Mobile platform coal quality remote detection system and method

Abstract

The invention discloses a mobile platform coal quality remote detection system and method, and solves the problem that the analysis precision and analysis efficiency of a model are affected by a spectral data acquisition error rate due to the fact that parameters of pulverized coal cannot be preprocessed when a model between a spectral line and a calorific value is established and a coefficient matrix and an error term of the model are obtained in an existing method. The method comprises the following steps: detecting stored coal in real time by adopting mobile coal quality detection equipment, pre-constructing a data processing model based on the combination of adaptive interactive multiple models and deep learning, preprocessing spectral sample data by the data processing model, and establishing a coal quality index quantitative model based on the combination of a convolutional neural network and a lightweight detection structure; according to the method, noise reduction, normalization and feature extraction preprocessing are performed on the real-time spectral data by pre-constructing the data processing model based on the combination of the adaptive interactive multiple models and deep learning, so that the spectral data acquisition error rate is reduced, and the analysis precision and analysis efficiency of the coal quality index quantitative model are ensured.

Description

Technical Field

[0001] The invention belongs to the technical field of coal quality detection, and in particular relates to a mobile platform coal quality remote detection system and method. Background Art

[0002] At present, coal platforms on railways usually use coal unloaders to unload coal on site, and the coal in each carriage is piled along the railway line. The coal stacking distance is far, the area occupied is large, and the coal storage volume is large. In order to avoid the accumulation of coal and occupying storage space, it is necessary to conduct rapid and accurate testing and analysis of the platform coal quality. The coal quality testing at the coal transfer platform still uses traditional chemical analysis methods. Usually, samples are taken from several tons of coal in each carriage, and then they are sent to the laboratory for analysis and testing item by item after mixing, grinding, sample preparation and other links. The test results have serious lags, and it is impossible to judge the coal quality in real time to guide loading and storage, and the representativeness of the sampling cannot be guaranteed.

[0003] In the prior art, Chinese patent CN116482080A discloses a coal quality analysis method based on LIBS technology, wherein the laser output by a laser is split to form a first light beam and a second light beam; the first light beam is incident on the coal powder, and a photodetector array receives scattered light signals and sends them to an analysis unit, which outputs the particle size distribution of the coal powder; the second light beam is incident on the coal powder to generate a plasma signal, which is sent to a spectrometer to obtain a spectrum of the coal powder, and the analysis unit uses an analysis model to process the spectrum and particle size distribution to obtain the parameters of the coal powder; however, the existing method cannot perform noise reduction, normalization and feature extraction preprocessing on the parameters of the coal powder when establishing a model between spectral lines and calorific value and obtaining the coefficient matrix and error terms of the model, so that the error rate of spectral data acquisition affects the model analysis accuracy and analysis efficiency. Based on this, we propose a mobile platform coal quality remote detection system and method. Summary of the invention

[0004] The purpose of the present invention is to address the shortcomings of the prior art and provide a mobile platform coal quality remote detection system and method, which solves the problem that the existing method cannot perform noise reduction, normalization and feature extraction preprocessing on the parameters of coal powder when establishing a model between spectral lines and calorific value and obtaining the coefficient matrix and error terms of the model, so that the error rate of spectral data acquisition affects the model analysis accuracy and analysis efficiency.

[0005] The present invention is implemented in this way: a mobile platform coal quality remote detection method, the mobile platform coal quality remote detection method comprising: Use mobile coal quality testing equipment to conduct real-time testing of stored coal and obtain real-time spectral data. During real-time testing, the laser pulses emitted by the laser induced breakdown spectrometer are enhanced based on dual-pulse laser technology. Pre-build a data processing model based on adaptive interactive multi-model combined with deep learning, and pre-process the real-time spectral data based on the data processing model to obtain a real-time data set; Acquire spectral sample data, input the spectral sample data into a pre-built data processing model, the data processing model pre-processes the spectral sample data to obtain a sample data set, divide the sample data set into a training set and a validation set, establish a coal quality index quantitative model based on a convolutional neural network combined with a lightweight detection structure, use the training set as input, iteratively train the coal quality index quantitative model until convergence, use the validation set to verify whether the model meets the preset detection accuracy, and if it meets the detection accuracy, output the trained coal quality index quantitative model; The coal quality index quantitative model is executed with the real-time data set as input. The coal quality index quantitative model analyzes the real-time data set to obtain a first detection result and a second detection result, wherein the first detection result includes a coal particle size detection result, a coal moisture detection result, a coal calorific value detection result, a coal volatile matter detection result, and a coal element detection result.

[0006] Preferably, the data processing model includes a preprocessing module, a data correction layer and a preprocessing output layer. The preprocessing module includes a normalization processing layer, a background noise reduction layer, a feature extraction layer and a baseline correction layer. The normalization processing layer is used to standardize the real-time spectral data and eliminate the influencing factors of the optical path and the data volume. The background noise reduction layer is used to perform background noise reduction on the real-time spectral data. The feature extraction layer is used to identify and extract spectral characteristic factors reflecting the quality of coal. The baseline correction layer is used to correct the spectral baseline. When pre-constructing a data processing model based on adaptive interactive multi-model combined with deep learning, the feature extraction layer is improved by combining the adaptive interactive multi-model and the PeleeNet detection structure, the feature interaction function Glou is set, and the feature extraction improvement layer is constructed.

[0007] Preferably, the method for preprocessing real-time spectral data based on the data processing model specifically includes: Loading real-time spectral data, and performing normalization processing on the real-time spectral data, wherein the normalization processing adopts Z-score normalization combined with vector normalization to normalize the data to obtain a normalized data set; Obtain a normalized data set, preset a wavelet denoising function and a wavelet basis threshold, use the wavelet denoising function to perform denoising on the normalized data set, and obtain a denoised output set; Among them, the wavelet denoising function is expressed as: (1) in, is the normalized input data set, is the mean variance of the normalized data set, is the normalization coefficient, is the wavelet coefficient, is the number of wavelet basis decomposition layers; Load the denoised output set, and use the Mosaic-Mixup hybrid enhancement algorithm to perform hybrid enhancement on the denoised output set to obtain a hybrid enhancement set; (2) is the mixed enhancement value, is the mixing coefficient of the Mosaic-Mixup hybrid enhancement algorithm, is a batch of data of the denoised output set collected by stratified sampling, Another batch of data for the denoising output set; (3) in, is the mixing parameter; A mixed enhancement set is obtained, and a singular value denoising algorithm is used to perform background denoising on the mixed enhancement set to obtain a background denoising set; Taking the background noise reduction set as input, the improved feature extraction layer performs feature extraction on the background noise reduction set to obtain the feature extraction set, and combines the adaptive Kalman filter algorithm to correct the element characteristic spectral lines and spectral baselines to output a real-time data set.

[0008] Preferably, the method of iteratively training the coal quality index quantitative model to convergence using the training set as input specifically includes: Pre-build a coal quality index quantitative model based on a convolutional neural network combined with a lightweight detection structure, and set an initial model of the coal quality index quantitative model, wherein the initial model of the coal quality index quantitative model includes an index input layer, a feature convolution architecture, a pooling layer, a feature decoder, and an index evaluator; The lightweight detection structure MobileNet is introduced into the feature convolution architecture, and the residual connection network is used to replace the pooling layer. The improved feature convolution architecture includes three layers of convolution and a MobileNet layer. The MobileNet layer contains 128 neurons and is combined with a ReLU activation function. Define the index evaluation function of the index evaluator, where the index evaluation function is defined as: (4) in, is the comprehensive evaluation score of coal quality, is the number of characteristic factors, is the model activation function, is the weight coefficient of the characteristic factor, Enter values ​​for the characteristic factor results, is the coupling coefficient of the index evaluation function, is the bias vector of the index evaluation function, The data dimension value for the input value of the characteristic factor result; Initialize the hyperparameters and learning rate of the initial model, set the loss function of the initial model, and use the SISoU loss function as the loss function; The loss function expression is as follows: (5) in, is the model loss rate, is the comprehensive evaluation score of coal quality, is the comprehensive real score of coal quality, is the number of characteristic factors, is the gradient of the SISoU loss function, is the momentum hyperparameter of the SISoU loss function.

[0009] Preferably, the method of iteratively training the coal quality index quantitative model to convergence using the training set as input specifically further includes: Taking the training set as input, setting the number of iterative training of the initial model, iteratively training the quantitative model of coal quality indicators, and ending the training when the loss function is at the minimum value; Use cross-validation to verify the comprehensive evaluation effect of the initial model. If the preset effect threshold is reached, the model is judged to be converged. The accuracy of the initial model is verified by the validation set. If the preset effect threshold is not reached, the Adam optimizer is used to iteratively update the hyperparameters and learning rate. Obtain a validation set, preset an accuracy threshold, and use the validation set to verify whether the model meets the preset detection accuracy. If it meets the detection accuracy, output the trained coal quality index quantitative model.

[0010] Preferably, the method in which the coal quality index quantitative model analyzes the real-time data set to obtain the first detection result and the second detection result specifically includes: The real-time data set is input into the indicator input layer, and the feature convolution architecture performs dilated convolution on the real-time data set to obtain the dilated convolution result; (6) in, Enter values ​​for the real-time dataset, is the convolution kernel size of the feature convolution architecture, is the expanded receptive field of the feature convolution architecture, is the expansion factor; (7) Obtain the dilated convolution result, and the residual connection network performs residual connection on the dilated convolution result to obtain the residual connection result; The residual connection result is loaded, and the feature decoder decodes the feature factor based on the support vector machine to obtain the feature factor result, and the feature factor result is output as the first detection result; Obtain a first detection result, define an indicator evaluator to evaluate the first detection result based on an indicator evaluation function, and obtain a second detection result; The indicator evaluation function is defined as: (4) in, The comprehensive evaluation score of coal quality is taken as the second test result. is the number of characteristic factors, is the model activation function, is the weight coefficient of the characteristic factor, Enter values ​​for the characteristic factor results, is the coupling coefficient of the index evaluation function, is the bias vector of the index evaluation function, The data dimension value for the input value of the characteristic factor result; (8) in, They are the hyper parameters of the quantitative model of coal quality indicators; (9) is the weight coefficient of the characteristic factor, Enter values ​​for the characteristic factor results, is the weight penalty factor.

[0011] The present invention also provides a mobile platform coal quality remote detection system, the mobile platform coal quality remote detection system specifically comprising: Mobile coal quality testing equipment, used for real-time testing of stored coal and obtaining real-time spectral data, wherein the real-time testing is based on the dual-pulse laser technology to enhance the laser pulses emitted by the laser induced breakdown spectrometer; A preprocessing module is used to pre-build a data processing model based on adaptive interactive multi-model combined with deep learning, and pre-process the real-time spectral data based on the data processing model to obtain a real-time data set; A model building module is used to obtain spectral sample data, input the spectral sample data into a pre-built data processing model, and the data processing model pre-processes the spectral sample data to obtain a sample data set, which is divided into a training set and a validation set. A coal quality index quantitative model based on a convolutional neural network combined with a lightweight detection structure is established. The training set is used as input, and the coal quality index quantitative model is iteratively trained until convergence. The validation set is used to verify whether the model meets the preset detection accuracy. If it meets the detection accuracy, the trained coal quality index quantitative model is output; The coal quality detection module takes the real-time data set as input and executes the coal quality index quantitative model. The coal quality index quantitative model analyzes the real-time data set to obtain a first detection result and a second detection result, wherein the first detection result includes a coal particle size detection result, a coal moisture detection result, a coal calorific value detection result, a coal volatile matter detection result, and a coal element detection result.

[0012] Preferably, the mobile coal quality detection equipment comprises: A mobile base, wherein a plurality of groups of mobile rollers are installed at the bottom of the mobile base, an equipment stand is fixedly installed on the mobile base, and a rotating drive unit is arranged on the equipment stand; A remote detection mechanism connected to the rotary drive unit, the remote detection mechanism being used for remote detection of coal quality; Wherein, the remote detection mechanism includes: A mechanism mounting seat, the mechanism mounting seat being connected to the rotation driving part; An environmental monitoring camera, which is used to obtain coal storage environment data and photograph the coal surface; A coal quality analyzer, which emits laser pulses in real time and collects real-time spectral data based on laser induced breakdown spectroscopy technology; a synchronous adjustment unit, the synchronous adjustment unit being arranged on the mechanism mounting seat, the synchronous adjustment unit being connected to the environment monitoring camera and the coal quality analyzer respectively, and being used for synchronously adjusting the tilt angles of the environment monitoring camera and the coal quality analyzer, and An auxiliary cleaning part, wherein the auxiliary cleaning part is arranged on the mechanism mounting seat and is used for auxiliary cleaning of the coal quality analyzer.

[0013] Preferably, the synchronization adjustment unit includes: An adjusting motor, wherein the adjusting motor is fixedly mounted on the mechanism mounting seat; A linkage gear fixedly connected to the output end of the regulating motor, a monitoring support seat is arranged above the linkage gear, and the bottom of the monitoring support seat is fixedly connected to the linkage gear through a connecting rod; A driven gear, the driven gear is arranged between the monitoring support seat and the linkage gear, the driven gear is rotatably connected to the upper surface of the mechanism mounting seat, and the coal quality analyzer is detachably connected to the outer side of the driven gear; The coal quality analyzer comprises: An analyzer mainframe, the analyzer mainframe being detachably connected to a side wall of the driven gear; A laser transmitter mounted on the side wall of the analyzer mainframe, and At least one group of spectral response detectors is installed on the side wall of the coal quality analyzer, and the spectral response detectors are used to capture the emission spectrum of coal plasma.

[0014] Preferably, the auxiliary cleaning unit comprises: A linkage round seat, wherein the linkage round seat is detachably connected to the side wall of the analyzer mainframe, and an arc-shaped driving groove is provided on the side wall of the linkage round seat, and an arc-shaped driving block is slidably embedded in the arc-shaped driving groove; A linkage swing rod fixedly connected to the arc-shaped driving block, the linkage swing rod is rotatably mounted on the cleaning limit seat, the linkage round seat is rotatably mounted on the cleaning limit seat, and the cleaning limit seat is fixedly mounted on the mechanism mounting seat; An arc gear, the arc gear is fixedly connected to the linkage swing rod, a linkage rack is provided on one side of the arc gear, the linkage rack is meshed with the arc gear, one end of the linkage rack is fixedly connected to the cleaning fan, and the linkage rack is slidably connected to the cleaning limit seat; A return spring rod is fixedly mounted on the cleaning limit seat, and one end of the return spring rod is fixedly connected to a linkage rack.

[0015] Compared with the prior art, the embodiments of the present application have the following beneficial effects: In an embodiment of the present invention, by pre-constructing a data processing model based on an adaptive interactive multi-model combined with deep learning, real-time spectral data can be pre-processed for denoising, normalization and feature extraction, thereby reducing the spectral data acquisition error rate, reducing the computational complexity and memory usage of the model, and ensuring the analysis accuracy and efficiency of the coal quality index quantitative model. By establishing a coal quality index quantitative model based on a convolutional neural network combined with a lightweight detection structure, the complementarity between the convolutional neural network and the lightweight detection structure MobileNet is utilized to improve the overall prediction accuracy and robustness, so that the contribution and weight of the characteristic factors to the influence of coal quality can be judged, thereby maximizing the prediction accuracy of the model.

[0016] In the embodiment of the present invention, by pre-constructing a data processing model including a normalization processing layer, a background noise reduction layer, a feature extraction layer and a baseline correction layer, background noise reduction processing can be performed on real-time spectral data, and spectral characteristic factors can be effectively extracted to achieve effective fusion of multi-level spectral characteristic factors to improve the expression ability of features, thereby overcoming the problem that the prior art cannot effectively eliminate background interference and outlier removal of real-time spectral data.

[0017] In an embodiment of the present invention, a coal quality index quantitative model and a training method thereof are provided. The coal quality index quantitative model consists of an index input layer, a feature convolution architecture, a pooling layer, a feature decoder, and an index evaluator. A lightweight detection structure MobileNet is introduced into the feature convolution architecture of the coal quality index quantitative model, and a residual connection network is used to replace the pooling layer. An adaptive interactive algorithm based on data mining technology is developed, and a variety of machine learning algorithms are used in combination to adapt to different data distribution situations to improve the prediction accuracy of coal detection indicators.

[0018] An embodiment of the present invention provides a mobile platform coal quality remote detection system, which consists of a mobile coal quality detection device, a preprocessing module, a model building module, and a coal quality detection module. The mobile platform coal quality remote detection system can meet the needs of the coal transportation market, improve the loading and unloading and transportation efficiency of coal railway platforms, realize the interconnection of transportation, unloading, detection and loading, and improve the overall transportation efficiency of coal.

[0019] In an embodiment of the present invention, a remote detection mechanism is provided that can synchronously emit laser pulses in real time and collect real-time spectral data. The remote detection mechanism is composed of an environmental monitoring camera, a coal quality analyzer, a synchronous adjustment unit, and an auxiliary cleaning unit. The synchronous adjustment of the environmental monitoring camera and the coal quality analyzer can be achieved simply by starting the synchronous adjustment unit. Compared with the prior art, the detection range is larger while ensuring the accuracy of coal quality analysis. The auxiliary cleaning unit can also work in conjunction with the coal quality analyzer. When coal is stored, excessive dust content in the environment interferes with data collection by the coal quality analyzer. The coal quality analyzer is synchronously cleaned, thereby further ensuring the accuracy of coal quality analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic diagram of the implementation process of the mobile platform coal quality remote detection method provided by the present invention.

[0021] Figure 2 It is a schematic diagram of the implementation flow of the method for preprocessing real-time spectral data based on the data processing model provided by the present invention.

[0022] Figure 3 It is a schematic diagram of the implementation flow of the method provided by the present invention for iteratively training the quantitative model of coal quality indicators until convergence using the training set as input.

[0023] Figure 4 It is a schematic diagram of the implementation process of a method for analyzing a real-time data set using the coal quality index quantitative model provided by the present invention to obtain a first detection result and a second detection result.

[0024] Figure 5 It is a control block diagram of the mobile platform coal quality remote detection system provided by the present invention.

[0025] Figure 6 It is a structural schematic diagram of the mobile coal quality detection equipment provided by the present invention.

[0026] Figure 7 It is a three-dimensional structural schematic diagram of the mobile coal quality detection equipment provided by the present invention.

[0027] Figure 8 It is a top view of the mobile coal quality detection equipment provided by the present invention.

[0028] Fig. 9 It is a structural schematic diagram of the remote detection mechanism provided by the present invention.

[0029] Fig.10 It is an axonometric diagram of the remote detection mechanism provided by the present invention.

[0030] Fig.11 It is a front view of the remote detection mechanism provided by the present invention.

[0031] In the figure: 10-mobile coal quality detection equipment, 11-mobile base, 12-equipment stand, 13-rotation drive unit, 131-rotation motor, 132-rotation gear, 133-side gear, 134-rotation support arm, 2-remote detection mechanism, 21-synchronous adjustment unit, 211-adjustment motor, 212-linkage gear, 213-monitoring support seat, 214-driven gear, 22-environmental monitoring camera, 23-coal quality analyzer, 231-analyzer Main unit, 232-laser transmitter, 233-spectral response detector, 24-auxiliary cleaning unit, 241-linked round seat, 242-arc-shaped driving groove, 243-arc-shaped driving block, 244-linked swing rod, 245-cleaning limit seat, 246-arc-shaped gear, 247-linked rack, 248-cleaning fan, 249-reset spring rod, 25-mechanism mounting seat, 20-pretreatment module, 30-model building module, 40-coal quality detection module. DETAILED DESCRIPTION

[0032] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of this application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0033] The existing methods cannot perform denoising, normalization and feature extraction preprocessing on the parameters of the coal powder when establishing a model between spectral lines and calorific value and obtaining the coefficient matrix and error terms of the model, so that the error rate of spectral data acquisition affects the analysis accuracy and efficiency of the model. To address the above problems, we propose a mobile platform coal quality remote detection system and method. In short, the method includes using a mobile coal quality detection device 10 to perform real-time detection on the stored coal to obtain real-time spectral data, and then pre-constructing a data processing model based on an adaptive interactive multi-model combined with deep learning, and establishing a coal quality index quantitative model based on a convolutional neural network combined with a lightweight detection structure. Finally, the real-time data set is used as input to execute the coal quality index quantitative model, and the coal quality index quantitative model analyzes the real-time data set to obtain a first detection result and a second detection result. In an embodiment of the present invention, by pre-constructing a data processing model based on an adaptive interactive multi-model combined with deep learning, real-time spectral data can be pre-processed for denoising, normalization and feature extraction, thereby reducing the spectral data acquisition error rate, reducing the computational complexity and memory usage of the model, and ensuring the analysis accuracy and efficiency of the coal quality index quantitative model. By establishing a coal quality index quantitative model based on a convolutional neural network combined with a lightweight detection structure, the complementarity between the convolutional neural network and the lightweight detection structure MobileNet is utilized to improve the overall prediction accuracy and robustness, so that the contribution and weight of the characteristic factors to the influence of coal quality can be judged, thereby maximizing the prediction accuracy of the model.

[0034] The embodiment of the present invention provides a mobile platform coal quality remote detection method, Figure 1 The schematic diagram of the implementation process of the mobile platform coal quality remote detection method is shown. The mobile platform coal quality remote detection method specifically includes: Step S10, using a mobile coal quality detection device 10 to detect the stored coal in real time to obtain real-time spectral data; It should be noted that with the increase of detection distance, the energy of the laser pulse emitted by the mobile coal quality detection equipment 10 will attenuate to a certain extent, and the laser focusing spot at a long distance will become larger and larger, and the laser power density will drop sharply, thereby affecting the generated plasma density, temperature and light intensity, etc., and the collection solid angle of the plasma spectrum will decrease with the square of the detection distance. These factors limit the detection distance of remote LIBS. In order to enable the remote LIBS system to achieve higher sensitivity and higher precision detection, the laser pulse emitted by the laser induced breakdown spectrometer is enhanced based on the dual-pulse laser technology during real-time detection in the embodiment of the present invention.

[0035] It should be noted that laser induced breakdown spectroscopy (LIBS) technology is a technology that measures coal without contact to realize coal quality detection and analysis. It uses high-energy laser pulses to excite coal into plasma, and determines the content of each element in the coal by analyzing the spectrum emitted by the plasma, thereby realizing rapid, real-time and non-contact measurement of coal.

[0036] Step S20, pre-constructing a data processing model based on adaptive interactive multi-model combined with deep learning, and pre-processing the real-time spectral data based on the data processing model to obtain a real-time data set; Step S30, obtaining spectral sample data, inputting the spectral sample data into a pre-built data processing model, the data processing model pre-processing the spectral sample data to obtain a sample data set, dividing the sample data set into a training set and a validation set, establishing a coal quality index quantitative model based on a convolutional neural network combined with a lightweight detection structure, taking the training set as input, iteratively training the coal quality index quantitative model until convergence, using the validation set to verify whether the model meets the preset detection accuracy, and if it meets the detection accuracy, outputting the trained coal quality index quantitative model; Step S40, using the real-time data set as input, executes the coal quality index quantitative model, the coal quality index quantitative model analyzes the real-time data set to obtain a first detection result and a second detection result, wherein the first detection result includes a coal particle size detection result, a coal moisture detection result, a coal calorific value detection result, a coal volatile matter detection result, and a coal element detection result.

[0037] In an embodiment of the present invention, by pre-constructing a data processing model based on an adaptive interactive multi-model combined with deep learning, real-time spectral data can be pre-processed for denoising, normalization and feature extraction, thereby reducing the spectral data acquisition error rate, reducing the computational complexity and memory usage of the model, and ensuring the analysis accuracy and efficiency of the coal quality index quantitative model. By establishing a coal quality index quantitative model based on a convolutional neural network combined with a lightweight detection structure, the complementarity between the convolutional neural network and the lightweight detection structure MobileNet is utilized to improve the overall prediction accuracy and robustness, so that the contribution and weight of the characteristic factors to the influence of coal quality can be judged, thereby maximizing the prediction accuracy of the model.

[0038] In the embodiment of the present invention, the data processing model includes a preprocessing module 20, a data correction layer and a preprocessing output layer. The preprocessing module 20 includes a normalization processing layer, a background noise reduction layer, a feature extraction layer and a baseline correction layer.

[0039] It should be noted that the normalization processing layer is connected to the background noise reduction layer, the background noise reduction layer is connected to the feature extraction layer, and the feature extraction layer is connected to the baseline correction layer, and the normalization processing layer includes an embedding Embedding, the number of channels of the normalization processing layer is 64, and the loss function of the data processing model is the loss function SIou. When the data processing model is trained, the normalization processing layer, background noise reduction layer, feature extraction layer and baseline correction layer of the model are subjected to ablation tests and compared with the Pareto analysis model and RFM model in the prior art.

[0040] The normalization processing layer is used to standardize the real-time spectral data and eliminate the factors affecting the optical path and data volume. The background noise reduction layer is used to perform background noise reduction on the real-time spectral data. The feature extraction layer is used to identify and extract spectral characteristic factors reflecting the quality of coal. The baseline correction layer is used to correct the spectral baseline. When pre-constructing a data processing model based on adaptive interactive multi-model combined with deep learning, the feature extraction layer is improved by combining the adaptive interactive multi-model and the PeleeNet detection structure, and the feature interaction function Glou is set to construct a feature extraction improvement layer.

[0041] In the embodiment of the present invention, by pre-constructing a data processing model including a normalization processing layer, a background noise reduction layer, a feature extraction layer and a baseline correction layer, background noise reduction processing can be performed on real-time spectral data, and spectral characteristic factors can be effectively extracted, so as to achieve effective fusion of multi-level spectral characteristic factors to improve the expression ability of features, thereby overcoming the problem that the prior art cannot effectively eliminate background interference of real-time spectral data and eliminate outliers, so that the embodiment of the present invention shows significant advantages in the field of coal measurement.

[0042] The embodiment of the present invention provides a method for preprocessing real-time spectral data based on a data processing model. Figure 2 The schematic diagram of the implementation flow of the method for preprocessing real-time spectral data based on the data processing model is shown. The method for preprocessing real-time spectral data based on the data processing model specifically includes: Step S101, loading real-time spectral data, and normalizing the real-time spectral data, wherein the normalization process uses Z-score normalization combined with vector normalization to normalize the data to obtain a normalized data set; It should be noted that the use of Z-score standardization combined with vector normalization to normalize data can not only effectively eliminate the dimension effect, thereby improving the model convergence speed when training the quantitative model of coal quality indicators, but also adapt to the needs of different algorithms, enhance the robustness of the quantitative model of coal quality indicators, simplify the data processing process, and is widely used in various scenarios.

[0043] Step S102, obtaining a normalized data set, presetting a wavelet denoising function and a wavelet basis threshold, and using the wavelet denoising function to perform denoising on the normalized data set to obtain a denoised output set; Among them, the wavelet denoising function is expressed as: (1) in, is the normalized input data set, is the mean variance of the normalized data set, is the normalization coefficient. In this embodiment, the normalization coefficient can be set to 2-6. is the wavelet coefficient, the wavelet coefficient is set to 1 or 2, is the number of wavelet basis decomposition layers, the number of wavelet basis decomposition layers is 6-10, is the data volume of the dataset; Step S103, loading the denoised output set, and performing hybrid enhancement on the denoised output set using a Mosaic-Mixup hybrid enhancement algorithm to obtain a hybrid enhancement set; (2) is the mixed enhancement value, is the mixing coefficient of the Mosaic-Mixup hybrid enhancement algorithm, is a batch of data of the denoised output set collected by stratified sampling, Another batch of data for the denoising output set; (3) in, is a mixing parameter. In this embodiment, the mixing parameter is set to 0.25-0.3; Step S104, obtaining a mixed enhancement set, and performing background noise reduction on the mixed enhancement set using a singular value denoising algorithm to obtain a background noise reduction set; It should be noted that the basic idea of ​​using the singular value denoising algorithm to perform background noise reduction on the mixed enhancement set is to represent the mixed enhancement set in matrix form and then perform SVD decomposition on the matrix. The singular values ​​obtained after decomposition represent the energy of each component in the mixed enhancement set. Larger singular values ​​correspond to important components in the mixed enhancement set, while smaller singular values ​​mainly correspond to noise components.

[0044] Step S105, taking the background noise reduction set as input, improving the feature extraction layer to extract features from the background noise reduction set to obtain a feature extraction set, and combining the adaptive Kalman filter algorithm to correct the element characteristic spectral lines and spectral baselines to output a real-time data set.

[0045] In this embodiment, in order to extract the characteristic spectral lines of elements and separate the overlapping spectral lines, in addition to calibrating the wavelength of the spectrometer, an adaptive Kalman filtering algorithm is used to resolve the severely interfering spectral lines, thereby achieving abnormal data elimination, spectral noise removal and characteristic data extraction.

[0046] The embodiment of the present invention provides a method for iteratively training a coal quality index quantitative model until convergence using a training set as input. Figure 3 The schematic diagram of the implementation process of the method for iteratively training the quantitative model of coal quality indicators to convergence using the training set as input is shown. The method for iteratively training the quantitative model of coal quality indicators to convergence using the training set as input specifically includes: Step S201, pre-constructing a coal quality index quantitative model based on a convolutional neural network combined with a lightweight detection structure, and setting an initial model of the coal quality index quantitative model, wherein the initial model of the coal quality index quantitative model includes an index input layer, a feature convolution architecture, a pooling layer, a feature decoder, and an index evaluator; It should be noted that the initial model can be a convolutional neural network model or a support vector machine model, and the data processing model preprocesses the spectral sample data in the same way as the data processing model preprocesses the real-time spectral data, and the sample data set is divided into a training set and a validation set, and the distribution ratio of the training set to the validation set can be 7:3, and the sample data set can obtain public coal quality spectral analysis data from China Coal Resources Network, the U.S. Energy Information Administration (EIA), etc. These databases usually contain a large amount of coal sample data, which can be compared and analyzed.

[0047] Step S202, a lightweight detection structure MobileNet is introduced into the feature convolution architecture, and a residual connection network is used to replace the pooling layer. The improved feature convolution architecture includes three layers of convolution and a MobileNet layer. The MobileNet layer contains 128 neurons and is combined with a ReLU activation function to define an indicator evaluation function of the indicator evaluator; It should be noted that the three convolution layers in the feature convolution architecture are divided into the first convolution layer, the second convolution layer, and the third convolution layer. The convolution kernel size of the first convolution layer, the second convolution layer, and the third convolution layer is 3*3, the number of channels of the first convolution layer = 64, the number of channels of the second convolution layer = 128, and the number of channels of the third convolution layer = 256.

[0048] Among them, the indicator evaluation function is defined as: (4) in, is the comprehensive evaluation score of coal quality, is the number of characteristic factors. In this embodiment, the characteristic factors include but are not limited to coal particle size, coal moisture, coal calorific value, coal volatile matter, and coal elements. The number of characteristic factors is 5-15. is the model activation function, is the weight coefficient of the characteristic factor, Enter values ​​for the characteristic factor results, is the coupling coefficient of the index evaluation function. In this embodiment, the coupling coefficient of the index evaluation function is set to 0.1-0.5. is the bias vector of the index evaluation function, The data dimension value for the characteristic factor result input value is set to 1-8; Step S203, initializing the hyperparameters and learning rate of the initial model, setting the loss function of the initial model, and the loss function is the SISoU loss function; The loss function expression is as follows: (5) in, is the model loss rate, is the comprehensive evaluation score of coal quality, is the comprehensive real score of coal quality, is the number of characteristic factors, is the gradient of the SISoU loss function, is the momentum hyperparameter of the SISoU loss function, set to 0.85.

[0049] Step S204, taking the training set as input, setting the number of iterative training of the initial model, iteratively training the quantitative model of coal quality indicators, and ending the training when the loss function reaches the minimum value; In this embodiment, the number of iterative training is set to 100-500 times.

[0050] Step S205, verifying the comprehensive evaluation effect of the initial model by cross-validation; Step S206: if the preset effect threshold is reached, the model is judged to be converged, and the initial model accuracy is verified by the validation set; if the preset effect threshold is not reached, the Adam optimizer is used to iteratively update the hyperparameters and learning rate, and the process returns to step S204; It should be noted that the preset effect threshold is set to 0.95-0.98.

[0051] Step S207, obtaining a validation set, setting a precision threshold, and using the validation set to verify whether the model meets the preset detection precision. In this embodiment, the precision threshold is set to 0.95-1; Step S208, if the detection accuracy is met, output the trained coal quality index quantitative model; if the detection accuracy is not met, return to step S204.

[0052] In an embodiment of the present invention, a coal quality index quantitative model and a training method thereof are provided. The coal quality index quantitative model consists of an index input layer, a feature convolution architecture, a pooling layer, a feature decoder, and an index evaluator. A lightweight detection structure MobileNet is introduced into the feature convolution architecture of the coal quality index quantitative model, and a residual connection network is used to replace the pooling layer. An adaptive interactive algorithm based on data mining technology is developed, and a variety of machine learning algorithms are used in combination to adapt to different data distribution situations to improve the prediction accuracy of coal detection indicators.

[0053] The embodiment of the present invention provides a method for analyzing a real-time data set using a coal quality index quantitative model to obtain a first detection result and a second detection result. Figure 4 The present invention shows a schematic diagram of the implementation process of the method for obtaining the first detection result and the second detection result by analyzing the real-time data set by the quantitative model of coal quality indicators. The method for obtaining the first detection result and the second detection result by analyzing the real-time data set by the quantitative model of coal quality indicators specifically includes: Step S301, input the real-time data set into the indicator input layer, and the feature convolution architecture performs dilated convolution on the real-time data set to obtain a dilated convolution result; (6) in, Enter values ​​for the real-time dataset, is the convolution kernel size of the feature convolution architecture, is the expanded receptive field of the feature convolution architecture, is the expansion factor; (7) Step S302, obtaining the dilated convolution result, and performing residual connection on the dilated convolution result by the residual connection network to obtain a residual connection result; Step S303, loading the residual connection result, the feature decoder decodes the feature factor based on the support vector machine to obtain the feature factor result, and outputs the feature factor result as the first detection result; Step S304, obtaining the first detection result, defining an indicator evaluator to evaluate the first detection result based on an indicator evaluation function, and obtaining a second detection result; The indicator evaluation function is defined as: (4) in, The comprehensive evaluation score of coal quality is taken as the second test result. is the number of characteristic factors, is the model activation function, is the weight coefficient of the characteristic factor, Enter values ​​for the characteristic factor results, is the coupling coefficient of the index evaluation function, is the bias vector of the index evaluation function, The data dimension value for the input value of the characteristic factor result; (8) in, They are the hyperparameters of the coal quality index quantitative model, and the hyperparameters of the coal quality index quantitative model are set to 0.1-0.3; (9) is the weight coefficient of the characteristic factor, Enter values ​​for the characteristic factor results, is a weight penalty factor. In this embodiment, the weight penalty factor is set to 0.1-1.

[0054] The embodiment of the present invention also provides a mobile platform coal quality remote detection system, Figure 5 The control block diagram of the mobile platform coal quality remote detection system is shown. The mobile platform coal quality remote detection system specifically includes: The mobile coal quality detection device 10 is used for real-time detection of stored coal and acquisition of real-time spectral data, wherein the real-time detection is based on the dual-pulse laser technology to enhance the laser pulses emitted by the laser induced breakdown spectrometer; A preprocessing module 20 is used to pre-build a data processing model based on an adaptive interactive multi-model combined with deep learning, and pre-process the real-time spectral data based on the data processing model to obtain a real-time data set; The model building module 30 is used to obtain spectral sample data, input the spectral sample data into a pre-built data processing model, and the data processing model pre-processes the spectral sample data to obtain a sample data set, divides the sample data set into a training set and a validation set, and establishes a coal quality index quantitative model based on a convolutional neural network combined with a lightweight detection structure. The training set is used as input, and the coal quality index quantitative model is iteratively trained until convergence. The validation set is used to verify whether the model meets the preset detection accuracy. If it meets the detection accuracy, the trained coal quality index quantitative model is output; The coal quality detection module 40 takes the real-time data set as input and executes the coal quality index quantitative model. The coal quality index quantitative model analyzes the real-time data set to obtain a first detection result and a second detection result, wherein the first detection result includes a coal particle size detection result, a coal moisture detection result, a coal calorific value detection result, a coal volatile matter detection result, and a coal element detection result.

[0055] An embodiment of the present invention provides a mobile platform coal quality remote detection system, which consists of a mobile coal quality detection device 10, a preprocessing module 20, a model building module 30, and a coal quality detection module 40. The mobile platform coal quality remote detection system can meet the needs of the coal transportation market, improve the loading and unloading and transportation efficiency of coal railway platforms, realize the interconnection of transportation, unloading, detection and loading, and improve the overall transportation efficiency of coal.

[0056] In a further preferred embodiment of the present invention, Figure 6-Figure 8 As shown, the mobile coal quality detection device 10 includes: A mobile base 11, wherein a plurality of groups of mobile rollers are installed at the bottom of the mobile base 11, a device stand 12 is fixedly installed on the mobile base 11, and a rotation driving part 13 is arranged on the device stand 12; In this embodiment, the movable base 11 can be a rectangular base or a round base, and the movable base 11 is rotatably connected to movable rollers through bearings or rollers, the number of movable rollers is four groups, and the bottom of the equipment stand 12 is fixedly connected to the upper surface of the movable base 11 by plugging or riveting.

[0057] It should be noted that the rotation driving unit 13 includes: Rotating motor 131; A rotating gear 132 connected to the output end of the rotating motor 131 by interference fit; The side gear 133 installed on the top of the equipment stand 12 is rotatably engaged with the rotating gear 132 for transmission. A rotating support arm 134 is fixedly installed on the upper part of the side gear 133 , and the top of the rotating support arm 134 is connected to the remote detection mechanism 2 .

[0058] In this embodiment, the rotating motor 131 can be a servo motor, and the rotating motor 131 is fixedly installed on the top of the equipment stand 12 by fastening bolts or clamps. When the position of the remote detection mechanism 2 needs to be adjusted, the rotating motor 131 is turned on. The starting of the rotating motor 131 can drive the rotating gear 132 to rotate, so that the rotating gear 132 drives the side gear 133, the rotating support arm 134, and the remote detection mechanism 2 to rotate, which is beneficial to the detection and analysis of coal at different positions.

[0059] A remote detection mechanism 2 connected to the rotary drive unit 13, wherein the remote detection mechanism 2 is used for remotely detecting coal quality; Among them, Figure 9-11 As shown, the remote detection mechanism 2 includes: A mechanism mounting seat 25, wherein the mechanism mounting seat 25 is connected to the rotation driving unit 13; The mechanism mounting seat 25 can be a rectangular seat or a “T”-shaped seat, and the mechanism mounting seat 25 is plugged or riveted to the top of the rotating support arm 134 .

[0060] An environmental monitoring camera 22, which is used to obtain coal storage environment data and photograph the coal surface; It should be noted that the environment monitoring cameras 22 may be CCD cameras, and the number of the environment monitoring cameras 22 is set to 3-6 groups.

[0061] A coal quality analyzer 23, wherein the coal quality analyzer 23 emits laser pulses in real time based on laser induced breakdown spectroscopy technology and collects real-time spectral data; a synchronous adjustment unit 21, the synchronous adjustment unit 21 being arranged on the mechanism mounting seat 25, and the synchronous adjustment unit 21 being connected to the environment monitoring camera 22 and the coal quality analyzer 23 respectively, and being used for synchronously adjusting the tilt angles of the environment monitoring camera 22 and the coal quality analyzer 23, and The auxiliary cleaning part 24 is disposed on the mechanism mounting seat 25 , and is used for auxiliary cleaning of the coal quality analyzer 23 .

[0062] In an embodiment of the present invention, a remote detection mechanism 2 is provided that can synchronously emit laser pulses in real time and collect real-time spectral data. The remote detection mechanism 2 is composed of an environmental monitoring camera 22, a coal quality analyzer 23, a synchronous adjustment unit 21, and an auxiliary cleaning unit 24. The synchronous adjustment of the environmental monitoring camera 22 and the coal quality analyzer 23 can be achieved by simply starting the synchronous adjustment unit 21. Compared with the prior art, the detection range is larger, and the accuracy of the coal quality analysis is ensured. The auxiliary cleaning unit 24 can also work in conjunction with the coal quality analyzer 23. When the dust content in the environment is too high during coal storage, which interferes with the data collection of the coal quality analyzer 23, the coal quality analyzer 23 is synchronously cleaned, thereby further ensuring the accuracy of the coal quality analysis.

[0063] In a further preferred embodiment of the present invention, Figure 9-10 As shown, the synchronization adjustment unit 21 includes: An adjusting motor 211, wherein the adjusting motor 211 is fixedly mounted on the mechanism mounting seat 25; A linkage gear 212 fixedly connected to the output end of the regulating motor 211, a monitoring support seat 213 is arranged above the linkage gear 212, and the bottom of the monitoring support seat 213 is fixedly connected to the linkage gear 212 through a connecting rod; The driven gear 214 is disposed between the monitoring support seat 213 and the linkage gear 212 . The driven gear 214 is rotatably connected to the upper surface of the mechanism mounting seat 25 , and the coal quality analyzer 23 is detachably connected to the outer side of the driven gear 214 .

[0064] In this embodiment, the adjusting motor 211 can be a single-phase asynchronous motor, and the adjusting motor 211 is fixedly installed in the mechanism mounting seat 25 by fastening bolts. The linkage gear 212 is rotatably connected to the upper surface of the mechanism mounting seat 25 through a bearing, and two or three groups of arc-shaped tooth groups are symmetrically arranged on the linkage gear 212. The arc-shaped tooth groups are meshed with the driven gear 214 for transmission, and the monitoring support seat 213 is fixedly connected to the environmental monitoring camera 22.

[0065] The coal quality analyzer 23 comprises: An analyzer mainframe 231, wherein the analyzer mainframe 231 is detachably connected to a side wall of the driven gear 214; A laser emitter 232 mounted on the side wall of the analyzer mainframe 231, and At least one group of spectral response detectors 233 is installed on the side wall of the coal quality analyzer 23, and the spectral response detectors 233 are used to capture the emission spectrum of coal plasma.

[0066] In this embodiment, the laser emitter 232 and the spectral response detector 233 are installed on the analyzer host 231 by fastening bolts or clamping. The spectral response detectors 233 are evenly distributed on the analyzer host 231 in a matrix or circumferentially. The number of spectral response detectors 233 can be 3-8 groups.

[0067] During measurement, the adjusting motor 211 is turned on. The starting of the adjusting motor 211 can drive the linkage gear 212 and the monitoring support seat 213 to rotate, so that the monitoring support seat 213 drives the environmental monitoring camera 22 to rotate, realizing variable range monitoring of the environmental monitoring camera 22. The rotation of the linkage gear 212 can drive the driven gear 214 to rotate, and then the driven gear 214 drives the analyzer host 231, the laser emitter 232, and the spectral response detector 233 to flip, so that the analyzer host 231, the laser emitter 232, and the spectral response detector 233 can perform variable range detection on coal.

[0068] In a further preferred embodiment of the present invention, Figure 9-10 As shown, the auxiliary cleaning part 24 includes: A linkage round seat 241, wherein the linkage round seat 241 is detachably connected to the side wall of the analyzer mainframe 231, and an arc-shaped driving groove 242 is provided on the side wall of the linkage round seat 241, and an arc-shaped driving block 243 is slidably embedded in the arc-shaped driving groove 242; In this embodiment, the end of the linkage round seat 241 is connected to the analyzer host 231 by plugging or snapping, the inner wall of the arc driving groove 242 is polished, and the arc driving block 243 can be a cylindrical block or a spherical block.

[0069] A linkage swing rod 244 fixedly connected to the arc-shaped driving block 243, the linkage swing rod 244 is rotatably mounted on the cleaning limit seat 245, the linkage round seat 241 is rotatably mounted on the cleaning limit seat 245, and the cleaning limit seat 245 is fixedly mounted on the mechanism mounting seat 25; It should be noted that the middle portion of the linkage swing rod 244 is rotatably connected to the cleaning limit seat 245 via a bearing, and the cleaning limit seat 245 is fixedly connected to the mechanism mounting seat 25 by means of clamping or riveting.

[0070] An arc gear 246, wherein the arc gear 246 is fixedly connected to the linkage swing rod 244, a linkage rack 247 is provided on one side of the arc gear 246, the linkage rack 247 is meshedly connected with the arc gear 246, one end of the linkage rack 247 is fixedly connected to a cleaning fan 248, and the linkage rack 247 is slidably connected to the cleaning limit seat 245; The return spring rod 249 is fixedly mounted on the cleaning limit seat 245 , and one end of the return spring rod 249 is fixedly connected to the linkage rack 247 .

[0071] In this embodiment, the arc gear 246 can be a half, one third or one quarter incomplete gear, and the end of the tilt limit seat can be provided with a rectangular rod or a square rod for guiding the linkage rack 247. The reset spring rod 249 is installed on the cleaning limit seat 245 by a clip or a bolt, and the telescopic end of the reset spring rod 249 is connected to the linkage rack 247 by a clamp or a bolt.

[0072] During operation, when the analyzer mainframe 231 rotates, it can drive the linkage round seat 241 to rotate, and the linkage round seat 241 drives the arc driving groove 242 to rotate, so that the arc driving groove 242 drives the arc driving block 243 and the linkage swing rod 244 to swing, and when the linkage swing rod 244 swings, it drives the arc gear 246 to swing, and then the arc gear 246 drives the linkage rack 247 and the cleaning fan 248 to intermittently translate to the front of the analyzer mainframe 231, the laser emitter 232, and the spectral response detector 233, so as to clean the surface of the analyzer mainframe 231, the laser emitter 232, and the spectral response detector 233.

[0073] In summary, the present invention provides a mobile platform coal quality remote detection system and method. In an embodiment of the present invention, by pre-constructing a data processing model based on an adaptive interactive multi-model combined with deep learning, real-time spectral data can be denoised, normalized and pre-processed with feature extraction, thereby reducing the spectral data acquisition error rate, reducing the computational complexity and memory occupancy of the model, while ensuring the analysis accuracy and efficiency of the coal quality index quantitative model. In addition, by establishing a coal quality index quantitative model based on a convolutional neural network combined with a lightweight detection structure, the complementarity between the convolutional neural network and the lightweight detection structure MobileNet is utilized to improve the overall prediction accuracy and robustness, so that the contribution and weight of the characteristic factors to the influence of coal quality can be judged, thereby maximizing the prediction accuracy of the model.

[0074] It should be noted that, for the above-mentioned embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described order of actions, because according to the present invention, some steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0075] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on these embodiments, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in this field can still combine, add, delete or make other adjustments to the features in the various embodiments of the present invention according to the circumstances without conflict, without making creative work, so as to obtain different other technical solutions that do not deviate from the concept of the present invention in essence, and these technical solutions also belong to the scope of protection of the present invention.

Claims

1. A mobile platform coal quality remote detection method, characterized in that: The mobile platform coal quality remote detection method comprises: Use mobile coal quality testing equipment to test stored coal in real time and obtain real-time spectral data; Pre-build a data processing model based on adaptive interactive multi-model combined with deep learning, and pre-process the real-time spectral data based on the data processing model to obtain a real-time data set; Acquire spectral sample data, input the spectral sample data into a pre-built data processing model, the data processing model pre-processes the spectral sample data to obtain a sample data set, divide the sample data set into a training set and a validation set, establish a coal quality index quantitative model based on a convolutional neural network combined with a lightweight detection structure, use the training set as input, iteratively train the coal quality index quantitative model until convergence, use the validation set to verify whether the model meets the preset detection accuracy, and if it meets the detection accuracy, output the trained coal quality index quantitative model; The coal quality index quantitative model is executed with the real-time data set as input. The coal quality index quantitative model analyzes the real-time data set to obtain a first detection result and a second detection result, wherein the first detection result includes a coal particle size detection result, a coal moisture detection result, a coal calorific value detection result, a coal volatile matter detection result, and a coal element detection result.

2. The mobile platform coal quality remote detection method according to claim 1, characterized in that: The data processing model includes a preprocessing module, a data correction layer and a preprocessing output layer. The preprocessing module includes a normalization processing layer, a background noise reduction layer, a feature extraction layer and a baseline correction layer. The normalization processing layer is used to standardize the real-time spectral data and eliminate the factors affecting the optical path and data volume. The background noise reduction layer is used to perform background noise reduction on the real-time spectral data. The feature extraction layer is used to identify and extract spectral characteristic factors reflecting the quality of coal. The baseline correction layer is used to correct the spectral baseline. When pre-constructing a data processing model based on adaptive interactive multi-model combined with deep learning, the feature extraction layer is improved by combining the adaptive interactive multi-model and the PeleeNet detection structure, and the feature interaction function Glou is set to construct a feature extraction improvement layer.

3. The mobile platform coal quality remote detection method according to claim 2, characterized in that: The method for preprocessing real-time spectral data based on the data processing model specifically includes: Loading real-time spectral data, and performing normalization processing on the real-time spectral data, wherein the normalization processing adopts Z-score normalization combined with vector normalization to normalize the data to obtain a normalized data set; Obtain a normalized data set, preset a wavelet denoising function and a wavelet basis threshold, use the wavelet denoising function to perform denoising on the normalized data set, and obtain a denoised output set; Among them, the wavelet denoising function is expressed as: （1） in, is the normalized input data set, is the mean variance of the normalized data set, is the normalization coefficient, is the wavelet coefficient, is the number of wavelet basis decomposition layers; Load the denoised output set, and use the Mosaic-Mixup hybrid enhancement algorithm to perform hybrid enhancement on the denoised output set to obtain a hybrid enhancement set; （2） is the mixed enhancement value, is the mixing coefficient of the Mosaic-Mixup hybrid enhancement algorithm, is a batch of data of the denoised output set collected by stratified sampling, Another batch of data for the denoising output set; （3） in, is the mixing parameter; A mixed enhancement set is obtained, and a singular value denoising algorithm is used to perform background denoising on the mixed enhancement set to obtain a background denoising set; Taking the background noise reduction set as input, the improved feature extraction layer performs feature extraction on the background noise reduction set to obtain the feature extraction set, and combines the adaptive Kalman filter algorithm to correct the element characteristic spectral lines and spectral baselines to output a real-time data set.

4. The mobile platform coal quality remote detection method according to claim 1, characterized in that: The method of iteratively training the coal quality index quantitative model until convergence using the training set as input specifically includes: Pre-build a coal quality index quantitative model based on a convolutional neural network combined with a lightweight detection structure, and set an initial model of the coal quality index quantitative model, wherein the initial model of the coal quality index quantitative model includes an index input layer, a feature convolution architecture, a pooling layer, a feature decoder, and an index evaluator; The lightweight detection structure MobileNet is introduced into the feature convolution architecture, and the residual connection network is used to replace the pooling layer. The improved feature convolution architecture includes three layers of convolution and a MobileNet layer. The MobileNet layer contains 128 neurons and is combined with a ReLU activation function. Define the index evaluation function of the index evaluator, where the index evaluation function is defined as: （4） in, is the comprehensive evaluation score of coal quality, is the number of characteristic factors, is the model activation function, is the weight coefficient of the characteristic factor, Enter values ​​for the characteristic factor results, is the coupling coefficient of the index evaluation function, is the bias vector of the index evaluation function, The data dimension value for the input value of the characteristic factor result; Initialize the hyperparameters and learning rate of the initial model, set the loss function of the initial model, and use the SISoU loss function as the loss function; The loss function expression is as follows: （5） in, is the model loss rate, is the comprehensive evaluation score of coal quality, is the comprehensive real score of coal quality, is the number of characteristic factors, is the gradient of the SISoU loss function, is the momentum hyperparameter of the SISoU loss function.

5. The mobile platform coal quality remote detection method according to claim 4, characterized in that: The method of iteratively training the coal quality index quantitative model to convergence using the training set as input specifically includes: Taking the training set as input, setting the number of iterative training of the initial model, iteratively training the quantitative model of coal quality indicators, and ending the training when the loss function is at the minimum value; Use cross-validation to verify the comprehensive evaluation effect of the initial model. If the preset effect threshold is reached, the model is judged to be converged, and the accuracy of the initial model is verified through the validation set. If the preset effect threshold is not reached, the Adam optimizer is used to iteratively update the hyperparameters and learning rate. Obtain a validation set, preset an accuracy threshold, and use the validation set to verify whether the model meets the preset detection accuracy. If it meets the detection accuracy, output the trained coal quality index quantitative model.

6. The mobile platform coal quality remote detection method according to claim 5, characterized in that: The method of using the coal quality index quantitative model to analyze the real-time data set to obtain the first detection result and the second detection result specifically includes: The real-time data set is input into the indicator input layer, and the feature convolution architecture performs dilated convolution on the real-time data set to obtain the dilated convolution result; （6） in, Enter values ​​for the real-time dataset, is the convolution kernel size of the feature convolution architecture, is the expanded receptive field of the feature convolution architecture, is the expansion factor; （7） Obtain the dilated convolution result, and the residual connection network performs residual connection on the dilated convolution result to obtain the residual connection result; The residual connection result is loaded, and the feature decoder decodes the feature factor based on the support vector machine to obtain the feature factor result, and the feature factor result is output as the first detection result; Obtain a first detection result, define an indicator evaluator to evaluate the first detection result based on an indicator evaluation function, and obtain a second detection result; The indicator evaluation function is defined as: （4） in, The comprehensive evaluation score of coal quality is taken as the second test result. is the number of characteristic factors, is the model activation function, is the weight coefficient of the characteristic factor, Enter values ​​for the characteristic factor results, is the coupling coefficient of the index evaluation function, is the bias vector of the index evaluation function, The data dimension value for the input value of the characteristic factor result; （8） in, They are the hyper parameters of the quantitative model of coal quality indicators; （9） is the weight coefficient of the characteristic factor, Enter values ​​for the characteristic factor results, is the weight penalty factor.

7. A mobile platform coal quality remote detection system, used to implement the mobile platform coal quality remote detection method according to any one of claims 1 to 6, characterized in that: The mobile platform coal quality remote detection system specifically includes: Mobile coal quality testing equipment, used for real-time testing of stored coal and obtaining real-time spectral data, wherein the real-time testing is based on the dual-pulse laser technology to enhance the laser pulses emitted by the laser induced breakdown spectrometer; A preprocessing module is used to pre-build a data processing model based on adaptive interactive multi-model combined with deep learning, and pre-process the real-time spectral data based on the data processing model to obtain a real-time data set; A model building module is used to obtain spectral sample data, input the spectral sample data into a pre-built data processing model, and the data processing model pre-processes the spectral sample data to obtain a sample data set, which is divided into a training set and a validation set. A coal quality index quantitative model based on a convolutional neural network combined with a lightweight detection structure is established. The training set is used as input, and the coal quality index quantitative model is iteratively trained until convergence. The validation set is used to verify whether the model meets the preset detection accuracy. If it meets the detection accuracy, the trained coal quality index quantitative model is output; The coal quality detection module takes the real-time data set as input and executes the coal quality index quantitative model. The coal quality index quantitative model analyzes the real-time data set to obtain a first detection result and a second detection result, wherein the first detection result includes a coal particle size detection result, a coal moisture detection result, a coal calorific value detection result, a coal volatile matter detection result, and a coal element detection result.

8. The mobile platform coal quality remote detection system according to claim 7, characterized in that: The mobile coal quality detection equipment comprises: A mobile base, wherein a plurality of groups of mobile rollers are installed at the bottom of the mobile base, an equipment stand is fixedly installed on the mobile base, and a rotating drive unit is arranged on the equipment stand; A remote detection mechanism connected to the rotary drive unit, the remote detection mechanism being used for remote detection of coal quality; Wherein, the remote detection mechanism includes: A mechanism mounting seat, the mechanism mounting seat being connected to the rotation driving part; An environmental monitoring camera, which is used to obtain coal storage environment data and photograph the coal surface; A coal quality analyzer, which emits laser pulses in real time and collects real-time spectral data based on laser induced breakdown spectroscopy technology; a synchronous adjustment unit, the synchronous adjustment unit being arranged on the mechanism mounting seat, the synchronous adjustment unit being connected to the environment monitoring camera and the coal quality analyzer respectively, and being used for synchronously adjusting the tilt angles of the environment monitoring camera and the coal quality analyzer, and An auxiliary cleaning part, wherein the auxiliary cleaning part is arranged on the mechanism mounting seat and is used for auxiliary cleaning of the coal quality analyzer.

9. The mobile platform coal quality remote detection system according to claim 8, characterized in that: The synchronization adjustment unit includes: An adjusting motor, wherein the adjusting motor is fixedly mounted on the mechanism mounting seat; A linkage gear fixedly connected to the output end of the regulating motor, a monitoring support seat is arranged above the linkage gear, and the bottom of the monitoring support seat is fixedly connected to the linkage gear through a connecting rod; A driven gear, the driven gear is arranged between the monitoring support seat and the linkage gear, the driven gear is rotatably connected to the upper surface of the mechanism mounting seat, and the coal quality analyzer is detachably connected to the outer side of the driven gear; The coal quality analyzer comprises: An analyzer mainframe, the analyzer mainframe being detachably connected to a side wall of the driven gear; A laser transmitter mounted on the side wall of the analyzer mainframe, and At least one group of spectral response detectors is installed on the side wall of the coal quality analyzer, and the spectral response detectors are used to capture the emission spectrum of coal plasma.

10. The mobile platform coal quality remote detection system according to claim 9, characterized in that: The auxiliary cleaning unit comprises: A linkage round seat, wherein the linkage round seat is detachably connected to the side wall of the analyzer mainframe, and an arc-shaped driving groove is provided on the side wall of the linkage round seat, and an arc-shaped driving block is slidably embedded in the arc-shaped driving groove; A linkage swing rod fixedly connected to the arc-shaped driving block, the linkage swing rod is rotatably mounted on the cleaning limit seat, the linkage round seat is rotatably mounted on the cleaning limit seat, and the cleaning limit seat is fixedly mounted on the mechanism mounting seat; An arc gear, the arc gear is fixedly connected to the linkage swing rod, a linkage rack is provided on one side of the arc gear, the linkage rack is meshed with the arc gear, one end of the linkage rack is fixedly connected to the cleaning fan, and the linkage rack is slidably connected to the cleaning limit seat; A return spring rod is fixedly mounted on the cleaning limit seat, and one end of the return spring rod is fixedly connected to a linkage rack.

Images