A coal composition analysis method based on coal spectral data

Abstract

The present invention relates to the technical field of coal composition analysis, and more specifically, to a coal composition analysis method based on coal spectral data, comprising: obtaining coal spectral data and coal composition of each coal sample; performing feature extraction on the coal spectral data to obtain a coal data set containing multiple features; training an extreme learning machine based on the coal data set and the coal composition, and optimizing using an artificial bee colony algorithm, comprising: determining a speed factor for this round of iteration, determining a variability correlation value of the coal data set, determining a weighted speed factor for this round of iteration, and constructing an update search function for this round of iteration; and completing the construction of a coal composition analysis model, wherein the coal composition analysis model is used for coal composition analysis. By constructing the update search function, the present invention enables the artificial bee colony algorithm to search more flexibly in the parameter space when optimizing the parameters of the extreme learning machine model, thereby reducing the possibility of falling into a local optimum.

Description

Technical Field

[0001] The present invention relates to the technical field of coal component analysis, and more particularly to a coal component analysis method based on coal spectral data. Background Art

[0002] With the continued development of industry, global requirements for coal quality are becoming increasingly stringent. High-quality coal is crucial for improving production efficiency and reducing environmental pollution. Therefore, industrial analysis of coal before use is extremely necessary. Traditional coal analysis relies heavily on chemical analysis methods, which, while ensuring accuracy, are costly and time-consuming. Spectral analysis technology, with its advantages of fast analysis speed, low detection costs, and high efficiency, has been widely used in recent years in fields such as ore analysis, grade identification, and food inspection. Numerous studies have demonstrated that the use of spectral characteristics can effectively measure coal indicators such as moisture, ash, volatile matter, fixed carbon, calorific value, and sulfur content, providing a new and efficient approach for industrial coal analysis.

[0003] The core principle of using spectral data to detect coal composition is that the spectral characteristics of a substance are directly related to its chemical composition and structure. Specifically, different components in coal have unique absorption, reflection or transmission characteristics for light of specific wavelengths, and these characteristics can be quantitatively characterized by spectral data. The Chinese patent document with announcement number CN108489912B discloses a coal composition analysis method based on coal spectral data, including: coal spectral data collection; coal composition prediction using a coal composition analysis model, the input of the model is the collected spectral data, and the output is the coal composition. This method uses spectral data and coal industry analysis and measurement results to establish a coal composition analysis model. The model uses a convolutional neural network to extract spectral feature data, and outputs the spectral feature data to the extreme learning machine to obtain the coal composition corresponding to the coal spectral data obtained by the coal industry analysis and measurement. In this prediction process, an artificial bee colony algorithm is used to optimize the weights and deviations of the extreme learning machine, thereby obtaining an optimized coal composition analysis model.

[0004] However, the aforementioned patent document does not address the problem that, in coal composition analysis, due to the high correlation between certain coal components in spectral feature data, the prediction error in certain regions of the extreme learning machine model's parameter space will change relatively slowly, causing the artificial bee colony algorithm to linger in these regions and fall into a local optimum. Furthermore, highly correlated components will limit the dimensionality of the parameter space (the dimensionality of the parameter space refers to the number of dimensions of the space consisting of the possible value ranges of all model parameters. When highly correlated features exist, the information between these features is highly overlapping, effectively limiting the dimensionality of the parameter space), making the optimization process limited to the combination of these features, making it difficult to explore a wider solution space. For example, assuming that ash and volatile matter are highly correlated, adjusting the ash parameters may be accompanied by adjusting the volatile matter parameters because their changes are interrelated. As a result, the extreme learning machine model cannot optimize the ash or volatile matter parameters separately during the optimization process, resulting in the exploration of the solution space being limited to the combination of these features. Summary of the Invention

[0005] To address the problem that the prediction error in certain areas of the parameter space of the extreme learning machine model changes relatively slowly, causing the artificial bee colony algorithm to remain in these areas and fall into local optimality; and that highly correlated features limit the dimensionality of the parameter space, making the optimization process limited to the combination of these features, making it difficult to explore a wider solution space, the present invention proposes a coal composition analysis method based on coal spectral data, which includes the following steps:

[0006] Coal spectral data and coal composition of each coal sample are obtained, and features are extracted from the coal spectral data to obtain a coal data set containing multiple features. An extreme learning machine is trained based on the coal data set and the coal composition, and the artificial bee colony algorithm is used for optimization to complete the construction of a coal composition analysis model, which is used for coal composition analysis. In each round of optimization using the artificial bee colony algorithm, an update search function is constructed to replace the weights and deviations of the extreme learning machine, including: correcting a preset initial speed factor according to the number of iterations in this round and a preset maximum number of iterations to obtain the speed factor for this round of iteration; determining a variability correlation value of the coal data set according to the variance of each feature of the coal data set and the correlation between each two features; weighting the speed factor of this round of iteration based on the variability correlation value to obtain a weighted speed factor for this round of iteration; and constructing an update search function for this round of iteration based on the value of each individual in each dimension of the current solution of the artificial bee colony algorithm, the value of any remaining individual in each dimension of the current solution, and the weighted speed factor.

[0007] The beneficial effects are: by introducing a variability correlation value based on feature variance and correlation, the speed factor of the artificial bee colony algorithm is weightedly adjusted, making the search process more flexible, helping to escape the local flat area of ​​the parameter space caused by high-correlation features, and improving the global optimization ability; using the correlation information between features to dynamically adjust the search step and direction, avoiding the limitation of the parameter space dimension due to the high coupling of features, which is conducive to a more comprehensive exploration of the parameter space, thereby obtaining better extreme learning machine model parameters; the optimized extreme learning machine model can more accurately capture the complex relationship between coal composition and spectral characteristics, and improve the accuracy and stability of component analysis; through iterative dynamic adjustment of the speed factor, the artificial bee colony algorithm has better adaptability to the characteristic structure of different coal data sets, which improves the robustness and reliability of the artificial bee colony algorithm in practical applications; in the iterative process, the search strategy is automatically adjusted according to the feature variability, which reduces the dependence on initial parameters and manual experience, and improves the automation and intelligence level of the optimization process.

[0008] Furthermore, the coal samples include anthracite, bituminous coal and lignite.

[0009] Furthermore, the coal spectral data is obtained by grinding each coal sample to obtain a coal powder sample piece, performing multiple spectral tests on each coal powder sample piece, and taking the average of all spectral tests of each coal powder sample piece as the coal spectral data of each coal sample.

[0010] The beneficial effects are: the grinding process makes the sample more uniform, reduces the influence of particle size and morphology differences on the spectral results, and makes the obtained spectral data better represent the overall properties of the coal sample; by performing multiple spectral tests on the coal powder sample slices and taking the average, it can effectively reduce the accidental errors and noise in a single measurement, and improve the stability and repeatability of the spectral data.

[0011] Furthermore, the coal composition is obtained by measuring each coal sample through coal industry analysis.

[0012] Furthermore, the coal components include moisture, volatile matter, ash, fixed carbon, lower calorific value and sulfur content.

[0013] Furthermore, the feature extraction adopts convolutional neural network.

[0014] Furthermore, the speed factor of this round of iteration satisfies:

[0015] Where, For the The speed factor of the round iteration, is the preset initial velocity factor, For the The number of round iterations, The preset maximum number of iterations.

[0016] The beneficial effects are: by linearly decreasing the speed factor with the number of iterations, the artificial bee colony algorithm adopts a larger step size in the early stage, which helps to quickly jump out of the local area and expand the search range. As the iteration gradually deepens, the step size decreases, which is conducive to fine search and convergence; dynamic adjustment of the speed factor avoids the oscillation caused by excessive step size in the later update, helps the artificial bee colony algorithm to converge stably to a better solution, and improves the tuning accuracy of model parameters; the decreasing mechanism of the speed factor effectively balances global exploration and local utilization, strengthens diversity exploration in the early stage, and strengthens the refinement of excellent solutions in the later stage, thereby improving optimization efficiency and effect.

[0017] Furthermore, the correlation is a Pearson correlation coefficient.

[0018] Furthermore, the variability association value satisfies:

[0019] Where, is the variability association value of the feature data, is the number of features of the coal dataset, The coal dataset The variance of the features, The coal dataset Features and The correlation of features, is the natural exponential function, is the absolute value symbol.

[0020] The beneficial effects are: the variability correlation value reflects both the variance (variability) of the feature and the degree of correlation between features, and can effectively quantify the independent information and redundancy of each feature in the coal data set; the contribution of the feature variation amplitude is represented by exponentially increasing the feature variance, and the correlation coefficient of highly correlated features is reduced by exponential decay, making the weighting process more scientific and avoiding excessive influence of high-correlation features on parameter optimization.

[0021] Furthermore, the update search function satisfies:

[0022] Where, For the A new solution to the update search function for round iterations, For the The current solution of the artificial bee colony algorithm in the first round of iteration Individuals in The value of the dimension, For the The weighted speed factor of the round iteration, For the The current solution of the artificial bee colony algorithm in the first round of iteration Individuals in The value of the dimension, is the preset interference factor.

[0023] The beneficial effects are: updating the search function to utilize the dimensional differences of different individuals in the current solution to guide the search direction, which helps the artificial bee colony algorithm to make full use of group information and improve the diversity and effectiveness of the search; introducing weighted speed factors and interference factors, so that the search step size and direction can be adaptively adjusted during the iteration process, enhancing the ability of the artificial bee colony algorithm to escape from local optimality and improving global optimization performance; introducing randomness through interference factors to avoid the search from falling into premature convergence, while the weighted speed factor ensures that the search tends to converge, achieving a dynamic balance between exploration and utilization.

[0024] The present invention has the following beneficial effects:

[0025] (1) By correcting the speed factor according to the number of iterations and determining the variability correlation value based on the variance and correlation of the coal dataset characteristics to weight the speed factor, the artificial bee colony algorithm can search more flexibly in the parameter space when optimizing the parameters of the extreme learning machine model, avoiding the artificial bee colony algorithm from excessively staying in the area where the prediction error changes slowly (due to the high correlation of coal characteristics) in the parameter space of the extreme learning machine model, reducing the possibility of falling into the local optimum, and thus improving the optimization effect and generalization ability of the model.

[0026] (2) Highly correlated features can limit the dimension of the parameter space, restricting the optimization process to feature combinations. However, the update search function constructed in this invention takes into account the variability correlation value of the coal dataset and can break the limitations on the exploration of the solution space caused by the high correlation of features. For example, when ash and volatile matter are highly correlated, the parameter space can be explored more freely, and there is a chance to find a better parameter combination. This allows the extreme learning machine model to optimize the parameters of related features individually or more flexibly, thus expanding the scope of exploration of the solution space.

[0027] (3) Since the optimization process is more effective, local optimality is avoided and the exploration of the knowledge space is expanded, the parameters of the extreme learning machine model can be adjusted more reasonably, which helps to improve the accuracy of the coal composition analysis model constructed based on coal spectral data, thereby more accurately analyzing the coal composition and providing a more reliable basis for coal quality assessment and rational utilization.

[0028] (4) The present invention improves the artificial bee colony algorithm to address the special case of high correlation of features in coal spectral data. By considering operations such as the variability correlation value of the coal data set, the artificial bee colony algorithm can better adapt to the characteristics of data in coal composition analysis, thereby enhancing the adaptability and effectiveness of the artificial bee colony algorithm in this specific field. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a flowchart of the steps of a coal composition analysis method based on coal spectral data according to an embodiment of the present invention.

[0030] Figure 2 This is a flowchart of step S3 in a method for analyzing coal components based on coal spectral data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The following is a clear and complete description of the technical solutions in the embodiments of the present invention. The described embodiments are part of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of the present invention.

[0032] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0033] See also Figure 1 , which shows a flowchart of a method for analyzing coal components based on coal spectral data provided by one embodiment of the present invention, including steps S1 to S4:

[0034] S1: Obtain coal spectrum data and coal composition of each coal sample.

[0035] It should be noted that coal is a complex solid, combustible mineral primarily composed of organic matter, inorganic matter, and water. Coal spectral data encompasses visible, near-infrared, mid-infrared, and short-wave infrared bands. Coal composition is analyzed by analyzing the relationship between the absorption, reflection, and transmission characteristics of specific wavelengths and coal components (such as moisture, ash, and volatile matter). Coal composition analysis can optimize coal utilization efficiency. For example, high-volatile coal (such as lignite) is suitable for layer-fired furnaces, while low-volatile coal (such as anthracite) is more suitable for pulverized coal furnaces. Coal characteristics can also be analyzed to determine subsequent processing options. For example, sulfur content directly affects sulfur dioxide emissions, necessitating desulfurization facilities for high-sulfur coal. The principle behind obtaining coal composition data is that extreme learning machines are supervised machine learning models that require labeled training data for model training.

[0036] Specifically, various coal samples (eg, 100 coal samples) are collected from multiple regions in China, including anthracite, bituminous coal, and lignite.

[0037] Specifically, the coal spectrum data is obtained in the following manner:

[0038] Each coal sample is ground to obtain a pulverized coal sample piece. An SVC HR-1024 portable full-spectrum ground feature spectrometer is used as a spectral data acquisition instrument to perform multiple (e.g., 5) spectral tests on each pulverized coal sample piece. The average value of all spectral tests on each pulverized coal sample piece is used as the coal spectral data of each coal sample.

[0039] Specifically, the coal composition is obtained by measuring each coal sample through coal industry analysis (mainly chemical methods, such as moisture content, ash content, and volatile matter content).

[0040] Specifically, the coal components include moisture, volatile matter, ash, fixed carbon, lower calorific value and sulfur content.

[0041] S2: Extract features from the coal spectral data to obtain a coal dataset containing multiple features.

[0042] It should be noted that coal spectral data comes directly from spectral acquisition equipment, contains a large amount of redundant information and complex spectral response patterns, and cannot be directly used for coal composition analysis. The association between coal composition and coal spectral data is implicit in the absorption, reflection or transmission characteristics of specific bands. These key information must be converted into effective features that can be used for model training through feature extraction. The extracted features can more accurately reflect the intrinsic correlation between coal composition and spectrum, providing high-quality input for model training. Therefore, this step requires feature extraction of coal spectral data to obtain effective feature spectral data, that is, a coal data set containing multiple features, thereby providing a basis for subsequent analysis.

[0043] Specifically, the feature extraction adopts convolutional neural network.

[0044] Among them, the convolutional neural network has a strong feature extraction capability, which can extract effective spectral feature data from the coal spectral data, provide a basis for subsequent analysis, and solve the problem of spectral data feature extraction; the coal spectral data is input into the convolutional neural network for feature extraction to obtain spectral feature data. For example, the convolutional neural network structure is selected as a 2-layer convolution. and 2-layer sampling ; Convolutional layer selection As the activation function, the sampling layer selects the mean sampling function; each spectral data has 1024 features, and after the convolutional neural network feature extraction, the output feature data is 300.

[0045] S3: Train an extreme learning machine based on the coal dataset and coal composition, and optimize it using the artificial bee colony algorithm.

[0046] It should be noted that the extreme learning machine is a key component in building a coal composition analysis model. A complex nonlinear relationship exists between coal spectral characteristics (such as absorbance at different wavelengths) and components (such as ash and volatile matter). The extreme learning machine uses a single-hidden-layer feedforward neural network to construct a nonlinear mapping function between coal spectral characteristic data and coal composition data. This allows the extreme learning machine to obtain coal composition data by inputting coal spectral characteristic data.

[0047] A coal dataset containing multiple features is used as the input data of the extreme learning machine, and the industrial analysis results of coal composition, including indicators such as moisture, volatile matter, ash, fixed carbon, lower calorific value and sulfur content, are used as the output data of the extreme learning machine. The model is trained using an extreme learning machine with a single hidden layer feedforward neural network.

[0048] It should be noted that when traditional artificial bee colony algorithms are used to analyze coal composition, because coal is a complex organic-inorganic mixture with natural chemical correlations between its components, resulting in highly correlated spectral features. This high correlation leads to different parameter combinations in the model parameter space producing similar predictions, forming "flat regions" with gently varying errors. In traditional artificial bee colony algorithms, individual (bee) position updates rely on random step sizes, which can easily cause them to stagnate within these "flat regions" and become trapped in local optima. This step introduces a weighted speed factor into the artificial bee colony algorithm to dynamically adjust the search step size. By adjusting the step size, the algorithm balances its global search capability with local optimization accuracy, addressing the problem of being trapped in local optima in coal composition analysis caused by feature correlation and data dispersion.

[0049] In each round of optimization using the artificial bee colony algorithm, an update search function is constructed to replace the weights and biases of the extreme learning machine, referring to Figure 2 , step S3 includes steps S301 to S304, which are specifically as follows:

[0050] S301: Determine the speed factor of this round of iteration.

[0051] It should be noted that the coal spectral characteristic data contains a variety of coal varieties. During the exploration process, it is necessary to cover different areas in the solution space as much as possible to avoid falling into local optimality. In the early exploration, it is necessary to cover the spectral differences of different coal types, so it is necessary to set a larger speed factor to control the large step size, allowing the artificial bee colony algorithm to perform "jump-type" exploration and break through the attraction of local optimal values. With the increase in the number of iterations, in the later exploration of the artificial bee colony algorithm, it is necessary to fine-tune the parameters in the optimal solution area located in the early stage to approach the global optimal, so a smaller speed factor is set to control the small step size, so that the artificial bee colony algorithm can perform more precise exploration in the current area.

[0052] The preset initial speed factor is modified according to the number of iterations in this round and the preset maximum number of iterations to obtain the speed factor of this round of iteration.

[0053] Specifically, the speed factor of this round of iteration satisfies:

[0054] ;

[0055] Where, For the The speed factor of the round iteration, is the preset initial velocity factor, For the The number of round iterations, The preset maximum number of iterations.

[0056] Implementers can set the initial speed factor and the maximum number of iterations according to specific implementation conditions. For example, the initial speed factor is 0.9 and the maximum number of iterations is 300.

[0057] Among them, as the number of iterations The increase, The value of gradually increases, making Initial velocity factor The reduction in Gradually decrease; in the early stage of artificial bee colony algorithm, when Much smaller than hour, Approaching initial velocity factor , ensuring that the artificial bee colony algorithm can move quickly in the solution space with a large step size, fully exploring the vast area corresponding to the spectral characteristics of different coal varieties, and avoiding falling into the local optimal trap encountered in the early stage due to too small a step size; and when the iteration approaches the preset maximum number of times, Approaching 0, the search step size of the artificial bee colony algorithm is greatly reduced, prompting the algorithm to conduct a refined search in the potential optimal solution area located in the early stage, and gradually approach the global optimal solution by fine-tuning the parameters of the extreme learning machine.

[0058] S302: Determine the variability correlation value of the coal data set.

[0059] It should be noted that the variability association value of the coal dataset is used to measure the degree of synergy between the discreteness of data features and the correlation between features. The degree of data discreteness is represented by the variance of the feature, which positively regulates the variability association value. High-variance features indicate dispersed data, requiring a larger step size, which results in a larger variability association value and a larger subsequent weighted acceleration factor. This allows for rapid coverage of different regions and avoids inefficient exploration due to small step sizes. Conversely, low-variance features indicate concentrated data and require a smaller step size, which results in a smaller variability association value and a smaller subsequent weighted acceleration factor. This allows for a more refined search and avoids skipping over the optimal solution due to large step sizes. The correlation between features reflects the degree of association between two feature data. The correlation has a negative correlation adjustment on the variability correlation value. The prediction error corresponding to the high correlation feature is insensitive to parameter adjustment. Different parameter combinations may produce similar prediction errors, which will increase the risk of the artificial bee colony algorithm falling into the local optimum. The smaller the variability correlation value, the smaller the variability correlation value. By calculating the feature correlation, the variability correlation value is reduced after identifying the high correlation, thereby reducing the subsequent weighted speed factor, intensively exploring in the one-dimensional subspace, and avoiding skipping the optimal solution with a large step size. Conversely, after identifying the low correlation feature, the variability correlation value is larger, thereby increasing the subsequent weighted speed factor, so that the artificial bee colony algorithm can cross a larger parameter range in one iteration, quickly access the area of ​​different feature combinations, and improve the global search efficiency. The above two factors will significantly affect the search efficiency and accuracy of the artificial bee colony algorithm when optimizing the parameters of the extreme learning machine. Therefore, by calculating the variability correlation value, this step can better adjust the algorithm's search strategy according to the characteristics of the data itself, making the algorithm more suitable for the characteristics of coal composition analysis data.

[0060] The variability correlation value of the coal dataset is determined based on the variance of each feature of the coal dataset and the correlation between every two features.

[0061] Specifically, the correlation is the Pearson correlation coefficient.

[0062] Specifically, the variability association value satisfies:

[0063] ;

[0064] Where, is the variability association value of the feature data, is the number of features of the coal dataset, The coal dataset The variance of the features, The coal dataset Features and The correlation of the features ( , that is, between different features), is the natural exponential function, is the absolute value symbol.

[0065] in, and are the variance of the feature and the mean of the Pearson correlation coefficient of the feature respectively; Indicates that whether the two features are positively correlated or negatively correlated, the larger the absolute value, the higher the degree of association between the two features, and the higher the correlation between the two features. When it is larger, it means that during the search process of the artificial bee colony algorithm, due to these highly correlated features, more "flat areas" will be formed in the model parameter space, making the artificial bee colony algorithm prone to falling into local optimality. , which can quantitatively suppress the negative impact of this high correlation on the artificial bee colony algorithm, combined with the feature variance mean term , which together provide a basis for the subsequent artificial bee colony algorithm to adjust the search step size, balance the global search and local optimization capabilities, and improve the performance of the artificial bee colony algorithm in dealing with coal composition analysis problems.

[0066] S303: Determine the weighted speed factor of this round of iteration.

[0067] It's important to note that in the artificial bee colony algorithm (ABCA) process of optimizing an extreme learning machine (ELM) for coal composition analysis, a weighted speed factor is a key parameter. It dynamically adjusts the ABCA's search step size to balance its global search capability and local optimization accuracy. Due to the high correlation and complex variability of coal dataset characteristics, traditional ABCAs are prone to falling into local optima. However, the weighted speed factor, by combining a speed factor related to the number of iterations with a variability-related value reflecting the data's characteristics, enables the ABCA to adaptively adjust the search step size based on the actual data and the iteration process, effectively escaping local optima and approaching the global optimal solution.

[0068] Specifically, the weighted acceleration factor satisfies:

[0069] ;

[0070] Where, For the The weighted speed factor of the round iteration, For the The speed factor of the round iteration, is the variability associated value of the feature data.

[0071] Among them, the speed factor It reflects the influence of the number of iterations on the search step size. In the early stage of the artificial bee colony algorithm iteration, in order to cover different areas in the solution space as much as possible and avoid falling into the local optimum, a larger speed factor is set to control the large step size, allowing the artificial bee colony algorithm to perform "jump-type" exploration; as the number of iterations increases, the artificial bee colony algorithm gradually locates the possible optimal solution area. At this time, it is necessary to fine-tune the parameters to approach the global optimum, so the speed factor gradually decreases; and the variability correlation value The variance of each feature of the coal dataset and the correlation between each two features are comprehensively considered, reflecting the degree of data dispersion and the degree of coordination between features. When the feature variance is large or the correlation between features is low, the variability correlation value is large, which means that the data distribution is more dispersed and a larger search step is required; otherwise, a smaller search step is required. Therefore, the weighted speed factor By multiplying the speed factor and the variability correlation value, the purpose of dynamically adjusting the search step size according to the iteration process and data characteristics is achieved.

[0072] S304: Construct an update search function for this round of iteration.

[0073] It should be noted that the update search function is the core formula used by the artificial bee colony algorithm to generate new solutions in each iteration. In the case of coal composition analysis, due to the complexity of the data and the high correlation of features, traditional artificial bee colony algorithm update methods can easily cause individuals to stagnate in "flat areas," leading to the artificial bee colony algorithm being trapped in local optima. However, the update search function constructed here, by introducing a weighted acceleration factor and a preset interference factor, enables the artificial bee colony algorithm to more flexibly explore the solution space when generating new solutions. The weighted acceleration factor dynamically adjusts the search step size by taking into account the influence of the number of iterations and data characteristics; the interference factor introduces a degree of randomness into the search process, helping to break free from local optima and enhance the artificial bee colony algorithm's global search capabilities. In this way, the update search function guides the artificial bee colony algorithm to more effectively find the optimal extreme learning machine weights and biases in a complex data space, thereby improving the accuracy of the coal composition analysis model.

[0074] Based on the value of each individual in each dimension of the current solution of the artificial bee colony algorithm, the value of each dimension of any remaining individuals in the current solution, and the weighted speed factor, an update search function for this round of iteration is constructed.

[0075] Specifically, the update search function satisfies:

[0076] ;

[0077] Where, For the A new solution to the update search function for round iterations, For the The current solution of the artificial bee colony algorithm in the first round of iteration Individuals in The value of the dimension, For the The weighted speed factor of the round iteration, For the The current solution of the artificial bee colony algorithm in the first round of iteration Individuals in The value of the dimension, is the preset interference factor.

[0078] Here, individual refers to a set of parameters to be optimized in the extreme learning machine model of bees in the artificial bee colony algorithm. In the present invention, it refers to the weight matrix from the input layer to the hidden layer and the deviation vector of the hidden layer. Dimension refers to the dimension of the parameter, that is, a single weight or deviation of the extreme learning machine model. Indicates the specific value of a dimension of a possible solution searched by the current artificial bee colony algorithm. is the current solution Individuals in The values ​​of the dimensions provide diversity for the generation of new solutions by selecting different individuals for comparison; the weighted speed factor Controlled from the current individual Towards The step size of the direction search is dynamically adjusted according to the number of iterations and data characteristics; the preset interference factor The randomness is introduced into the search process, which enables the employed bees to explore new feasible solutions in the neighborhood of the existing food source (i.e., the weights and bias of the extreme learning machine); the new solution In the current individual Based on and The direction of the difference ( , that is, different individuals to ensure is not zero to avoid search stagnation), and is updated with a step size determined by the weighted speed factor and the interference factor, so that each round of iteration can generate a new possible solution based on the current solution and data characteristics, gradually approaching the global optimal solution.

[0079] S4: Complete the construction of a coal composition analysis model, which is used for analyzing coal composition.

[0080] Through the above steps, the construction of the coal composition analysis model is completed. Subsequently, the newly collected coal spectral data is input into the coal composition analysis model to obtain the corresponding coal composition, completing the coal composition analysis method based on coal spectral data.

[0081] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A coal composition analysis method based on coal spectral data, characterized in that: include: Obtaining coal spectral data and coal composition for each coal sample, performing feature extraction on the coal spectral data to obtain a coal dataset containing multiple features, training an extreme learning machine based on the coal dataset and coal composition, and optimizing using an artificial bee colony algorithm to complete the construction of a coal composition analysis model, which is used for coal composition analysis; In each round of optimization using the artificial bee colony algorithm, an update search function is constructed to replace the weights and biases of the extreme learning machine, including: The preset initial speed factor is modified according to the number of iterations of this round and the preset maximum number of iterations to obtain the speed factor of this round of iteration; Based on the variance of each feature of the coal dataset and the correlation between every two features, the variability correlation value of the coal dataset is determined to meet the following requirements: , is the variability association value of the feature data, is the number of features of the coal dataset, The coal dataset The variance of the features, The coal dataset Features and The correlation of features, is the natural exponential function, is the absolute value symbol; The speed factor of this round of iteration is weighted based on the variability association value to obtain the weighted speed factor of this round of iteration; based on the value of each individual in each dimension of the current solution of the artificial bee colony algorithm, the value of each dimension of any remaining individuals of the current solution, and the weighted speed factor, an update search function of this round of iteration is constructed.

2. The method for analyzing coal composition based on coal spectral data according to claim 1, characterized in that: The coal samples include anthracite, bituminous coal and lignite.

3. The method for analyzing coal composition based on coal spectral data according to claim 1, characterized in that: The coal spectrum data is obtained in the following manner: Each coal sample is ground to obtain a coal powder sample piece, and each coal powder sample piece is subjected to multiple spectral tests, and the average value of all spectral tests of each coal powder sample piece is used as the coal spectral data of each coal sample.

4. The method for analyzing coal composition based on coal spectral data according to claim 1, characterized in that: The coal composition is determined by coal industry analysis of each coal sample.

5. The method for analyzing coal composition based on coal spectral data according to claim 1, characterized in that: The coal components include moisture, volatile matter, ash, fixed carbon, lower calorific value and sulfur content.

6. The method for analyzing coal composition based on coal spectral data according to claim 1, characterized in that: The feature extraction adopts convolutional neural network.

7. The method for analyzing coal composition based on coal spectral data according to claim 1, characterized in that: The speed factor of this round of iteration satisfies: ； Where, For the The speed factor of the round iteration, is the preset initial velocity factor, For the The number of round iterations, The preset maximum number of iterations.

8. The method for analyzing coal composition based on coal spectral data according to claim 1, characterized in that: The correlation is the Pearson correlation coefficient.

9. The method for analyzing coal components based on coal spectrum data according to claim 1, characterized in that: The update search function satisfies: ； Where, For the A new solution to the update search function for round iterations, For the The current solution of the artificial bee colony algorithm in the first round of iteration Individuals in The value of the dimension, For the The weighted speed factor of the round iteration, For the The current solution of the artificial bee colony algorithm in the first round of iteration Individuals in The value of the dimension, is the preset interference factor.

Images