocessing and LIBS spectral feature peak intensity correction. The device includes a control host, a spectrometer, a laser generator, and a rotary sample table. However, when the existing method uses wavelet transform to denoise the spectral signal of the sample and reduces the interference of adverse factors on the spectral signal, the fixed threshold rule used in wavelet threshold denoising cannot adapt to the changes in noise characteristics in different temperature intervals, and the influence of related environmental information (such as temperature) on the spectral signal is not considered, leading to drift and distortion of the measured spectral signal, thereby affecting the accuracy and stability of the detection results. To address the above problems, we propose a multi-channel coal quality online detection system and method based on signal enhancement. SUMMARY

[0005] The present application aims to address the shortcomings of the prior art by providing a multi-channel coal quality online detection system and method based on signal enhancement, which solves the problem of the existing method using wavelet transform to denoise the spectral signal of the sample and reducing the interference of adverse factors on the spectral signal, and the fixed threshold rule used in wavelet threshold denoising cannot adapt to the changes in noise characteristics in different temperature intervals, and the influence of related environmental information on the spectral signal is not considered, leading to drift and distortion of the measured spectral signal, thereby affecting the accuracy and stability of the detection results.

[0006] The application is implemented based on a signal-enhanced multi-channel coal quality online detection method, which comprises the following steps:

[0007] Based on the real-time acquisition of coal quality multi-channel spectral information and associated environmental information by a coal quality detection device, the coal quality multi-channel spectral information and the associated environmental information are combined into a real-time information set, and the real-time information set is subjected to signal enhancement preprocessing to output a signal enhancement set.

[0008] An index recognition model based on a support vector machine is pre-constructed, and the signal enhancement set is iteratively trained until convergence. The signal enhancement set is loaded, and the signal enhancement set is subjected to initial feature selection based on the index recognition model. The feature importance is determined by improving the random forest algorithm, a multi-channel weighted feature map combined with the feature importance is constructed, and the index recognition model outputs an initial index set combined with the multi-channel weighted feature map.

[0009] The initial index set is obtained, the detection confidence of the initial index set is determined based on a multi-head attention mechanism, the drift risk of the initial index set is dynamically evaluated, the index is cooperatively corrected based on the drift risk of the initial index set and the initial index set, and the coal quality index detection result after cooperative correction is output.

[0010] Preferably, the signal enhancement preprocessing method for the real-time information set comprises:

[0011] The real-time information set is loaded, and the real-time information set is subjected to abnormal value cleaning and elimination based on a boundary hybrid sampling strategy to obtain the real-time information set after abnormal value cleaning and elimination, and the real-time information set is subjected to normalization processing.

[0012] The real-time information set after normalization processing is obtained, the signal band of the real-time information set is identified based on a spectral analysis method, the peak characteristics of the signal band are determined, the coal quality multi-channel spectral information in the real-time information set is adaptively time period cut based on the peak characteristics of the signal band, and the associated environmental information and a plurality of adaptive time periods are aligned by using a timestamp alignment method to obtain an adaptive time period set.

[0013] The adaptive time period set is dynamically filtered based on a BEADS algorithm combined with an S-G filtering method to output the adaptive time period set after dynamic filtering.

[0014] The adaptive time period set is loaded, a field three-dimensional temperature field is constructed based on the adaptive time period set, the field three-dimensional temperature field is coupled with the adaptive time period to obtain derived enhanced characteristics, and the derived enhanced characteristics, the field three-dimensional temperature field and the adaptive time period are combined to obtain the signal enhancement set.

[0015] Preferably, the abnormal value cleaning and elimination method for the real-time information set based on the boundary hybrid sampling strategy comprises:

[0016] An acquisition real-time information set, and a boundary point detection algorithm is used to determine the coefficient of variation of data points in the real-time information set;

[0017] The coefficient of variation of the data points is calculated by the following formula:

[0018]

[0019] The coefficient of variation of the data points is calculated by the following formula: The coefficient of variation of the data points is calculated by the following formula: The density of the data points, the number of adjacent data points in the real-time information set with a distance of from the distance data point;

[0020] A variation threshold of the real-time information set is set based on the coefficient of variation of the data points, and data points with a coefficient of variation greater than the variation threshold in the real-time information set are divided into boundary data points, and the remaining data points are divided into center data points;

[0021] The boundary data points are loaded, a preset oversampling rate is set based on the weight of the boundary data points, the synthetic minority over-sampling technique is used to copy the boundary data points, and new boundary data points after copying are obtained;

[0022] The center data points are loaded, an undersampling rate is set based on the density of the center data points, the center data points are traversed, the center data points are pruned and cleaned based on the Tomek-Links method combined with the undersampling rate, and new center data after reconstruction is obtained;

[0023] The new boundary data points after copying and the new center data after reconstruction are obtained, and the real-time information set after removal of outliers is integrated based on the new boundary data points after copying and the new center data after reconstruction;

[0024] The real-time information set after removal of outliers is loaded, and the real-time information set is normalized.

[0025] Preferably, the method for dynamically filtering the adaptive time period set based on the BEADS algorithm combined with the S-G filtering method comprises:

[0026] The adaptive time period set is loaded, the parameters of the BEADS algorithm are initialized, and the number of iterations is set;

[0027] The difference operator of the multi-channel spectral information of the coal quality in the adaptive time period is calculated based on the BEADS algorithm, and the spectral information estimate is updated, and the baseline component and the noise component of the multi-channel spectral information of the coal quality in the adaptive time period are determined after a preset number of iterations;

[0028] The estimated baseline component and the noise component are subtracted from the original signal to obtain the adaptive time period after denoising;

[0029] The window size and polynomial order of the S-G filter corresponding to different adaptive time periods are determined based on the smoothness of the adaptive time period, and the S-G filter method realizes smoothing of the multi-channel spectral information of the coal quality in the adaptive time period through local polynomial fitting, and outputs a set of dynamically filtered adaptive time periods.

[0030] Preferably, the method for coupling the on-site three-dimensional temperature field with the adaptive time period comprises:

[0031] The adaptive time period set is loaded, the associated environmental information in the adaptive time period set is extracted, the coal field grid model is established based on the associated environmental information, the coal field grid model is divided, and the internal temperature field space and the boundary temperature field space of the coal field grid model sub-grid are obtained;

[0032] The internal temperature field space and the boundary temperature field space of the coal field grid model sub-grid are obtained, the dynamically filtered associated temperature parameters are distributed to the internal temperature field space and the boundary temperature field space based on the velocity gradient and the temperature gradient equivalence principle, the internal temperature field space and the boundary temperature field space are integrated, and the on-site three-dimensional temperature field is obtained;

[0033] A nonlinear relationship formula is established with the coal quality multi-channel spectral information and the derived enhanced feature as output variables, and the nonlinear relationship formula includes temperature correlation coefficients and spectral correlation coefficients;

[0034] The weight coefficients of the temperature correlation coefficients and the spectral correlation coefficients are adjusted based on the particle swarm optimization algorithm, and the weight coefficients of the temperature correlation coefficients and the spectral correlation coefficients are optimized by introducing Gaussian disturbance and amplitude change rate based on the particle swarm optimization algorithm;

[0035] The Bayesian information criterion of the nonlinear relationship formula is minimized by using iterative operation, the temperature correlation coefficients and the spectral correlation coefficients that meet the convergence condition are determined based on the minimization of the Bayesian information criterion, the derived enhanced feature is obtained, and the signal enhancement set is obtained by merging the derived enhanced feature, the on-site three-dimensional temperature field, and the adaptive time period.

[0036] Preferably, when the index recognition model based on the support vector machine is pre-constructed, the index recognition model takes the support vector machine model as a basic framework, and further includes an input layer and an output layer, a singular value decomposition algorithm is introduced in the space-time feature extraction operator layer between the input layer and the support vector machine, an improved random forest algorithm is introduced in the support vector machine model, and an industrial analysis index of a multi-channel weighted feature map is predicted by a parallel SVM prediction head, and the industrial analysis index is taken as an initial index set output.

[0037] Preferably, the method for outputting an initial index set by the index recognition model based on a multi-channel weighted feature map comprises:

[0038] characteristic combination set comprising at least one group of index characteristics, and mutual information between the index characteristics in the characteristic combination set and other index characteristics is calculated;

[0039] The mutual information between the index characteristics is taken as the distance of the index characteristics, M index characteristics closest to the distance of the index characteristics are determined, the feature weight update probability of the M index characteristics is determined based on the Relief algorithm, the M index characteristics are updated in weight based on the feature weight update probability, and an initial feature set after weight update is obtained;

[0040] The initial feature set after the elimination of redundant index characteristics is subjected to index characteristic extraction, and at least one index decision tree is constructed, the error value and the maximum information coefficient of the index decision tree are weighted, the error value and the maximum information coefficient of the index decision tree are taken as the importance of the index characteristics, and a multi-channel weighted feature map combined with the importance of the characteristics is constructed;

[0041] The initial feature set after the elimination of redundant index characteristics is subjected to index characteristic extraction, and at least one index decision tree is constructed, the error value and the maximum information coefficient of the index decision tree are weighted, the error value and the maximum information coefficient of the index decision tree are taken as the importance of the index characteristics, and a multi-channel weighted feature map combined with the importance of the characteristics is constructed;

[0042] The spectrum channel RBF kernel function, the environment channel linear kernel function and the derivative channel polynomial kernel function are set for the multi-channel weighted feature map, the spectrum channel RBF kernel function, the environment channel linear kernel function and the derivative channel polynomial kernel function are weighted and fused based on the attention mechanism, and the industrial analysis index of the multi-channel weighted feature map is predicted through the parallel SVM prediction head, and the industrial analysis index is taken as the initial index set output.

[0043] In another aspect, the present application also provides a multi-channel coal quality online detection system based on signal enhancement, comprising:

[0044] A signal enhancement module, which combines the coal quality multi-channel spectral information and the associated environmental information into a real-time information set based on the real-time acquisition of the coal quality multi-channel spectral information and the associated environmental information by the coal quality detection equipment, and outputs a signal enhancement set after signal enhancement preprocessing of the real-time information set;

[0045] A multi-channel analysis module, which is used for pre-building an index recognition model based on a support vector machine, iteratively training the signal enhancement set until convergence, loading the signal enhancement set, selecting initial features of the signal enhancement set based on the index recognition model, determining feature importance through an improved random forest algorithm, constructing a multi-channel weighted feature map combined with the feature importance, and outputting an initial index set by the index recognition model combined with the multi-channel weighted feature map;

[0046] The index correction module is configured to acquire an initial index set, determine a detection confidence of the initial index set based on a multi-head attention mechanism, dynamically evaluate a drift risk amount of the initial index set, perform index collaborative correction based on the drift risk amount of the initial index set and the initial index set, and output a coal quality index detection result after collaborative correction.

[0047] Preferably, the signal enhancement module comprises:

[0048] The signal cleaning unit is configured to load a real-time information set, perform outlier cleaning and elimination on the real-time information set based on a boundary hybrid sampling strategy, obtain the real-time information set after outlier cleaning and elimination, and perform normalization processing on the real-time information set.

[0049] The adaptive interception unit is configured to acquire the real-time information set after normalization processing, identify a signal waveband of the real-time information set based on a spectral analysis method, determine a peak value feature of the signal waveband, perform adaptive time period interception on coal quality multi-channel spectral information in the real-time information set in combination with the peak value feature of the signal waveband, and align associated environmental information with multiple groups of the adaptive time period by using a timestamp alignment method to obtain an adaptive time period set.

[0050] The dynamic filtering unit is configured to perform dynamic filtering on the adaptive time period set based on a BEADS algorithm in combination with an S-G filtering method, and output the adaptive time period set after dynamic filtering.

[0051] The signal coupling unit is configured to load the adaptive time period set, construct a field three-dimensional temperature field based on the adaptive time period set, couple the field three-dimensional temperature field with the adaptive time period, obtain derived enhanced features, and combine the derived enhanced features, the field three-dimensional temperature field, and the adaptive time period to obtain a signal enhancement set.

[0052] Preferably, the adaptive interception unit comprises:

[0053] The spectral analysis module is configured to acquire the real-time information set after normalization processing, identify a signal waveband of the real-time information set based on a spectral analysis method, and determine a peak value feature of the signal waveband.

[0054] The time period interception module is configured to perform adaptive time period interception on coal quality multi-channel spectral information in the real-time information set in combination with the peak value feature of the signal waveband.

[0055] The time period alignment module is configured to align associated environmental information with multiple groups of the adaptive time period by using a timestamp alignment method to obtain an adaptive time period set.

[0056] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0057] In the embodiment of the present application, the real-time collected coal quality multi-channel spectral information and associated environmental information are subjected to signal enhancement preprocessing, and the interference of environmental factors on the spectral signal is considered during the signal enhancement preprocessing, the derived enhanced features are obtained by coupling the on-site three-dimensional temperature field with the adaptive time period, the interference of environmental factors on the signal can be effectively compensated, the signal after environmental factor compensation is more stable, the signal drift and distortion caused by environmental changes are reduced, and thus the stability and accuracy of the signal are improved.

[0058] In the embodiment of the present application, when the real-time information set is subjected to signal enhancement preprocessing, the abnormal value cleaning and elimination of the real-time information set based on the boundary mixed sampling strategy can effectively propose non-Gaussian distribution data, so as to ensure that the samples in the real-time information set are representative of coal quality while avoiding the problem of data elimination misjudgment, the BEADS algorithm is combined with the S-G filtering method to dynamically filter the adaptive time period set, the signal is more stable, the signal fluctuation caused by baseline drift is reduced, and the derived enhanced features are obtained by coupling the on-site three-dimensional temperature field with the adaptive time period, and the feature expression ability of the signal is further enhanced.

[0059] In the embodiment of the present application, when the real-time information set is subjected to signal enhancement preprocessing, the abnormal value cleaning and elimination of the real-time information set based on the boundary mixed sampling strategy can effectively propose non-Gaussian distribution data, so as to ensure that the samples in the real-time information set are representative of coal quality while avoiding the problem of data elimination misjudgment, the BEADS algorithm is combined with the S-G filtering method to dynamically filter the adaptive time period set, the signal is more stable, the signal fluctuation caused by baseline drift is reduced, and the derived enhanced features are obtained by coupling the on-site three-dimensional temperature field with the adaptive time period, and the feature expression ability of the signal is further enhanced.

[0060] In the embodiment of the present application, when the real-time information set is subjected to signal enhancement preprocessing, the abnormal value cleaning and elimination of the real-time information set based on the boundary mixed sampling strategy can effectively propose non-Gaussian distribution data, so as to ensure that the samples in the real-time information set are representative of coal quality while avoiding the problem of data elimination misjudgment, the BEADS algorithm is combined with the S-G filtering method to dynamically filter the adaptive time period set, the signal is more stable, the signal fluctuation caused by baseline drift is reduced, and the derived enhanced features are obtained by coupling the on-site three-dimensional temperature field with the adaptive time period, and the feature expression ability of the signal is further enhanced.

[0061] In the embodiment of the present application, the influence of environmental factors on the spectral signal is comprehensively considered by coupling the on-site three-dimensional temperature field with the adaptive time period, thereby improving the accuracy and reliability of signal processing. The modeling accuracy of the three-dimensional temperature field is improved by accurately establishing a coal on-site grid model. The accurate distribution of the associated temperature parameters is realized by combining the velocity gradient and the temperature gradient equivalence principle. The temperature distribution on site is comprehensively simulated, the interference of temperature change on the spectral signal is reduced, the stability of the signal is enhanced, and the combination of the temperature correlation coefficient and the spectral correlation coefficient determined by the Bayesian information criterion minimization generates a derived enhanced feature, further enhances the feature expression ability of the signal, provides more rich feature information for subsequent signal enhancement and coal quality index detection, and improves the accuracy and reliability of detection.

[0062] In the embodiment of the present application, when the index recognition model outputs an initial index set in combination with a multi-channel weighted feature map, a space-time feature extraction operator layer is arranged between the input layer and the support vector machine of the index recognition model, which can extract the time and space features in the signal and enhance the processing capability of the model on complex data. The mutual information between the features is calculated, and the mutual information is used as the distance between the features, which can more accurately evaluate the correlation between the features. The feature weight update probability is determined based on the Relief algorithm, which can dynamically adjust the feature weight and update the weight of the features, enhance the contribution of important features, and reduce the influence of redundant features. Finally, the support vector machine model sets a spectral channel RBF kernel function, an environment channel linear kernel function and a derived channel polynomial kernel function for the multi-channel weighted feature map, which can better process different types of features. And the kernel functions of different channels are weighted and fused based on the attention mechanism, which can dynamically adjust the contribution of each channel and improve the adaptability and prediction ability of the model. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 is the implementation flowchart of the multi-channel coal quality online detection method based on signal enhancement provided by the present application.

[0064] Figure 2 The signal enhancement preprocessing method implementation flowchart for the real-time information set is shown.

[0065] Figure 3 The abnormal value cleaning and elimination method implementation flowchart based on the boundary mixed sampling strategy for the real-time information set is shown.

[0066] Figure 4 The dynamic filtering method implementation flowchart based on the BEADS algorithm combined with the S-G filtering method for the adaptive time period set is shown.

[0067] Figure 5 The three-dimensional temperature field and adaptive time period coupling method implementation flowchart is shown.

[0068] Figure 6 A method for realizing an initial index set by combining an index recognition model with a multi-channel weighted feature map is shown.

[0069] Figure 7 A schematic diagram of the architecture of a multi-channel coal quality online detection system based on signal enhancement is shown. DETAILED DESCRIPTION

[0070] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used in the description herein is for describing specific embodiments only and is not intended to be limiting of the application; the description and the drawings are to be regarded as illustrative in nature and are not intended to limit the application; the terminology used in the description and the claims of the present application and the above description of the drawings includes the terms specifically mentioned above as well as any equivalents thereof.

[0071] The existing method reduces the noise of the spectrum signal of the sample by wavelet transform, and reduces the interference of the spectrum signal by adverse factors, but the wavelet threshold denoising adopts a fixed threshold rule, which cannot adapt to the change of noise characteristics in different temperature intervals, and the influence of the related environmental information on the spectrum signal is not considered, so that the measured spectrum signal is prone to drift and distortion, thereby affecting the accuracy and stability of the detection result, in view of the above problems, we propose a multi-channel coal quality online detection system and method based on signal enhancement, briefly speaking, when the method is implemented, first, the real-time information set is preprocessed by signal enhancement, the initial feature selection is performed on the signal enhancement set based on the index recognition model, the feature importance is determined by improving the random forest algorithm, the multi-channel weighted feature map combined with the feature importance is constructed, the initial index set is output by the index recognition model combined with the multi-channel weighted feature map, and the index is cooperatively corrected based on the drift risk quantity of the initial index set and the initial index set.

[0072] The embodiment of the present application provides a multi-channel coal quality online detection method based on signal enhancement, Figure 1 A method for realizing an initial index set by combining an index recognition model with a multi-channel weighted feature map is shown.

[0073] S10, based on the coal quality detection equipment real-time acquisition coal quality multi-channel spectral information and associated environmental information, the coal quality multi-channel spectral information and associated environmental information is merged into real-time information set, the signal enhancement preprocessing is carried out to real-time information set, and signal enhancement set is outputted;

[0074] Wherein, the coal quality detection equipment includes but is not limited to laser-induced breakdown spectroscopy (LIBS) equipment, near-infrared spectroscopy (NIR) equipment, X-ray fluorescence (XRF) equipment, gas chromatography-mass spectrometry (GC-MS) equipment, and coal quality multi-channel spectral information includes but is not limited to ultraviolet-visible spectral information, near-infrared spectral information, Raman spectral information, fluorescence spectral information, and associated environmental information includes but is not limited to associated temperature parameters, associated humidity parameters, associated air pressure parameters and coal quality sample position parameters.

[0075] S20, pre-constructing an index recognition model based on a support vector machine, and iteratively training the signal enhancement set to convergence, loading the signal enhancement set, performing initial feature selection on the signal enhancement set based on the index recognition model, determining feature importance by improving the random forest algorithm, constructing a multi-channel weighted feature map combined with the feature importance, and the index recognition model outputs an initial index set combined with the multi-channel weighted feature map;

[0076] S30, obtaining the initial index set, determining the detection confidence of the initial index set based on the multi-head attention mechanism, dynamically evaluating the drift risk amount of the initial index set, performing index coordination correction based on the drift risk amount of the initial index set and the initial index set, and outputting the coal quality index detection result after coordination correction.

[0077] In the embodiment of the application, the real-time acquisition of coal quality multi-channel spectral information and associated environmental information is signal enhanced and preprocessed, the interference of environmental factors on spectral signals is considered during signal enhancement preprocessing, the derived enhancement features are obtained by coupling the on-site three-dimensional temperature field with the adaptive time period, the interference of environmental factors on signals can be effectively compensated, the signals after environmental factor compensation are more stable, the signal drift and distortion caused by environmental changes are reduced, and the stability and accuracy of the signals are improved.

[0078] The embodiment of the application provides a signal enhancement preprocessing method for real-time information set, Figure 2 A signal enhancement preprocessing method for real-time information set is shown, and the signal enhancement preprocessing method for real-time information set specifically includes:

[0079] S101, load real-time information set, perform outlier cleaning and removal on real-time information set based on boundary hybrid sampling strategy, obtain real-time information set after outlier cleaning and removal, and perform normalization processing on real-time information set;

[0080] S102, obtain the normalized real-time information set, identify the signal band of the real-time information set based on the spectral analysis method, determine the peak value characteristics of the signal band, combine the peak value characteristics of the signal band to adaptively cut the time period of the coal quality multi-channel spectral information in the real-time information set, and align the associated environmental information with the multiple adaptive time periods by using the time stamp alignment method to obtain the adaptive time period set.

[0081] S103, dynamically filter the adaptive time period set based on the BEADS algorithm combined with the S-G filtering method, and output the dynamically filtered adaptive time period set.

[0082] S104, load the adaptive time period set, construct the on-site three-dimensional temperature field based on the adaptive time period set, couple the on-site three-dimensional temperature field with the adaptive time period, obtain the derived enhanced characteristics, and combine the derived enhanced characteristics, the on-site three-dimensional temperature field, and the adaptive time period to obtain the signal enhancement set.

[0083] In the embodiment of the present application, when the real-time information set is preprocessed for signal enhancement, the abnormal value cleaning and elimination of the real-time information set based on the boundary mixed sampling strategy can effectively eliminate non-Gaussian distribution data, thereby ensuring that the samples in the real-time information set are representative of coal quality while avoiding data elimination errors. By dynamically filtering the adaptive time period set by the BEADS algorithm combined with the S-G filtering method, the signal is more stable, and the signal fluctuation caused by baseline drift is reduced. By coupling the on-site three-dimensional temperature field with the adaptive time period, the derived enhanced characteristics can be obtained, further enhancing the feature expression ability of the signal.

[0084] The embodiment of the present application provides a method for cleaning and eliminating abnormal values of a real-time information set based on a boundary mixed sampling strategy, Figure 3 A flowchart for implementing a method for cleaning and eliminating abnormal values of a real-time information set based on a boundary mixed sampling strategy is shown, and the method for cleaning and eliminating abnormal values of a real-time information set based on a boundary mixed sampling strategy specifically includes:

[0085] S1011, obtain the real-time information set, and determine the coefficient of variation of data points in the real-time information set by using a boundary point detection algorithm;

[0086] Wherein, the coefficient of variation of data points is calculated by the following formula:

[0087]

[0088] Wherein, a coefficient of variation of data points, respectively, a number of adjacent data points in the real-time information set whose distance from the data point is a distance between data points in the real-time information set;

[0089] S1012, setting a variation threshold of the real-time information set based on the coefficient of variation of the data points, dividing data points in the real-time information set whose coefficient of variation is greater than the variation threshold into boundary data points and the rest into center data points through the variation threshold of the real-time information set;

[0090] S1013, loading the boundary data points, copying the boundary data points using the SMOTE technique based on a preset oversampling rate of the boundary data point weight to obtain new boundary data points after copying;

[0091] wherein the oversampling rate is determined by the following formula:

[0092]

[0093]

[0094] wherein, is the oversampling rate, is a rounding up function, respectively, a weight coefficient of the data type of the data point, a maximum weight coefficient of the data type, is a total amount of data types, is a total amount of boundary data points, is an average coefficient of variation of data points associated with the data type;

[0095] S1014, loading the center data points, traversing the center data points based on a preset undersampling rate of the center data point density, and deleting and cleaning the center data points based on the Tomek-Links method combined with the undersampling rate to obtain new center data after reconstruction;

[0096] wherein the undersampling rate is preset by the following formula:

[0097]

[0098] wherein, is the undersampling rate, respectively, a center data point density, a center data point average density, is a density retention coefficient;

[0099] S1015, obtaining the new boundary data points after copying and the new center data after reconstruction, and integrating the real-time information set after cleaning and removing outliers based on the new boundary data points after copying and the new center data after reconstruction.

[0100] S1016, load the real-time information set after the abnormal value cleaning and elimination, and normalize the real-time information set.

[0101] In the embodiment of the application, when the real-time information set is cleaned and eliminated based on the boundary mixed sampling strategy, the variation threshold of the real-time information set is set based on the coefficient of variation of data points, and the oversampling multiple is preset based on the weight of boundary data points. On the one hand, the data points can be accurately and explicitly divided into boundary data points and center data points. On the other hand, through the synthesis of the minority oversampling technology, the quality of the boundary data points can be increased, and the representativeness of the boundary data points can be improved. Through the Tomek-Links method combined with the undersampling rate, the redundant data in the center data points is deleted, the distribution of the center data points is significantly optimized, and through the integration of the copied boundary new data points and the reconstructed center new data, a high-quality real-time information set is formed, which provides a reliable data basis for subsequent signal enhancement preprocessing.

[0102] The embodiment of the application provides a dynamic filtering method for an adaptive time period set based on a BEADS algorithm combined with an S-G filtering method, Figure 4 The embodiment of the application provides a dynamic filtering method for an adaptive time period set based on a BEADS algorithm combined with an S-G filtering method,

[0103] S1031, load the adaptive time period set, initialize the parameters of the BEADS algorithm, and set the iteration number, wherein the parameters of the BEADS algorithm are , respectively, the iteration number is 15-20 times, the BEADS algorithm allows the initialization parameters and sets the iteration number, which enables the algorithm to dynamically adjust the parameters according to the characteristics of the data to achieve the best noise reduction effect. Compared with wavelet transform, the BEADS algorithm provides more flexibility and can better adapt to different types of spectral signals;

[0104] wherein the initial regularization weight is determined by the following formula:

[0105]

[0106] wherein, represents the initial regularization weight, is the signal-to-noise ratio of the adaptive time period, is the weight scaling coefficient;

[0107] S1032, the difference operator of the multi-channel spectral information of the coal quality in the adaptive time period is calculated based on the BEADS algorithm, and the spectral information estimation is updated, the baseline component and the noise component of the multi-channel spectral information of the coal quality in the adaptive time period are determined after a preset number of iterations, the BEADS algorithm can accurately separate the baseline component and the noise component in the spectral signal by calculating the difference operator and iterative updating, so that the signal-to-noise ratio of the signal is improved;

[0108] S1033, the estimated baseline component and the noise component are subtracted from the original signal to obtain the adaptive time period after denoising;

[0109] S1034, the window size and the polynomial order of the S-G filter corresponding to different adaptive time periods are determined based on the smoothness of the adaptive time period, the S-G filter method realizes smoothing of the multi-channel spectral information of the coal quality in the adaptive time period through local polynomial fitting, and outputs a set of adaptive time periods after dynamic filtering, it should be noted that the selection of the polynomial order should be dynamically adjusted according to the complexity of the signal. The higher the polynomial order, the stronger the fitting ability, but the higher the calculation complexity, and it may lead to overfitting, in the embodiment, the window size and the polynomial order of the S-G filter corresponding to different adaptive time periods are determined based on the smoothness of the adaptive time period, which can realize adaptive smoothing processing of signals with different smoothness. This method not only can improve the smoothing effect of the signal, but also can reduce signal distortion, improve the quality and stability of the signal.

[0110] In the embodiment of the application, when the adaptive time period set is dynamically filtered based on the BEADS algorithm combined with the S-G filter method, the BEADS algorithm can accurately separate the baseline component and the noise component in the spectral signal through the difference operator and iterative updating, thereby reducing signal distortion, and the S-G filter method can dynamically adjust parameters according to the smoothness of the adaptive time period, realize smoothing processing through local polynomial fitting, reduce the interference of high-frequency noise, and at the same time, retain the characteristic peaks of the signal, enhance the signal clarity, and effectively reduce the noise interference.

[0111] The embodiment of the application provides a method for coupling a field three-dimensional temperature field and an adaptive time period, Figure 5 A flowchart for realizing the method for coupling the field three-dimensional temperature field and the adaptive time period is shown, and the method for coupling the field three-dimensional temperature field and the adaptive time period specifically comprises:

[0112] S1041, load the adaptive time period set, extract the associated environmental information in the adaptive time period set, by extracting the associated environmental information in the adaptive time period set, the influence of environmental factors on the spectrum signal can be comprehensively considered, the accuracy and reliability of signal processing are improved, a coal site grid model is established based on the associated environmental information, the coal site grid model is divided, and the internal temperature field space and the boundary temperature field space of the coal site grid model sub-grid are obtained;

[0113] S1042, obtain the internal temperature field space and the boundary temperature field space of the coal site grid model sub-grid, combine the velocity gradient, the temperature gradient equivalent principle, and distribute the dynamic filtered associated temperature parameters to the internal temperature field space and the boundary temperature field space, integrate the internal temperature field space and the boundary temperature field space, obtain the site three-dimensional temperature field, by dividing the coal site grid model, the internal temperature field space and the boundary temperature field space are obtained, the distribution characteristics of the temperature field can be analyzed in more detail, and the modeling accuracy of the temperature field is improved, and by combining the velocity gradient and the temperature gradient equivalent principle, the dynamic filtered associated temperature parameters are distributed to the internal temperature field space and the boundary temperature field space, the dynamic change of the temperature field can be more accurately reflected, the internal temperature field space and the boundary temperature field space are integrated, the site three-dimensional temperature field is obtained, the temperature distribution of the site can be comprehensively simulated, and the integrity and accuracy of the temperature field modeling are improved;

[0114] S1043, establish a nonlinear relationship of coal quality multi-channel spectrum information with associated temperature parameters and adaptive time periods as output variables and derived enhanced features as output variables, the nonlinear relationship includes temperature correlation coefficients and spectrum correlation coefficients;

[0115] S1044, adjust the weight coefficients of the temperature correlation coefficients and the spectrum correlation coefficients based on the particle swarm optimization algorithm, by introducing Gaussian disturbance and amplitude change rate, the optimization process can be avoided from falling into local optimum, the globality of optimization is improved, and the weight coefficients of the temperature correlation coefficients and the spectrum correlation coefficients are optimized by introducing Gaussian disturbance and amplitude change rate based on the particle swarm optimization algorithm;

[0116] S1045, the Bayesian information criterion of the nonlinear relationship is minimized by using iterative operation, the combination of the temperature correlation coefficient and the spectrum correlation coefficient satisfying the convergence condition is determined based on the Bayesian information criterion minimization, the derived enhanced features are obtained, the derived enhanced features, the on-site three-dimensional temperature field and the adaptive time period are combined, and the signal enhancement set is obtained, and it should be noted that the derived enhanced features refer to a new set of features generated by processing and analyzing original data, which can better reflect the internal structure and pattern of data, thereby improving the performance and prediction ability of the model, in the embodiment, the derived enhanced features can be temperature compensation spectrum features, temperature anti-interference features, multi-sensor collaborative features and adversarial generation features.

[0117] In the embodiment of the present application, the influence of environmental factors on the spectrum signal is comprehensively considered by coupling the on-site three-dimensional temperature field and the adaptive time period, thereby improving the accuracy and reliability of signal processing, and the modeling accuracy of the three-dimensional temperature field is improved by accurately establishing the on-site grid model, the accurate distribution of the related temperature parameters is realized by combining the velocity gradient and the temperature gradient equivalence principle, the temperature distribution of the site is finally simulated comprehensively, the interference of temperature change on the spectrum signal is reduced, the stability of the signal is enhanced, and the combination of the temperature correlation coefficient and the spectrum correlation coefficient determined by the Bayesian information criterion minimization generates the derived enhanced features, further enhances the feature expression ability of the signal, provides more rich feature information for subsequent signal enhancement and coal quality index detection, and improves the accuracy and reliability of detection.

[0118] In the embodiment, when the index recognition model based on the support vector machine is pre-constructed, the index recognition model is based on a support vector machine model as a basic framework, and the index recognition model further includes an input layer and an output layer. A spatio-temporal feature extraction operator layer is arranged between the input layer and the support vector machine. The spatio-temporal feature extraction operator layer introduces a singular value decomposition algorithm. The spatio-temporal feature extraction operator layer combines the singular value decomposition algorithm to perform feature extraction on the signal enhancement set, to obtain a feature combination set including at least one group of index features. Mutual information between the index features in the feature combination set and other index features is calculated. The mutual information between the index features is used as the distance of the index features. M index features closest to the index features are determined. The feature weight update probability of the M index features is determined based on a Relief algorithm. The M index features are updated in weight based on the feature weight update probability, to obtain an initial feature set after weight update. An improved random forest algorithm is introduced in the support vector machine model. The improved random forest algorithm introduces a Chebyshev chaotic mapping network and a maximum information coefficient based on a random forest algorithm. The Chebyshev chaotic mapping network maps the initial feature set to a weight value of 0 to 1 through an approximate impulse function. Redundant index features are eliminated based on a preset weight value screening threshold. Index feature extraction is performed on the initial feature set after elimination of the redundant index features, and at least one index decision tree is constructed. The error value and the maximum information coefficient of the index decision tree are weighted, and the error value and the maximum information coefficient of the index decision tree are used as the importance of the index features. A multi-channel weighted feature map combined with the feature importance is constructed. The support vector machine model sets a spectral channel RBF kernel function, an environment channel linear kernel function, and a derivative channel polynomial kernel function for the multi-channel weighted feature map. The spectral channel RBF kernel function, the environment channel linear kernel function, and the derivative channel polynomial kernel function are weighted and fused based on an attention mechanism, and an industrial analysis index of the multi-channel weighted feature map is predicted through a parallel SVM prediction head. The industrial analysis index is output as an initial index set, wherein the industrial analysis index is a coal ash, volatile matter, and sulfur index.

[0119] The index recognition model training method includes:

[0120] S201, at least one group of modeling samples is generated based on a PSASP simulation software, and the modeling samples are subjected to disturbance processing to obtain a modeling sample set after disturbance processing;

[0121] S202, a pre-constructed index recognition model is loaded, a training round and a hyperparameter of the index recognition model are preset, and the pre-constructed index recognition model is subjected to ablation processing;

[0122] S203, the modeling sample set is subjected to signal enhancement preprocessing, and the index recognition model is trained through the modeling sample set to obtain a training output result;

[0123] S204, calculating a mean square error value of the training output result and the true result;

[0124] S205, judging whether the mean square error value is less than a preset difference threshold value;

[0125] S206, if the mean square error value is less than the preset difference threshold value, outputting a converged index recognition model;

[0126] If the mean square error value is not less than the preset difference threshold value, adjusting the hyperparameters of the index recognition model through the Adam optimizer, returning to S203, and continuing to iteratively train the model.

[0127] The embodiment of the present application provides a method for outputting an initial index set by an index recognition model in combination with a multi-channel weighted feature map, Figure 6 The embodiment of the present application provides a method for outputting an initial index set by an index recognition model in combination with a multi-channel weighted feature map,

[0128] S301, performing feature extraction on the signal enhancement set to obtain a feature combination set including at least one group of index features, and calculating mutual information between the index features in the feature combination set and other index features;

[0129] S302, taking the mutual information between the index features as the distance of the index features, determining M index features closest to the index features, determining the feature weight update probability of the M index features based on the Relief algorithm, performing weight update on the M index features based on the feature weight update probability, and obtaining an initial feature set after weight update;

[0130] S303, using a Chebyshev chaotic mapping network based on an improved random forest algorithm on the initial feature set, mapping the initial feature set to a weight value of 0 to 1 through an approximate impulse function, and eliminating redundant index features based on a preset weight value threshold;

[0131] S304, performing index feature extraction on the initial feature set after eliminating the redundant index features, and constructing at least one index decision tree, weighting the error value and the maximum information coefficient of the index decision tree, taking the error value and the maximum information coefficient of the index decision tree as the importance of the index features, and constructing a multi-channel weighted feature map in combination with the importance of the features;

[0132] S305, setting a spectral channel RBF kernel function, an environment channel linear kernel function and a derived channel polynomial kernel function for the multi-channel weighted feature map, weighting and fusing the spectral channel RBF kernel function, the environment channel linear kernel function and the derived channel polynomial kernel function based on an attention mechanism, and predicting the industrial analysis index of the multi-channel weighted feature map through a parallel SVM prediction head to output the industrial analysis index as an initial index set.

[0133] In the embodiment of the present application, when the index recognition model outputs the initial index set based on the multi-channel weighted feature map, the index recognition model sets a time-space feature extraction operator layer between the input layer and the support vector machine, which can extract the time and space features in the signal, enhance the processing capacity of the model for complex data, calculate the mutual information between the features, and use the mutual information as the distance between the features to more accurately evaluate the correlation between the features, determine the feature weight update probability based on the Relief algorithm, dynamically adjust the feature weight, and update the weight of the features, which can enhance the contribution of important features and reduce the influence of redundant features, and finally the support vector machine model sets the spectral channel RBF kernel function, the environment channel linear kernel function and the derived channel polynomial kernel function for the multi-channel weighted feature map, which can better process different types of features, and based on the attention mechanism, the kernel functions of different channels are weighted and fused, which can dynamically adjust the contribution of each channel and improve the adaptability and prediction ability of the model.

[0134] In the embodiment of the present application, the index collaborative correction method based on the drift risk quantity and the initial index set includes:

[0135] S401, calculating the event decay weight of the industrial analysis index in the initial index set based on the multi-head attention mechanism combined with the exponential decay model;

[0136] S402, constructing a time-space-environment attention architecture through the multi-head attention mechanism, fusing the event decay weight of the industrial analysis index with the time-space-environment attention architecture, and calculating the detection confidence of the industrial analysis index through attention weight aggregation;

[0137] S403, constructing a partial least squares regression function based on a drift compensation strategy with the detection confidence of the industrial analysis index as a constraint, introducing a GAN architecture to generate a drift risk quantity compensation value in the partial least squares regression function, and performing index collaborative correction on the initial index set based on the smoothing correction method combined with the drift risk quantity.

[0138] On the other hand, the embodiment of the present application also provides a multi-channel coal quality online detection system based on signal enhancement, Figure 7 The architecture schematic diagram of the multi-channel coal quality online detection system based on signal enhancement is shown, and the multi-channel coal quality online detection system based on signal enhancement specifically comprises:

[0139] The signal enhancement module 100 is based on the real-time acquisition of coal quality multi-channel spectral information and associated environmental information by a coal quality detection device, combines the coal quality multi-channel spectral information and the associated environmental information into a real-time information set, performs signal enhancement preprocessing on the real-time information set, and outputs a signal enhancement set;

[0140] The signal enhancement module 100 includes:

[0141] The signal cleaning unit 110 is configured to load the real-time information set, perform outlier cleaning and removal on the real-time information set based on a boundary mixed sampling strategy, obtain the real-time information set after outlier cleaning and removal, and perform normalization processing on the real-time information set.

[0142] The adaptive interception unit 120 is configured to obtain the normalized real-time information set, identify a signal waveband of the real-time information set based on a spectral analysis method, determine a peak value feature of the signal waveband, perform adaptive time period interception on the coal quality multi-channel spectral information in the real-time information set in combination with the peak value feature of the signal waveband, and align the associated environmental information with multiple groups of the adaptive time period by using a timestamp alignment method to obtain an adaptive time period set.

[0143] The dynamic filtering unit 130 is configured to perform dynamic filtering on the adaptive time period set based on a BEADS algorithm in combination with an S-G filtering method, and output the adaptive time period set after dynamic filtering.

[0144] The signal coupling unit 140 is configured to load the adaptive time period set, construct a field three-dimensional temperature field based on the adaptive time period set, couple the field three-dimensional temperature field with the adaptive time period, obtain derived enhancement features, and combine the derived enhancement features, the field three-dimensional temperature field, and the adaptive time period to obtain the signal enhancement set.

[0145] The adaptive interception unit 120 includes:

[0146] The spectral analysis module 121 is configured to obtain the normalized real-time information set, identify a signal waveband of the real-time information set based on a spectral analysis method, and determine a peak value feature of the signal waveband.

[0147] The time period interception module 122 is configured to perform adaptive time period interception on the coal quality multi-channel spectral information in the real-time information set in combination with the peak value feature of the signal waveband.

[0148] The time period alignment module 123 is configured to align the associated environmental information with multiple groups of the adaptive time period by using a timestamp alignment method to obtain an adaptive time period set.

[0149] The multi-channel analysis module 200 is used for pre-building a support vector machine-based index identification model, iteratively training a signal enhancement set to convergence, loading the signal enhancement set, performing initial feature selection on the signal enhancement set based on the index identification model, determining feature importance by improving a random forest algorithm, constructing a multi-channel weighted feature map combined with the feature importance, and outputting an initial index set by the index identification model combined with the multi-channel weighted feature map.

[0150] The index correction module 300 is used for acquiring the initial index set, determining a detection confidence of the initial index set based on a multi-head attention mechanism, dynamically evaluating a drift risk amount of the initial index set, performing index collaborative correction based on the drift risk amount of the initial index set and the initial index set, and outputting a coal quality index detection result after collaborative correction.

[0151] In summary, the application provides a multi-channel coal quality online detection system and method based on signal enhancement. In the embodiment of the application, signal enhancement preprocessing is performed on real-time collected coal quality multi-channel spectral information and associated environmental information. The interference of environmental factors on spectral signals is considered during the signal enhancement preprocessing. By coupling a three-dimensional temperature field with an adaptive time period, derived enhancement features are obtained, which can effectively compensate for the interference of environmental factors on signals. The signals after environmental factor compensation are more stable, and signal drift and distortion caused by environmental changes are reduced, thereby improving the stability and accuracy of the signals.

[0152] It should be noted that, for the foregoing embodiments, in order to simply describe, they are all expressed as a series of action combinations, but those skilled in the art should know that the application is not limited by the action sequence described, because according to the application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions and modules involved are not necessarily required by the application.

[0153] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the protection scope of the application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application. Although the application has been described in detail with reference to the above embodiments, those skilled in the art can still make some modifications to the features of the embodiments in the application without creative labor, such as mutual combination, addition or deletion or other adjustments according to the situation, so as to obtain different other technical solutions which do not deviate from the concept of the present application in essence. These technical solutions also belong to the scope of the present application.

Claims

1. A multi-channel coal quality on-line detection method based on signal enhancement, characterized in that, The method comprises: Based on the coal quality detection equipment, real-time acquisition of coal quality multi-channel spectral information and associated environmental information, merging the coal quality multi-channel spectral information and the associated environmental information into a real-time information set, signal enhancement preprocessing of the real-time information set, and output of a signal enhancement set; Pre-constructing a support vector machine-based index recognition model, iteratively training the signal enhancement set until convergence, loading the signal enhancement set, performing initial feature selection on the signal enhancement set based on the index recognition model, determining feature importance by improving the random forest algorithm, constructing a multi-channel weighted feature map combining the feature importance, and the index recognition model outputting an initial index set combining the multi-channel weighted feature map; Obtaining the initial index set, determining the detection confidence of the initial index set based on the multi-head attention mechanism, dynamically evaluating the drift risk of the initial index set, performing index collaborative correction based on the drift risk of the initial index set and the initial index set, and outputting the coal quality index detection result after collaborative correction; The signal enhancement preprocessing method of the real-time information set comprises: Loading the real-time information set, performing outlier cleaning and removing of the real-time information set based on a boundary hybrid sampling strategy, obtaining the real-time information set after outlier cleaning and removing, and performing normalization processing on the real-time information set; Obtaining the real-time information set after normalization processing, identifying the signal band of the real-time information set based on a spectral analysis method, determining the peak characteristics of the signal band, adaptively time period cutting of the coal quality multi-channel spectral information in the real-time information set combining the peak characteristics of the signal band, and aligning the associated environmental information with multiple sets of the adaptive time period using a timestamp alignment method to obtain an adaptive time period set; Dynamically filtering the adaptive time period set based on the BEADS algorithm combining the S-G filtering method, and outputting the adaptive time period set after dynamic filtering; Loading the adaptive time period set, constructing a field three-dimensional temperature field based on the adaptive time period set, coupling the field three-dimensional temperature field with the adaptive time period, obtaining derived enhanced features, merging the derived enhanced features, the field three-dimensional temperature field, and the adaptive time period, and obtaining the signal enhancement set.

2. The signal enhancement based multi-channel coal quality on-line detection method according to claim 1, characterized in that: The method for performing outlier cleaning and removing of the real-time information set based on the boundary hybrid sampling strategy comprises: Obtaining the real-time information set, and determining the coefficient of variation of the data points in the real-time information set using a boundary point detection algorithm; Setting a variation threshold of the real-time information set based on the coefficient of variation of the data points, dividing the data points in the real-time information set with a coefficient of variation greater than the variation threshold into boundary data points, and dividing the remaining data points into center data points; Loading the boundary data points, pre-setting an oversampling rate based on the weight of the boundary data points, replicating the boundary data points using a synthetic minority oversampling technique to obtain new boundary data points after replication; Loading the center data points, pre-setting an undersampling rate based on the density of the center data points, traversing the center data points, and reducing and cleaning the center data points based on the Tomek-Links method combining the undersampling rate to obtain new center data after reconstruction; Obtaining the new boundary data points after replication and the new center data after reconstruction, and integrating the new boundary data points after replication and the new center data after reconstruction into the real-time information set after outlier cleaning and removing. Load the real-time information set after the abnormal value cleaning and elimination, and normalize the real-time information set.

3. The signal enhancement based multi-channel coal quality on-line detection method according to claim 2, characterized in that: The dynamic filtering method based on the BEADS algorithm combined with the S-G filtering method for the adaptive time period set comprises: Load the adaptive time period set, initialize the parameters of the BEADS algorithm, and set the number of iterations; Calculate the difference operator of the multi-channel spectral information of the coal quality in the adaptive time period based on the BEADS algorithm and update the spectral information estimate, and determine the baseline component and the noise component of the multi-channel spectral information of the coal quality in the adaptive time period after a preset number of iterations; Subtract the estimated baseline component and noise component from the original signal to obtain the denoised adaptive time period; Determine the window size and the polynomial order of the S-G filtering corresponding to different adaptive time periods based on the smoothness of the adaptive time period, and realize the smoothing of the multi-channel spectral information of the coal quality in the adaptive time period through local polynomial fitting by the S-G filtering method, and output the dynamic filtered adaptive time period set.

4. The signal enhancement based multi-channel coal quality on-line detection method of claim 3, wherein: The method for coupling the field three-dimensional temperature field with the adaptive time period comprises: Load the adaptive time period set, extract the associated environmental information in the adaptive time period set, establish a coal field grid model based on the associated environmental information, divide the coal field grid model to obtain the internal temperature field space and the boundary temperature field space of the coal field grid model sub-grid; Obtain the internal temperature field space and the boundary temperature field space of the coal field grid model sub-grid, distribute the dynamically filtered associated temperature parameters to the internal temperature field space and the boundary temperature field space based on the velocity gradient and the temperature gradient equivalence principle, integrate the internal temperature field space and the boundary temperature field space to obtain the field three-dimensional temperature field; Establish a nonlinear relationship formula taking the associated temperature parameters and the multi-channel spectral information of the coal quality in the adaptive time period as output variables and the derived enhanced features as output variables, wherein the nonlinear relationship formula comprises temperature correlation coefficients and spectral correlation coefficients; Adjust the weight coefficients of the temperature correlation coefficients and the spectral correlation coefficients based on the particle swarm optimization algorithm, and introduce Gaussian disturbance and amplitude change rate based on the particle swarm optimization algorithm to optimize the weight coefficients of the temperature correlation coefficients and the spectral correlation coefficients; Minimize the Bayesian information criterion of the nonlinear relationship formula by using iterative operation, determine the combination of the temperature correlation coefficients and the spectral correlation coefficients that meet the convergence condition based on the minimization of the Bayesian information criterion, obtain the derived enhanced features, and combine the derived enhanced features, the field three-dimensional temperature field, and the adaptive time period to obtain the signal enhancement set.

5. The signal enhancement based multi-channel coal quality on-line detection method of claim 1, wherein: When pre-establishing the index recognition model based on the support vector machine, the index recognition model takes the support vector machine model as the basic framework, further comprises an input layer and an output layer, sets a singular value decomposition algorithm in the time-space feature extraction operator layer between the input layer and the support vector machine, introduces an improved random forest algorithm in the support vector machine model, and predicts the industrial analysis index of the multi-channel weighted feature map through the parallel SVM prediction head to output the initial index set.

6. The signal enhancement based multi-channel coal quality on-line detection method of claim 5, wherein: The method for outputting the initial index set by the index recognition model combined with the multi-channel weighted feature map comprises: The signal enhancement set is subjected to feature extraction to obtain a feature combination set including at least one group of index features, and mutual information between the index features in the feature combination set and other index features is calculated; The mutual information between the index features is taken as the distance of the index features, M index features with the shortest distance are determined, the feature weight update probability of the M index features is determined based on the Relief algorithm, the M index features are subjected to weight update based on the feature weight update probability, and an initial feature set after weight update is obtained; The initial feature set is subjected to Chebyshev chaotic mapping network based on the improved random forest algorithm, the initial feature set is mapped to a weight value of 0 to 1 by an approximate impulse function, and redundant index features are eliminated based on a preset weight value screening threshold; Index feature extraction is performed on the initial feature set after elimination of the redundant index features, at least one index decision tree is constructed, the error value and the maximum information coefficient of the index decision tree are weighted, the error value and the maximum information coefficient of the index decision tree are taken as the importance of the index features, and a multi-channel weighted feature map combined with the feature importance is constructed; Spectrum channel RBF kernel function, environment channel linear kernel function and derivative channel polynomial kernel function are set for the multi-channel weighted feature map, the spectrum channel RBF kernel function, the environment channel linear kernel function and the derivative channel polynomial kernel function are weighted and fused based on the attention mechanism, and the industrial analysis index of the multi-channel weighted feature map is predicted through a parallel SVM prediction head, and the industrial analysis index is taken as an initial index set output.

7. The multi-channel coal quality on-line detection system based on signal enhancement for implementing the multi-channel coal quality on-line detection method based on signal enhancement according to any one of the preceding claims 1-6, characterized in that: The multi-channel coal quality online detection system based on signal enhancement comprises: A signal enhancement module, which is configured to collect coal quality multi-channel spectrum information and associated environment information based on a coal quality detection device in real time, combine the coal quality multi-channel spectrum information and the associated environment information into a real-time information set, perform signal enhancement preprocessing on the real-time information set, and output a signal enhancement set; A multi-channel analysis module, which is configured to pre-construct an index recognition model based on a support vector machine, iteratively train the signal enhancement set until convergence, load the signal enhancement set, perform initial feature selection on the signal enhancement set based on the index recognition model, determine feature importance by using an improved random forest algorithm, construct a multi-channel weighted feature map combined with the feature importance, and output an initial index set by the index recognition model in combination with the multi-channel weighted feature map; An index correction module, which is configured to obtain the initial index set, determine a detection confidence of the initial index set based on a multi-head attention mechanism, dynamically evaluate a drift risk amount of the initial index set, perform index collaborative correction based on the drift risk amount of the initial index set and the initial index set, and output a coal quality index detection result after collaborative correction.

8. The signal enhancement based multi-channel coal quality on-line monitoring system according to claim 7, wherein: The signal enhancement module comprises: A signal cleaning unit, which is configured to load the real-time information set, perform outlier cleaning and elimination on the real-time information set based on a boundary mixed sampling strategy, obtain the real-time information set after outlier cleaning and elimination, and perform normalization processing on the real-time information set; An adaptive intercepting unit is configured to acquire the normalized real-time information set, identify a signal band of the real-time information set based on a spectral analysis method, determine a peak value feature of the signal band, adaptively intercept coal quality multi-channel spectral information in the real-time information set based on the peak value feature of the signal band, and align associated environmental information with multiple groups of the adaptive time periods using a timestamp alignment method to obtain an adaptive time period set. A dynamic filtering unit is configured to dynamically filter the adaptive time period set based on a BEADS algorithm combined with an S-G filtering method, and output the dynamically filtered adaptive time period set. A signal coupling unit is configured to load the adaptive time period set, construct a field three-dimensional temperature field based on the adaptive time period set, couple the field three-dimensional temperature field with the adaptive time period, obtain a derived enhanced feature, and combine the derived enhanced feature, the field three-dimensional temperature field, and the adaptive time period to obtain a signal enhancement set.

9. The signal enhancement based multi-channel coal quality on-line monitoring system according to claim 8, wherein: The adaptive intercepting unit includes: A spectral analysis module is configured to acquire the normalized real-time information set, identify a signal band of the real-time information set based on a spectral analysis method, and determine a peak value feature of the signal band. A time period intercepting module is configured to adaptively intercept coal quality multi-channel spectral information in the real-time information set based on the peak value feature of the signal band. A time period aligning module is configured to align associated environmental information with multiple groups of the adaptive time periods using a timestamp alignment method to obtain an adaptive time period set.

Images