A coal gangue classification label generation method and system based on a clustering algorithm

Abstract

The application is based on a coal gangue classification label generation method and system based on a clustering algorithm: N pieces of feature data related to coal gangue spontaneous combustion are obtained, and M different features are screened out; a large number of literatures related to coal gangue spontaneous combustion are investigated, all high-frequency features related to coal gangue spontaneous combustion in the literatures are captured, and L features are screened out by combining the M features and expert evaluation; the features data of each feature belonging to the L features in the N pieces of feature data are summarized, and H pieces of feature data are obtained by preprocessing any piece of feature data with missing values after summarization, each including L feature data; the correlation coefficient between each two features in the L features is calculated, and the L features are reduced and optimized according to the correlation coefficient to obtain I features; the weight of each feature is calculated; the H pieces of feature data after negative correlation feature data reverse processing are clustered by using a clustering model combined with the weight, and the final coal gangue spontaneous combustion classification label of each of the H pieces of feature data is obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and resources and environment technology, in particular to a coal gangue classification label generation method and system based on a clustering algorithm. BACKGROUND

[0002] In the current national bulk solid waste treatment environment, coal gangue is one of the industrial wastes with the largest emission and cumulative amount and serious harm in China. The existing gangue in China accumulates more than 6 billion tons (about 1 / 2 in North China), forming 1600-1800 gangue dumps, occupying and polluting a large amount of land resources, and more than 300 of them are in long-term self-ignition state due to the long-term accumulation of self-ignition and heat accumulation. The annual self-ignition gangue dump will emit 10.8g of CO, about 2g of H2S and NO X , about 6.5g of SO2, and polycyclic aromatic hydrocarbon organic pollutants such as benzopyrene and benzo-fluoranthene, particulate matter, polluting the atmosphere and endangering health. Even more than 50 explosions have occurred, nearly 10 of which are major accidents. The self-igniting coal gangue dump is prone to landslides, collapses, explosions and other disasters in the rainy season due to a large amount of rainwater erosion and thermal expansion, not only affecting the physical and mental health of local residents, but also seriously threatening the safety of coal mines and society. Therefore, the coal gangue treatment and disposal in China is urgent.

[0003] Due to the ecological and environmental damage caused by high-intensity and large-scale coal mining in China's main coal mining areas, problems such as surface subsidence and collapse and groundwater drawdown have occurred. In combination with the main coal gangue treatment and disposal method of "underground filling + ground backfilling", the gangue reduction is promoted. Based on this treatment method, the influence of coal gangue self-ignition on the setting of the treatment scheme and the treatment effect is considered. Therefore, it is necessary to classify the coal gangue according to the self-ignition characteristics in advance, and then correspond the coal gangue of different combustion grades to different backfilling schemes, so as to save manpower, material resources and financial resources and improve efficiency.

[0004] However, current methods for identifying the spontaneous combustion tendency of coal gangue mainly rely on experimental data and empirical formulas. For example, one method for identifying the spontaneous combustion tendency of coal gangue involves measuring the mass percentage of sulfur in the coal gangue sample and the cross-point temperature, and then deriving an identification formula for the spontaneous combustion tendency of coal gangue. However, these methods have problems such as incomplete data collection, inaccurate feature selection, and insufficient model validation, making it difficult to meet the high precision requirements in practical applications. Today, the PSO / Z method (Particle Swarm Optimization and Z-Score Method) is commonly used to predict the spontaneous combustion tendency of coal gangue. The PSO / Z method combines the particle swarm optimization algorithm (PSO) and the Z-Score analysis method. In predicting the spontaneous combustion tendency of coal gangue, PSO is used to optimize the parameters in the prediction model, enhancing the model's predictive ability, while the Z-Score analysis method is used to assess the spontaneous combustion risk of different coal gangue samples. Specifically, PSO is used to optimize the hyperparameters of neural networks, support vector machines (SVM), or other machine learning algorithms, thereby improving the accuracy of spontaneous combustion tendency prediction; the Z-Score method can effectively eliminate some noisy or extreme data through standardization and anomaly detection, thereby improving the stability of the prediction model. By combining PSO and Z-Score, researchers can extract key features that affect coal gangue spontaneous combustion from a large amount of raw data, and then establish a more accurate prediction model. Although the PSO / Z method has improved the accuracy of prediction to some extent, it still has room for improvement in dealing with complex and dynamically changing environmental factors. Moreover, the method studies fewer coal gangue feature quantities, which has certain limitations.

[0005] Therefore, in combination with machine learning algorithms, coal gangue is classified according to its spontaneous combustion characteristics, but the class labels of each data need to be determined in advance before classification. Therefore, the present application uses a clustering algorithm to integrate multiple coal gangue component attribute features to cluster coal gangue, and the clustering results are verified using internal evaluation indicators of the clustering algorithm and visual scatter plots of the clustering results to obtain the final coal gangue spontaneous combustion class labels, thereby providing prior technical support for subsequent labeled coal gangue classification prediction and backfill scheme issuance. SUMMARY

[0006] The present application provides a coal gangue classification label generation method and system based on a clustering algorithm to address the problems and deficiencies of the prior art.

[0007] The present application solves the above technical problems by the following technical solutions:

[0008] The present application provides a coal gangue classification label generation method based on a clustering algorithm, characterized by the following steps:

[0009] S1, collect a plurality of feature data related to coal gangue spontaneous combustion in each of N historical coal gangue spontaneous combustion related materials, obtain N feature data, and select M different features based on the N feature data;

[0010] S2, through research on a large number of coal gangue spontaneous combustion related literatures, all high frequency features related to coal gangue spontaneous combustion in the literatures are captured, and L features are selected by combining the M features and expert evaluation, L

[0011] S3, the feature data belonging to the L features in each of the N feature data is summarized, and any feature data with missing values in the N feature data after summarization is deleted or filled, H feature data is obtained, and each feature data includes L feature data;

[0012] S4, the correlation coefficient between each two features in the L features is calculated using the H feature data, the L features are optimized according to the correlation coefficient, I features are obtained, I

[0013] S5, research K articles related to coal gangue spontaneous combustion, respectively, count the frequency of each feature in the I features in the K articles, and calculate the feature weight of each feature;

[0014] S6, in the H feature data, the feature data of the features negatively correlated with coal gangue spontaneous combustion is inverted to obtain positively correlated feature data, combined with the feature weight, the inverted H feature data is clustered by using a weighted WK-means clustering model, and is clustered into three clusters, namely not easy to spontaneous combustion, easy to spontaneous combustion and extremely easy to spontaneous combustion, and a clustering result is obtained;

[0015] S7, the clustering result is verified multiple times, and the final coal gangue spontaneous combustion category label of each feature data in the H feature data is obtained.

[0016] The application also provides a coal gangue classification label generation method based on a clustering algorithm, which has the characteristics that it comprises:

[0017] A data collection module is used for collecting a plurality of feature data related to coal gangue spontaneous combustion in each of N historical coal gangue spontaneous combustion related materials, obtaining N feature data, and selecting M different features based on the N feature data;

[0018] A feature selection module is used for capturing all high frequency features related to coal gangue spontaneous combustion in the literatures by researching a large number of coal gangue spontaneous combustion related literatures, combining the M features, and selecting L features by expert evaluation, L

[0019] A preprocessing module is used for collecting feature data of features belonging to L features in each of N pieces of feature data, and performing deletion or filling preprocessing on feature data with missing values in any of the N pieces of feature data after collection, to obtain H pieces of feature data, each of which includes L pieces of feature data;

[0020] A correlation calculation module is used for calculating correlation coefficients between two features in L features by using the H pieces of feature data, and performing reduction optimization on the L features according to the correlation coefficients to obtain I features, I≤L, each piece of feature data containing I pieces of feature data;

[0021] A weight calculation module is used for investigating K pieces of literature related to coal gangue spontaneous combustion, respectively counting the frequency of each feature in I features in the K pieces of literature, and calculating the feature weight of each feature;

[0022] A clustering module is used for inverting the feature data of features negatively correlated with coal gangue spontaneous combustion in the H pieces of feature data to obtain feature data positively correlated with coal gangue spontaneous combustion, combining the feature weight, and using a weighted WK-means clustering model to cluster the H pieces of feature data after inversion, and clustering them into three clusters of not easy to spontaneous combustion, easy to spontaneous combustion and extremely easy to spontaneous combustion, to obtain a clustering result;

[0023] A category acquisition module is used for multiple verification of the clustering result, to obtain the final coal gangue spontaneous combustion category label of each piece of feature data in the H pieces of feature data.

[0024] The positive progress effect of the present application is that:

[0025] The present application proposes a coal gangue classification label generation method and system based on a clustering algorithm, which breaks through the classification limitation of traditional classification standards for single features, avoids the subjectivity and limitation of manual clustering, and provides a pilot condition for the construction of subsequent classification prediction models and the issuance of backfilling schemes.

[0026] The application collects characteristic data related to coal gangue spontaneous combustion in advance by literature retrieval on a well-known literature database combined with historical experimental data; the collected data is subjected to characteristic screening to screen out characteristic data highly related to coal gangue spontaneous combustion; the screened data is subjected to pretreatment, including deleting data with serious missing rows and predicting and filling missing values; the finally determined data is subjected to characteristic correlation evaluation; the frequency-based simple weighting method is used to calculate the weight of each characteristic relative to coal gangue spontaneous combustion; the coal gangue data is divided into three clustering clusters by combining the weight and the weighted WK-means clustering algorithm; and the clustering result is evaluated by using the clustering result visual scatter plot, Calinski-Harabasz index and the like. The method provides effective technical support for the establishment of a classification data label based on coal gangue spontaneous combustion, and provides a pilot basis for the subsequent construction of a classification and prediction model based on the attribute characteristics of coal gangue composition and the determination of a backfill scheme. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 The flowchart of the coal gangue classification label generation method based on the clustering algorithm of the preferred embodiment of the application.

[0028] Figure 2 The clustering result visual scatter plot of the coal gangue classification label generation method based on the clustering algorithm of the preferred embodiment of the application.

[0029] Figure 3 The clustering result Calinski-Harabasz index column chart of the coal gangue classification label generation method based on the clustering algorithm of the preferred embodiment of the application.

[0030] Figure 4 The block diagram of the coal gangue classification label generation system based on the clustering algorithm of the preferred embodiment of the application. DETAILED DESCRIPTION

[0031] In order to make the objects, technical solutions and advantages of the embodiments of the application clearer, the technical solutions in the embodiments of the application will be described below in connection with the drawings of the embodiments of the application. Obviously, the described embodiments are only some of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0032] For ease of description, only parts related to the application are shown in the drawings. The first, second and the like involved in the application are only for the convenience of describing the technical solutions of the application, and do not have a specific limiting effect, but are generic, and do not constitute a limiting effect on the technical solutions of the application.

[0033] As Figure 1As shown, the embodiment of the present application provides a coal gangue classification label generation method based on a clustering algorithm, including the following steps:

[0034] Step 101, collect N pieces of historical coal gangue spontaneous combustion related materials each with multiple feature data related to coal gangue spontaneous combustion, obtain N pieces of feature data, and filter out M different features based on the N pieces of feature data.

[0035] Wherein N is the number of collected historical coal gangue spontaneous combustion related materials, and M is the number of collected features related to coal gangue spontaneous combustion. One historical coal gangue spontaneous combustion related material collects one piece of feature data, and each piece of feature data includes multiple feature data.

[0036] In step 101, the existing published literature in each database is searched with the vocabulary related to coal gangue spontaneous combustion as the keyword, multiple feature data related to coal gangue spontaneous combustion is found in each literature searched, and multiple feature data obtained from any literature is taken as one piece of feature data. For example: taking "coal gangue", "coal gangue spontaneous combustion", "coal gangue comprehensive utilization" and the like as the keyword, the published literature in well-known databases such as Web of Science, Wanfang, CNKI, Scopus and the like is comprehensively and deeply reviewed, multiple features related to spontaneous combustion are found from a large number of related literature according to the keyword, and 135 pieces of feature data are screened out after screening, including 14 features.

[0037] Obtain multiple experimental data of coal mines from the ecological environment bureau, and the experimental data includes multiple feature data related to coal gangue spontaneous combustion, and multiple feature data obtained from any experimental data is taken as one piece of feature data.

[0038] Obtain N pieces of feature data, and filter out M different features based on the N pieces of feature data. For example, N is 135, and M is 14.

[0039] Step 102, through research on a large number of coal gangue spontaneous combustion related literature, all high frequency features related to coal gangue spontaneous combustion in all literature are grabbed, combined with the M features, and L features are screened out through expert evaluation, L < M, for example, L is 9.

[0040] In this step, through research on a large number of coal gangue spontaneous combustion related literature, all high frequency features related to coal gangue spontaneous combustion in all literature are grabbed, combined with the collected coal gangue spontaneous combustion related features, and 9 features more important for coal gangue spontaneous combustion are selected as the final input data.

[0041] Step 103, aggregate the feature data included in each of the N pieces of feature data belonging to the L features, and perform deletion or imputation preprocessing on the feature data with missing values in any of the N pieces of aggregated feature data, to obtain H pieces of feature data, each of which includes L pieces of feature data. In the H pieces of feature data, each piece of feature data includes L features with data and no missing values.

[0042] In step 103, the feature data included in each of the N pieces of feature data belonging to the L features is aggregated, a first Excel table containing N rows and L columns of feature data is constructed, for any row, whether the row feature data has a missing value is analyzed, if there is no missing value, the row is retained, if there is a missing value, further analysis is performed, if the missing feature data reaches two or more, the row is directly deleted, if the missing feature data is only one, a regression imputation method is used to predict and fill in the missing feature data, based on which a second Excel table containing H rows and L columns of feature data is obtained, for example, L is 9 and H is 85.

[0043] The regression imputation method includes:

[0044] S31, select all row feature data without missing values from the first Excel table as a training sample, use the missing feature in the row as a target variable, and use the other L-1 features as explanatory variables to establish a regression model.

[0045] Common regression models include linear regression, ridge regression, Lasso regression, etc. The linear regression model is used in this embodiment. Based on the training set (data without missing values), the selected explanatory variables (fixed carbon, sulfur content, moisture, ash, volatile matter, Al2O3, SiO2, CaO) are used to predict the target variable (sulfur trioxide SO3). The specific steps are as follows:

[0046] Use known explanatory variables (X n ) to predict the value of the target variable (Y), train the regression model, and fit an equation: Y = β0+ β1*X1+ β2*X2+... + β n *X n .

[0047] S32, use the feature data corresponding to the explanatory variables of each row in the training sample as input, and use the feature data corresponding to the target variable as output, to train the regression model using the training sample, to obtain a trained regression model.

[0048] S33, substitute the other L-1 feature data of the row into the trained regression model to predict the target variable value, which is used to fill in the feature data of the feature missing in the row.

[0049] For the target variable containing missing values, the regression model and the known explanatory variables (Xn ) to predict the missing values of the target variable (Y: SO3); and filling the predicted missing values into the missing positions in the original data set.

[0050] In this step, for the data items with severe missing values (more than two missing values for each of the 9 features), the rows containing missing values are deleted; for the data items with less missing values (only one missing value for each of the 9 features), the regression imputation method is used to predict and fill the missing values, and finally 85 data items containing 9 features (fixed carbon, sulfur content, moisture content, ash content, volatile matter, Al203, Si02, CaO, SO3) are obtained.

[0051] In step 104, the correlation coefficients between each two of the L features are calculated using the H feature data, and the L features are optimized by deletion according to the correlation coefficients, so that I features are obtained, and I≤L, and each feature data contains I feature data.

[0052] In step 104, for the H feature data, an H*L matrix is constructed, the H*L matrix is input into the Origin professional function drawing software, the Pearson correlation coefficient formula is used to calculate the correlation coefficients between each two of the features, each calculated correlation coefficient is analyzed, and when any correlation coefficient is greater than a set threshold (such as 0.9), it is further analyzed whether the two features corresponding to the correlation coefficient are positively or negatively correlated with coal gangue spontaneous combustion. If the two features are positively and negatively correlated with coal gangue spontaneous combustion respectively, the two features are retained, and if the two features are positively or negatively correlated with coal gangue spontaneous combustion, one of the two features is deleted, and then each feature data contains L-1 features. After the above operation, none of the 9 features is deleted in this embodiment, and finally there are still 9 features, that is, I=L.

[0053] In this step, the Pearson correlation coefficient is used to evaluate the correlation between the 9 features. The Pearson correlation coefficient is used to measure the linear relationship between two continuous variables, and the value ranges from -1 to 1 (+1 indicates complete positive correlation; -1 indicates complete negative correlation; and 0 indicates no linear relationship).

[0054] Pearson correlation coefficient formula:

[0055]

[0056] x i , y i is a sample value, is a sample mean.

[0057] Step 105, investigate K papers related to coal gangue spontaneous combustion, respectively count the frequency of each of the L features appearing in the K papers, and calculate the feature weight of each feature for subsequent weighted processing. For example, K is 50.

[0058] For example: Investigate 50 papers on coal gangue spontaneous combustion, count the frequency of 9 features (fixed carbon 36 times, sulfur content 43 times, moisture content 38 times, ash content 12 times, volatile content 17 times, Al2O3 6 times, SiO2 6 times, CaO 6 times, SO3 3 times), input the statistical data into SPSS, and get the weight of each feature relative to coal gangue spontaneous combustion (fixed carbon: 0.215, sulfur content: 0.257, moisture content: 0.228, ash content: 0.072, volatile content: 0.102, Al2O3: 0.036, SiO2: 0.036, CaO: 0.036, SO3: 0.018) by simple weighting method based on frequency.

[0059] Step 106, among the H feature data, the feature data of the features negatively correlated with coal gangue spontaneous combustion is inverted to obtain positively correlated feature data, combined with the feature weight, the inverted H feature data is clustered using the weighted WK-means clustering model, and is clustered into 3 clusters, namely not easy to spontaneous combustion, easy to spontaneous combustion and extremely easy to spontaneous combustion, to obtain the clustering result.

[0060] In step 106, S61, for the second Excel table containing H rows and L columns of feature data, the feature data of the features negatively correlated with coal gangue spontaneous combustion is inverted using the maximum value subtraction method to obtain positively correlated feature data, the feature weight of each feature is read, and the H rows and L columns of feature data after inversion are weighted by feature weight to obtain H rows and L columns of weighted feature values.

[0061] S62, use the data standardization processing (StandardScaler) method to standardize the H rows and L columns of weighted feature values to obtain H rows and L columns of standardized feature values; standardization processing unifies the scale of each feature, avoiding the influence of scale difference on clustering effect by weighted WK-means clustering algorithm.

[0062] S63, set the number of clusters to 3 and the random seed to 42 to ensure the reproducibility of the result, input the H rows and L columns of standardized feature values into the weighted WK-means clustering model for clustering to obtain the clustering result label of each row, the clustering result label is not easy to spontaneous combustion, easy to spontaneous combustion and extremely easy to spontaneous combustion, the three clusters are simplified as 0, 1 and 2 respectively, the clustering result label is added as the L+1 column in the second Excel table, and the simplified clustering result label is added as the L+2 column, forming H rows and (L+2) columns of data.

[0063] For example: reverse the two negatively correlated features "ash content (%)" and "CaO (%)" to make them have the same change trend as other positively correlated features (in this embodiment, the reverse method of negatively correlated features is to subtract the maximum value of the feature from the value of each sample); traverse each column of features in the data, if the feature is in the weight dictionary, multiply it by the corresponding weight to get the weighted feature value.

[0064] Step 107, visualize the scatter plot of the clustering results, Calinski-Harabasz index, etc. to perform multiple verifications on the clustering results, and obtain the final class label of each feature data in H feature data of coal gangue spontaneous combustion.

[0065] In step 107,

[0066] S71, obtain the center point coordinates of each cluster from the weighted WK-means clustering model after clustering, which reflect the average level of each cluster feature.

[0067] S72, output the complete data (H rows (L+2) columns of data) containing the clustering results and the clustering center coordinates of each cluster, and save the complete data with the clustering results to a new Excel file for subsequent viewing and analysis.

[0068] S73, use t-SNE algorithm to reduce the dimension of H rows L columns of data to two-dimensional space for visualization; add the t-SNE reduced dimension result to the new Excel file to form new "t-SNE_1" and "t-SNE_2" columns, mark the data points of different clusters with different colors according to the t-SNE reduced dimension result, and intuitively display the clustering effect, draw the clustering center point on the scatter plot, and mark it with a red cross shape to highlight the core position of each cluster.

[0069] S74, install matplotlib, seaborn and scikit-learn libraries by pip install command, which are used for data visualization and clustering result analysis.

[0070] S75, use calinski_harabasz_score function to calculate Calinski-Harabasz index to measure the variance ratio of clustering, and the larger the value, the better the clustering effect.

[0071] S76, display the clustering results in a scatter plot, as shown in Figure 2 ; and display the calculated Calinski-Harabasz index performance indicator in the form of a bar chart, as shown inFigure 3 The Calinski-Harabasz index CH Index is 20.06587142368264, and the Calinski-Harabasz index CH Index of other clustering models is 10.17225; the two indicators represent that the method has a good clustering effect, and provides label data for subsequent classification of coal gangue.

[0072] As shown in Figure 4 The embodiment of the application also provides a coal gangue classification label generation method based on a clustering algorithm, which comprises a data collection module 1, a feature screening module 2, a preprocessing module 3, a correlation calculation module 4, a weight calculation module 5, a clustering module 6 and a category acquisition module 7.

[0073] The data collection module 1 is used for pre-collecting a plurality of feature data related to coal gangue spontaneous combustion in each of N historical coal gangue spontaneous combustion related materials, obtaining N feature data, and screening M different features based on the N feature data.

[0074] The feature screening module 2 is used for screening L features by investigating a large number of coal gangue spontaneous combustion related literatures, grabbing high frequency features related to coal gangue spontaneous combustion in all the literatures, and combining the M features and expert evaluation.

[0075] The preprocessing module 3 is used for aggregating feature data belonging to features in the L features in each of the N feature data, deleting or filling in the pre-processing of any feature data with missing values in the aggregated N feature data, and obtaining H feature data, each of which comprises L feature data.

[0076] The correlation calculation module 4 is used for calculating the correlation coefficient between each two features in the L features by using the H feature data, optimizing the L features according to the correlation coefficient, obtaining I features, and I≤L, each feature data containing I feature data. In this embodiment, none of the 9 features is deleted, and finally there are still 9 features, that is, I=L.

[0077] The weight calculation module 5 is used for investigating K articles related to coal gangue spontaneous combustion, respectively counting the frequency of each feature in the L features in the K articles, and calculating the feature weight of each feature.

[0078] The clustering module 6 is used for the H feature data, and the feature data of the features negatively correlated with coal gangue spontaneous combustion is inverted to obtain positively correlated feature data, combined with the feature weight, and the weighted WK-means clustering model is used to cluster the inverted H feature data, which is clustered into three clusters of not easy to spontaneous combustion, easy to spontaneous combustion and extremely easy to spontaneous combustion, and the clustering result is obtained.

[0079] The category acquisition module 7 is configured to perform multiple verifications on the clustering result, and obtain a final coal gangue spontaneous combustion category label of each feature data in the H feature data.

[0080] The embodiment of the present application further provides an electronic device, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored in the memory to execute the foregoing method.

[0081] The embodiment of the present application further provides a computer readable storage medium, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the foregoing method.

[0082] The present application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions stored therein for implementing various aspects of the present application.

[0083] Although the specific embodiments of the present application are described above, those skilled in the art should understand that these are only illustrative, and the protection scope of the present application is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of the present application, and these changes and modifications all fall within the protection scope of the present application.

Claims

1. A method for generating classification labels for coal gangue based on a clustering algorithm, characterized in that, include: S1. Collect multiple feature data related to spontaneous combustion of coal gangue from N historical materials related to spontaneous combustion of coal gangue in advance, obtain N feature data, and select M different features based on the N feature data; S2. By surveying a large number of literatures related to spontaneous combustion of coal gangue, we captured the high-frequency features related to spontaneous combustion of coal gangue in all literatures. At the same time, combined with M features, and after expert evaluation, L features were selected, where L < M. S3. Summarize the feature data of each of the N feature data that belong to the L features. For any feature data with missing values ​​in the summarized N feature data, perform deletion or imputation preprocessing to obtain H feature data, each of which includes L feature data. S4. Calculate the correlation coefficient between each pair of features in the L features using H feature data. Based on the correlation coefficient, optimize the L features by reducing and eliminating them to obtain I features, where I≤L. Each feature data contains I feature data. S5. Investigate K articles related to spontaneous combustion of coal gangue, and count the frequency of each feature in the K articles for each of the I features, and calculate the feature weight of each feature accordingly. In the S6 and H feature data, the feature data that is negatively correlated with the spontaneous combustion of coal gangue is reversed to obtain the feature data that is positively correlated. Combined with the feature weights, the weighted WK-means clustering model is used to cluster the H feature data after the reversed processing, and they are clustered into 3 clusters: not easy to spontaneously combust, easy to spontaneously combust, and extremely easy to spontaneously combust, to obtain the clustering results. S7. Perform multiple verifications on the clustering results to obtain the final category label of spontaneous combustion of coal gangue for each of the H feature data.

2. The method for generating coal gangue classification labels based on clustering algorithm as described in claim 1, characterized in that, in In S1, using words related to spontaneous combustion of coal gangue as keywords, the existing published literature in various databases is searched. From each retrieved literature, multiple feature data related to spontaneous combustion of coal gangue are found, and the multiple feature data obtained from any one literature is used as one feature data. Experimental data from multiple coal mines were obtained from the Ecological and Environmental Protection Bureau. The experimental data included multiple characteristic data related to spontaneous combustion of coal gangue. Multiple characteristic data obtained based on any one experimental data were used as one characteristic data. Obtain N feature data points, and then select M different features based on these N feature data points.

3. The method for generating coal gangue classification labels based on clustering algorithm as described in claim 1, characterized in that, In S3, the feature data of each of the N feature data points belonging to the L features is summarized to construct a first Excel table with N rows and L columns of feature data. For any row, it is analyzed whether there are any missing feature data. If there are no missing features, the row is kept. If there are missing features, further analysis is performed. If there are two or more missing feature data, the row is deleted directly. If there is only one missing feature data, the missing feature data is filled in. Based on this, a second Excel table with H rows and L columns of feature data is obtained.

4. The method for generating coal gangue classification labels based on clustering algorithms as described in claim 3, characterized in that, In S3, since there is only one missing feature data in this row, the regression imputation method is used to predict and fill in the missing feature data. Regression imputation methods include: S31. Select all row feature data without missing values ​​from the first Excel sheet as training samples, and use the missing feature of the row as the target variable and the other L-1 features as explanatory variables to establish a regression model. S32. The feature data corresponding to the explanatory variables in each row of the training sample is used as input, and the feature data corresponding to the target variable is used as output. The training sample is used to train the regression model to obtain the trained regression model. S33. Substitute the other L-1 feature data of this row into the trained regression model to predict the target variable value. The target variable value is used to fill in the feature data that is the missing feature of this row.

5. The method for generating coal gangue classification labels based on clustering algorithm as described in claim 1, characterized in that, in In S4, for H feature data, an H*L matrix is ​​constructed. The H*L matrix is ​​input into the Origin professional function plotting software, and the correlation coefficient between each pair of features is calculated using the Pearson correlation coefficient formula. Each calculated correlation coefficient is analyzed. If any correlation coefficient is greater than a set threshold, it is further analyzed whether the two features corresponding to that correlation coefficient are positively or negatively correlated with coal gangue spontaneous combustion. If the two features are positively and negatively correlated with coal gangue spontaneous combustion respectively, the two features are retained. If both features are positively or negatively correlated with coal gangue spontaneous combustion, one of the features is deleted. Thus, each feature data contains L-1 features. After this operation, no feature is deleted, and there are still L features in the end, i.e., I=L.

6. The method for generating coal gangue classification labels based on clustering algorithms as described in claim 5, characterized in that, In S6, S61, for the second Excel table containing H rows and L columns of feature data, the feature data of the feature that is negatively correlated with spontaneous combustion of coal gangue is reversed by the maximum value subtraction method to obtain the feature data that is positively correlated. The feature weight of each feature is read, and the feature data of H rows and L columns after the reversal process is weighted by the feature weight to obtain the weighted feature value of H rows and L columns. S62. Use data standardization processing methods to standardize the weighted feature values ​​of H rows and L columns to obtain standardized feature values ​​of H rows and L columns. S63. Set the number of clusters to 3 and the random seed to 42. Input the standardized feature values ​​of row H and column L into the weighted WK-means clustering model to obtain the clustering result label for each row. The clustering result label is not easily spontaneously combustible, easily spontaneously combustible, and extremely easily spontaneously combustible. Add the clustering result label as the L+1 column of the second Excel sheet to form H rows (L+1) columns of data.

7. The method for generating coal gangue classification labels based on clustering algorithms as described in claim 6, characterized in that, The three clusters are simplified to 0, 1, and 2, respectively; In S63, the simplified label of the clustering result is added as the L+2th column, forming H rows (L+2) columns of data.

8. The method for generating coal gangue classification labels based on clustering algorithm as described in claim 7, characterized in that, in In S7, S71, obtain the coordinates of the center point of each cluster from the weighted WK-means clustering model after clustering; S72. Output the H rows (L+2) columns of data containing the clustering results and the coordinates of the cluster centers of each cluster, and save the H rows (L+2) columns of data containing the clustering results to a new Excel file; S73. Use the t-SNE algorithm to reduce the dimensionality of the data in row H and column L to a two-dimensional space. Add the t-SNE dimensionality reduction result to a new Excel file to form new columns "t-SNE_1" and "t-SNE_2". Based on the t-SNE dimensionality reduction result, mark the data points of different clusters with different colors to visually demonstrate the clustering effect. Plot the cluster center points on the scatter plot and mark them with a red cross shape to highlight the core position of each cluster. S74. Install the matplotlib, seaborn, and scikit-learn libraries using the pip install command; S75. Use the calinski_harabasz_score function to calculate the Calinski-Harabasz exponent; S76. Display the clustering results as a scatter plot and present the Calinski-Harabasz index performance index as a bar chart.

9. A coal gangue classification label generation system based on clustering algorithm, characterized in that it includes: A data collection module is used to collect multiple feature data related to spontaneous combustion of coal gangue from N historical coal gangue spontaneous combustion related materials in advance, obtain N feature data, and filter out M different features based on N feature data; A feature filtering module is used to capture the high-frequency features related to spontaneous combustion of coal gangue from a large number of literatures on spontaneous combustion of coal gangue. At the same time, based on M features, L features are selected by experts, where L < M. A preprocessing module is used to summarize the feature data of each of the N feature data that belongs to the L features. For any feature data with missing values ​​in the summarized N feature data, the module performs deletion or imputation preprocessing to obtain H feature data, each of which includes L feature data. A correlation calculation module is used to calculate the correlation coefficient between each pair of features among L features using H feature data. Based on the correlation coefficient, the L features are reduced and optimized to obtain I features, where I≤L, and each feature data contains I feature data. A weight calculation module is used to investigate K articles related to spontaneous combustion of coal gangue, and to count the frequency of each feature in the K articles for each of the I features, thereby calculating the feature weight of each feature. A clustering module is used to reverse the feature data that is negatively correlated with spontaneous combustion of coal gangue in H feature data to obtain positively correlated feature data. Combined with feature weights, a weighted WK-means clustering model is used to cluster the H feature data after reversal processing into 3 clusters: not easily spontaneously combusted, easily spontaneously combusted, and extremely easily spontaneously combusted, to obtain the clustering results. A category acquisition module is used to perform multiple verifications on the clustering results and obtain the final category label of spontaneous combustion of coal gangue for each feature data in H feature data.

10. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in memory to execute the method of any one of claims 1-7.

11. A computer-readable storage medium storing computer program instructions thereon, characterized in that, When computer program instructions are executed by a processor, they implement the method of any one of claims 1-7.

Images