A mine filling body spontaneous combustion tendency analysis method based on rapid identification
By using spectral analysis and neural network simulation to simulate the composition of mine backfill materials, the problem of rapidly identifying spontaneous combustion trends on-site has been solved, enabling accurate prediction of spontaneous combustion risks and improving the level of mine safety management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ANHUI UNIV OF SCI & TECH
- Filing Date
- 2025-09-24
- Publication Date
- 2026-07-07
AI Technical Summary
Existing technologies struggle to quickly and accurately identify the complex components and interactions of filling materials at the mine site, leading to inaccurate predictions of spontaneous combustion trends and failing to meet the real-time response requirements of mine safety management. This is especially true in deep mining environments where risk prediction is even more uncertain.
By acquiring on-site sample data of mine backfill materials, using spectral analysis algorithms to extract mineral proportions and additive content characteristics, combining neural networks to simulate the oxidation reaction pathways between minerals, screening high-risk interaction combinations, optimizing temperature and gas concentration correlation subsequences, generating dynamic trend indicators, determining the type of spontaneous combustion trend, and outputting accurate prediction results.
It significantly improves the accuracy and real-time performance of spontaneous combustion risk prediction, reduces the probability of spontaneous combustion accidents in mineral filling bodies, and provides technical support for safe production in mines.
Smart Images
Figure CN121090437B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of mine backfill material analysis technology, and in particular to a method for analyzing the spontaneous combustion trend of mine backfill based on rapid identification. Background Technology
[0002] Research on spontaneous combustion of mine backfill materials is a key area in mine safety, as it directly impacts the safety and sustainability of mine production. Backfill materials, used to fill goaf areas, not only support ground pressure but also prevent fires caused by spontaneous combustion, which could result in casualties, equipment damage, and environmental pollution. In recent years, with increasing mining depth and exposure to complex geological conditions, the risk of spontaneous combustion of backfill materials has significantly increased, necessitating efficient and accurate analytical methods to predict and control potential threats. Technological advancements in this field can not only protect miners' lives but also reduce economic losses and maintain the stability of mine production.
[0003] Existing methods have significant limitations in addressing the risk of spontaneous combustion in backfill materials. Traditional detection methods rely heavily on manual sampling and laboratory analysis, which are time-consuming and complex, failing to meet the demands of rapid response in mines. Furthermore, many methods focus only on single factors, such as temperature or gas concentration, neglecting the interactive effects of the complex composition of the backfill material itself. This one-sidedness leads to incomplete predictions and an inability to accurately reflect the dynamic changes in spontaneous combustion trends. For example, in some deep mines, the composition of backfill materials is complex due to varying geological conditions, making it difficult for traditional methods to quickly detect their flammability, thus missing the optimal window for early intervention.
[0004] The primary technical challenge lies in rapidly identifying the complex composition of the backfill material. Backfill materials typically consist of various minerals and additives, with varying proportions of sulfides, organic matter, and oxidizing minerals. The distribution and content of these components directly determine the potential risk of spontaneous combustion. However, existing technologies struggle to achieve rapid and accurate detection of these components in the field environment, especially under the high temperature and pressure conditions of deep mines, where conventional testing equipment cannot operate stably. This further exacerbates another core challenge: accurately determining the interaction between different components and its impact on the spontaneous combustion trend, given the complex composition. For example, a high sulfide content may trigger an oxidation reaction, but the presence of organic matter may accelerate the reaction process. Existing technologies cannot effectively distinguish the relative contributions of these components, leading to biased predictions.
[0005] In real-world mining scenarios, real-time monitoring and rapid decision-making are crucial. For example, after filling a goaf in a mine, workers need to quickly determine if the backfill material poses a spontaneous combustion risk in order to adjust ventilation or take other preventative measures. However, due to a lack of technology for rapid component identification, on-site personnel often rely on outdated laboratory analysis results. This not only delays preventative measures but may also lead to spontaneous combustion accidents due to misjudgments. Therefore, how to quickly and accurately identify the flammable properties of backfill materials in the field environment and accurately predict spontaneous combustion trends based on the interactions of complex components has become a key issue in mine safety management.
[0006] This problem is particularly prominent in deep mines, where the high temperatures and pressures of the deep environment further amplify the difficulty of component analysis. Simultaneously, the complex interactions of components make the prediction of spontaneous combustion risk even more uncertain. Research urgently needs to overcome the technical bottleneck of how to quickly obtain information on the composition of backfill materials on-site and accurately analyze their flammability. Only by solving these problems can we provide mines with real-time and reliable predictions of spontaneous combustion trends, thereby improving overall safety management. Summary of the Invention
[0007] To address the technical problems existing in the prior art, this invention proposes a method for analyzing the spontaneous combustion trend of mine backfill bodies based on rapid identification, which effectively reduces the probability of spontaneous combustion accidents of mineral backfill bodies and provides technical support for safe production.
[0008] To achieve the above objectives, this invention provides a method for analyzing the spontaneous combustion trend of mine backfill bodies based on rapid identification, comprising:
[0009] Obtain on-site sample data of mine backfill material, process the sample data using a spectral analysis algorithm to extract mineral proportions, obtain component distribution information, and, based on the component distribution information, use a neural network to simulate the oxidation reaction paths between different minerals to determine the interaction intensity value;
[0010] Based on the interaction strength value, a subset of sulfide and organic matter ratios is obtained, and high-risk interaction combinations are identified through logical screening rules to obtain risk classification labels;
[0011] For the aforementioned risk classification label, a flammable characteristic parameter sequence is extracted from historical spontaneous combustion data. From the flammable characteristic parameter sequence, a temperature and gas concentration correlation subsequence is obtained, and the subsequence is optimized to obtain a dynamic change trend index.
[0012] Based on the dynamic trend indicators, the spontaneous combustion trend type is determined, an early warning level label is obtained, and based on the early warning level label, the mine site environmental parameters are integrated to determine the final spontaneous combustion trend prediction result.
[0013] Compared with the prior art, the present invention has the following advantages and technical effects:
[0014] This invention addresses the problem of spontaneous combustion risk assessment for mineral-filled bodies in complex environments. It acquires on-site sample data and uses spectral analysis algorithms to extract mineral proportions and additive content characteristics, generating component distribution information. Then, it utilizes neural networks to simulate oxidation reaction pathways between minerals, determining the intensity of interactions and screening out high-risk subsets of sulfide and organic matter ratios. By matching historical spontaneous combustion data, it extracts flammability characteristic parameter sequences and performs secondary spectral optimization on temperature and gas concentration correlation subsequences to generate dynamic trend indicators, determining whether the spontaneous combustion trend is accelerating. Finally, it integrates on-site environmental parameters to output accurate spontaneous combustion prediction results. This invention significantly improves the accuracy and real-time performance of spontaneous combustion risk prediction, effectively reducing the probability of spontaneous combustion accidents in mineral-filled bodies and providing technical support for safe production. Attached Figure Description
[0015] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:
[0016] Figure 1 This is a flowchart of a method for analyzing the spontaneous combustion trend of mine backfill based on rapid identification, according to an embodiment of the present invention. Detailed Implementation
[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0018] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.
[0019] This embodiment proposes a method for analyzing the spontaneous combustion trend of mine backfill bodies based on rapid identification, such as... Figure 1 ,include:
[0020] Obtain on-site sample data of mine backfill material, process the sample data using a spectral analysis algorithm to extract mineral proportions, obtain component distribution information, and, based on the component distribution information, use a neural network to simulate the oxidation reaction paths between different minerals to determine the interaction intensity value;
[0021] Based on the interaction strength value, a subset of sulfide and organic matter ratios is obtained, and high-risk interaction combinations are identified through logical screening rules to obtain risk classification labels;
[0022] For the aforementioned risk classification label, a flammable characteristic parameter sequence is extracted from historical spontaneous combustion data. From the flammable characteristic parameter sequence, a temperature and gas concentration correlation subsequence is obtained, and the subsequence is optimized to obtain a dynamic change trend index.
[0023] Based on the dynamic trend indicators, the type of spontaneous combustion trend is determined, and an early warning level label is obtained. Through the early warning level label, the integrated environmental parameters of the mine site are obtained to determine the final spontaneous combustion trend prediction result.
[0024] Further, obtaining the component distribution information includes:
[0025] Acquire field sample data of mine backfill material, collect raw spectral data through sensors to obtain spectral dataset, process the spectral dataset using spectral analysis algorithms, analyze spectral band characteristics, extract mineral proportion characteristics, and determine feature dataset;
[0026] If the mineral proportion feature value in the feature dataset exceeds the preset feature threshold, the feature dataset is dimensionality reduced by principal component analysis algorithm to obtain a dimensionality-reduced feature set. Based on the dimensionality-reduced feature set, the mineral proportion and additive content are classified by support vector machine algorithm to determine the classification result set.
[0027] The classification result set is processed by a clustering analysis algorithm to determine the component distribution information and obtain a component distribution dataset. If there are outliers in the component distribution dataset, the outliers are filtered by a preset anomaly detection rule to obtain a filtered component distribution dataset. The component distribution information is generated based on the filtered component distribution dataset.
[0028] Preferably, spectral analysis is widely used in the composition analysis of filling materials. High-dimensional spectral datasets can be obtained by acquiring raw spectral data through sensors. For example, when collecting filling samples from a mine in the field, sensors acquire spectral reflectance data in the 400-2500 nm wavelength range, forming a spectral dataset containing hundreds of bands. When processing this dataset, spectral analysis algorithms can use partial least squares regression to analyze characteristic bands such as the clay mineral absorption peak at 900 nm and the moisture absorption peak at 1900 nm, extracting mineral proportion features such as clay and quartz content, and additive content features such as the proportion of cement or chemical stabilizers, generating a feature dataset. It should be noted that feature extraction needs to be combined with spectral library comparison to ensure accuracy. If the mineral proportion feature values in the feature dataset exceed a preset feature threshold, for example, if the clay content exceeds 30%, then dimensionality reduction is performed using principal component analysis (PCA). Specifically, PCA can compress the high-dimensional feature dataset into a few principal components, retaining more than 90% of the data variance. For example, the original dataset contains 100 band features, which are reduced to 5 principal components after principal component analysis, forming a dimensionality-reduced feature set. This process reduces computational complexity and improves subsequent classification efficiency. Based on the dimensionality-reduced feature set, the support vector machine algorithm can classify mineral proportions and additive contents. For example, based on clay content and cement proportion, samples can be divided into categories such as high clay-low cement and low clay-high cement. The classification result set reflects differences in material properties; for example, high clay samples may affect the strength of the infill.
[0029] Support Vector Machines (SVMs) employ radial basis kernel functions to optimize classification boundaries and improve classification accuracy. Clustering analysis algorithms further process the classification result set to determine component distribution information. For example, using the K-means clustering algorithm, the classification result set is divided into three clusters, representing high-clay, high-cement, and balanced-ratio backfill materials, respectively, generating a component distribution dataset. This process reveals the spatial distribution patterns of material components, which helps optimize mine backfilling processes. If outliers exist in the component distribution dataset, such as a sample with an abnormally high clay content of 50%, they can be filtered using preset anomaly detection rules, such as those based on the 3σ criterion, to remove outlier samples and obtain a filtered component distribution dataset. The filtered dataset better reflects the true material characteristics and avoids outliers interfering with the analysis results. Based on the filtered component distribution dataset, component distribution information is generated. For example, analysis shows that the clay content in mine backfill is mainly concentrated between 20% and 30%, and the cement content is concentrated between 5% and 10%, which determines the final material composition analysis results. This result guides ratio optimization, reduces costs, and improves the stability of the backfill.
[0030] The above method, through multi-step collaborative analysis, ensures the accuracy of the results, provides a scientific basis for mine backfilling technology, and significantly improves engineering efficiency and safety.
[0031] Furthermore, based on the component distribution information, a neural network is used to simulate the oxidation reaction pathways between different minerals to determine the interaction strength values, including:
[0032] Chemical reaction data from component distribution information is collected by sensors to obtain raw reaction datasets. Based on the raw reaction datasets, data preprocessing rules are used to clean and standardize the data to obtain standard reaction datasets.
[0033] If there are outliers in the standard reaction dataset, they are filtered by a preset anomaly detection rule to obtain a filtered reaction dataset. Based on the filtered reaction dataset, a neural network algorithm is used to extract reaction path features to obtain a feature reaction dataset.
[0034] Based on the characteristic reaction dataset, the oxidation reaction pathways between different minerals are analyzed to obtain a set of reaction pathways. If the path feature values in the set of reaction pathways exceed a preset path feature threshold, the reaction pathways are classified using a clustering analysis algorithm to obtain a classified path dataset.
[0035] The interaction strength value is calculated based on the classification path dataset.
[0036] Specifically, when sensors collect chemical reaction data from on-site samples, molecular vibration information during the reaction process can be obtained using an infrared spectrometer to form a raw reaction dataset. This raw dataset may contain noise or incomplete data, requiring preprocessing. Data cleaning removes abnormal peaks caused by sensor jitter, while standardization normalizes the data to the 0-1 range, resulting in a standard reaction dataset. It should be noted that outlier detection can employ a method based on the absolute deviation of the median. If the reaction rate of a sample deviates from the median by more than three times, it is considered an anomaly and filtered, generating a filtered reaction dataset.
[0037] In one embodiment, a convolutional neural network (CNN) is used to extract reaction path features based on the filtered reaction dataset. The CNN captures reaction rate variation patterns through multiple convolutional kernels, extracting features such as peak oxidation reaction rates and reaction durations to form a feature-based reaction dataset. For example, a mine sample shows that the peak oxidation rate of clay minerals is 0.02 mol / s, and the reaction duration is 120 seconds. These features reflect the differences in reactivity among different minerals. Based on the feature-based reaction dataset, when analyzing oxidation reaction paths among different minerals, the focus can be on the oxidation reaction between clay and quartz. Clay may preferentially oxidize to form alumina, while quartz reacts more slowly, forming silica. The reaction path set can contain multiple paths, such as a rapid clay oxidation path and a slow quartz oxidation path. It should be noted that if a path feature value, such as a reaction rate, exceeds 0.03 mol / s, the path is classified using a K-means clustering algorithm. For example, the paths are divided into three categories: high activity, medium activity, and low activity, resulting in a categorized path dataset.
[0038] Based on the classification path dataset, the synergistic effect of oxidation reactions between minerals can be analyzed when calculating the interaction strength values. For example, clay oxidation may accelerate quartz reactions. The interaction strength value is quantified by the reaction rate ratio; for instance, the interaction strength between clay and quartz is 1.5, indicating that clay promotes quartz oxidation. The final strength distribution information shows that in samples with a clay content of 20-30%, the interaction strength is concentrated between 1.2 and 1.8, reflecting the stability of the material's chemical properties. This analysis helps optimize the design of infill material proportions.
[0039] Interaction intensity distribution information can guide the selection of mix proportions with low interaction intensity, reducing material property fluctuations caused by excessive reaction. Understandably, through multi-level analysis, from data acquisition to intensity distribution, the interconnected methods ensure the reliability of the analysis results and provide data support for optimizing mine backfilling processes.
[0040] Furthermore, based on the interaction strength values, a subset of sulfide and organic matter ratios is obtained, and high-risk interaction combinations are identified through logical filtering rules to obtain risk classification labels, including:
[0041] The ratio subset of sulfides and organic matter is obtained from the interaction intensity values. The ratio subset of organic matter is standardized using data filtering rules to generate a standardized ratio dataset. High-risk interaction combinations are marked by logical judgment rules to obtain a high-risk marked dataset.
[0042] Based on the high-risk labeled dataset, the high-risk interaction combinations are classified using the K-means clustering algorithm to obtain a classification interaction dataset. Chemical reaction data of high-risk combinations are extracted from the classification interaction dataset, and the reaction features are decomposed using the principal component analysis algorithm to obtain a feature decomposition dataset.
[0043] Based on the feature decomposition dataset, feature vectors of mineral interaction paths are obtained. The similarity of interaction paths is determined by vector comparison rules to obtain a set of similar paths. If the path similarity in the set of similar paths exceeds a preset similarity threshold, the co-occurrence relationship between paths is extracted by association analysis rules to obtain a co-occurrence relationship dataset.
[0044] Based on the co-occurrence relationship dataset, the distribution features of high-risk interaction combinations are extracted to obtain risk classification labels.
[0045] Specifically, when obtaining a subset of sulfide and organic matter ratios based on interaction strength values, the proportions of sulfides and organic matter in a sample can be analyzed using chemical composition detection equipment such as X-ray fluorescence spectrometry. Assuming a sample shows a sulfide content of 15% and an organic matter content of 25%, an initial subset of ratios is formed. This subset may vary depending on the sample collection point, requiring data filtering. Data filtering rules can be based on ratio ranges, for example, removing outlier data where the sulfide content is below 5% or above 50%, thus obtaining an effective subset of ratios. In one embodiment, when standardizing the comparison set, a min-max standardization method can be used to map the ratio values to the 0-1 range. For example, a sulfide content of 15% is standardized to 0.3, and an organic matter content of 25% is standardized to 0.5, generating a standardized ratio dataset. If a ratio threshold of 0.6 is set, when the standardized ratio of sulfides or organic matter exceeds 0.6, it is marked as a high-risk interaction combination using logical judgment rules. For example, a sample with a sulfide standardization ratio of 0.7 is marked as high risk, generating a high-risk labeled dataset.
[0046] Based on a high-risk labeled dataset, when classifying high-risk interaction combinations using the K-means clustering algorithm, the number of clusters can be set to 3, dividing the data into high, medium, and low-risk categories. Assuming a sample with a sulfide-to-organic ratio combination is clustered into a high-risk category with a reaction rate of 0.04 mol / s, a classification interaction dataset is generated. It should be noted that the clustering results reflect the differences in reactivity between different ratio combinations. In one embodiment, chemical reaction data of high-risk combinations are extracted from the classification interaction dataset. When using principal component analysis (PCA) to decompose reaction features, features such as reaction rate and reaction temperature can be extracted. For example, a high-risk combination shows a reaction rate of 0.04 mol / s and a reaction temperature of 200℃. After PCA, two principal feature vectors are obtained, reflecting the correlation between reaction rate and temperature, generating a feature decomposition dataset.
[0047] When obtaining feature vectors of mineral interaction paths based on the eigenvalue decomposition dataset, path similarity can be calculated using vector comparison rules. Assuming the cosine similarity between the feature vector of the clay-sulfide path and the quartz-organic path is 0.85, exceeding a preset threshold of 0.8, it is included in the similar path set. It should be noted that the similar path set reflects the commonalities of different mineral interactions. In one embodiment, when extracting co-occurrence relationships from the similar path set, the co-occurrence frequency between paths can be calculated using association analysis rules. For example, if the clay-sulfide path and the quartz-organic path appear simultaneously in 50% of the samples, a co-occurrence relationship dataset is generated. Based on the co-occurrence relationship dataset, statistical analysis methods such as analysis of variance are used to evaluate the stability of the distribution characteristics. For example, if the variance of a certain co-occurrence relationship is 0.02, below the threshold of 0.05, it is labeled as a risk classification label. The above analysis method, through multi-level data processing, ensures the logical rigor from proportional subsets to risk classification labels, which helps to accurately identify high-risk interaction combinations and optimize the design of mine backfill material proportions.
[0048] Furthermore, the flammability characteristic parameter sequence extracted from historical spontaneous combustion data includes:
[0049] The flammable characteristic parameter sequence is normalized using data standardization rules to obtain a standardized parameter dataset. If the parameter value in the standardized parameter dataset exceeds the preset parameter value, it is marked as a high-risk characteristic combination through logical judgment rules to obtain a high-risk characteristic marked dataset.
[0050] Based on the high-risk characteristic label dataset, the high-risk characteristic combinations are classified using a decision tree algorithm to obtain a classification characteristic dataset. The distribution pattern of the high-risk characteristic combinations is extracted, and the stability of the distribution pattern is determined using statistical analysis methods to obtain stable distribution labels.
[0051] Based on the stable distribution labels, the associated features of high-risk characteristic combinations are obtained. The similarity between features is calculated by vector comparison rules to obtain a set of similar features. The co-occurrence relationship between features is extracted by association analysis rules to obtain a co-occurrence relationship dataset.
[0052] Predictors of high-risk trait combinations are extracted from the co-occurrence relation dataset. Regression analysis is used to determine the weights of the predictors, resulting in a weight distribution dataset. The probability of occurrence of high-risk trait combinations is then predicted based on the weight distribution dataset.
[0053] Specifically, when normalizing the parameter sequence, the Z-score standardization method can be used to convert the parameter values into a distribution with a mean of 0 and a standard deviation of 1. For example, a volatile matter content of 20% is normalized to 0.4, and a calorific value of 5000 kJ / kg is normalized to 0.6, forming a standardized parameter dataset. If the preset threshold is 0.7, when any parameter value exceeds 0.7, it is marked as a high-risk characteristic combination through logical judgment. For example, a sample with a calorific value normalized to 0.8 is marked as high-risk, generating a high-risk characteristic label dataset. For example, when using a decision tree algorithm for classification based on the high-risk characteristic label dataset, the entropy value can be set as the splitting criterion to generate high, medium, and low-risk categories. Suppose a sample is classified as high-risk due to its high calorific value and low ignition temperature, generating a classification characteristic dataset.
[0054] In one embodiment, when extracting distribution patterns from a classification feature dataset, the stability of the distribution can be evaluated using statistical analysis methods, such as the chi-square test. For example, if a distribution pattern has a p-value of 0.03, which is less than the threshold of 0.05, it is labeled as a stable distribution. In another embodiment, based on the stable distribution labels, the similarity of associated features is calculated using vector comparison rules. For example, if the cosine similarity of the feature vectors of volatile substance content and ignition temperature is 0.9, which exceeds the threshold of 0.8, it is included in the similar feature set. It should be noted that the similar feature set reflects the inherent correlation between different parameters. For example, when extracting co-occurrence relationships from the similar feature set, the co-occurrence frequency of features can be calculated using the Apriori algorithm. Assuming that volatile substances and low ignition temperature appear simultaneously in 60% of the samples, a co-occurrence relationship dataset is generated.
[0055] When extracting predictive factors from a co-occurrence relation dataset, volatile matter content and calorific value can be selected as factors, and their weights determined through linear regression analysis. Assuming a weight of 0.6 for volatile matter and 0.4 for calorific value, a weighted distribution dataset is generated. It should be noted that the weighted distribution dataset can be used to predict the probability of occurrence of high-risk characteristic combinations, which helps optimize the selection and proportioning design of mine backfill materials.
[0056] Furthermore, based on the aforementioned dynamic trend indicators, the type of spontaneous combustion trend is determined, including:
[0057] If the rate of increase of the dynamic trend indicator is higher than the preset rate, the spontaneous combustion trend is determined to be accelerating.
[0058] Furthermore, obtain warning level labels, including:
[0059] Time series data is extracted from the dynamic trend indicators to obtain a time series dataset. Based on the time series dataset, the periodic components are decomposed using the Fourier transform algorithm to obtain a periodic component dataset.
[0060] If the frequency components of the periodic component dataset are higher than the preset frequency threshold, the frequency components are grouped by cluster analysis to obtain a frequency group dataset. Based on the frequency group dataset, the stability index of each group is calculated to obtain a stability index dataset. Based on the stability index dataset, the correlation strength between the dominant group and the temperature sequence is calculated by regression analysis to obtain a correlation strength dataset.
[0061] Based on the correlation strength dataset, highly correlated features are selected, and a decision tree algorithm is used to construct a warning level prediction model. The prediction model parameters are obtained, and the dynamic trend indicators are monitored in real time through the prediction model parameters to obtain real-time warning level labels.
[0062] Specifically, volatile gas concentration data of backfill material in a mine were collected, and daily concentration values were recorded over six months, forming a time-series dataset containing 180 data points. The data reflects the fluctuation pattern of gas concentration over time. The generation of the time-series dataset relies on high-precision sensors to ensure data continuity and accuracy. By decomposing periodic components using the Fourier transform algorithm, the time-series dataset can be divided into periodic components of different frequencies, forming a periodic component dataset. For example, Fourier transform analysis shows a significant periodic component with a frequency of 0.05 times / day, reflecting a concentration peak every 20 days. The periodic component dataset extracts periodic patterns by decomposing the original data, facilitating subsequent analysis. If the frequency components in the periodic component dataset exceed a preset frequency threshold, such as 0.04 times / day, cluster analysis is used to group the frequency components, generating a frequency-grouped dataset. For example, using the K-means clustering algorithm, the frequency components are divided into a high-frequency group and a low-frequency group. The high-frequency group includes components with 0.04-0.06 times / day, accounting for 70%, while the low-frequency group includes components with 0.01-0.03 times / day. This grouping method helps identify major fluctuation patterns. Based on the frequency grouping dataset, stability indices are calculated for each group, forming a stability index dataset.
[0063] Statistical analysis of the high-frequency group's variance revealed relatively small fluctuations and a stability index of 0.9, significantly higher than the preset threshold of 0.7. This stability index dataset reflects the reliability of the frequency groupings, providing a foundation for subsequent correlation analysis. If the dominant group in the stability index dataset meets the preset stability conditions, regression analysis is used to calculate the correlation strength between the dominant group and the temperature sequence, generating a correlation strength dataset. For example, regression analysis shows that the correlation strength between the frequency components of the high-frequency group and the ignition temperature sequence is 0.88, indicating a high correlation. The correlation strength dataset highlights the potential impact of key features on the warning level. Based on the correlation strength dataset, highly correlated features are selected, and a decision tree algorithm is used to construct a warning level prediction model, generating prediction model parameters.
[0064] The decision tree model uses gas concentration frequency and ignition temperature as primary features to generate a rule: when the frequency is higher than 0.05 times / day and the ignition temperature is lower than 300℃, the warning level is high-risk. The prediction model parameters are optimized through multiple rounds of training to ensure prediction accuracy. By using the prediction model parameters, dynamic trend indicators are monitored in real time to generate real-time warning level labels.
[0065] Furthermore, the final spontaneous combustion trend prediction results are determined, including:
[0066] Temperature, humidity, and air pressure data are acquired from on-site data acquisition equipment. An environmental parameter dataset is generated through data integration methods, and a time series analysis method is used to process the environmental parameter dataset to obtain a trend dataset.
[0067] Based on the change trend dataset, the main change patterns are decomposed using principal component analysis to obtain a key feature dataset. Dominant feature components are extracted from the key feature dataset. If the contribution rate of the dominant feature components is higher than a preset contribution rate threshold, the key feature dataset is grouped using cluster analysis to obtain a feature group dataset.
[0068] Based on the feature grouping dataset, the feature stability index of each group is calculated to obtain the stability index dataset. If the stability index of the dominant group in the stability index dataset meets the preset stability condition, the correlation strength between the dominant group and the historical spontaneous combustion records is calculated by regression analysis to obtain the correlation strength dataset.
[0069] Based on the correlation strength dataset, highly correlated feature components are selected to determine the final spontaneous combustion trend prediction result.
[0070] Specifically, temperature, humidity, and air pressure data are collected hourly using high-precision sensors on-site, continuously monitored for three months, forming an environmental parameter dataset containing 2160 data points. Temperature data reflects changes in heat within the filling material, humidity data indicates the impact of moisture content on spontaneous combustion, and air pressure data is related to gas diffusion. It is important to note that consistent sensor calibration must be ensured during data integration to avoid data deviations affecting subsequent analysis. If the fluctuation range of the environmental parameter dataset exceeds a preset threshold, such as temperature fluctuations exceeding 5°C or humidity changes exceeding 10%, time series analysis is required. For example, using the sliding window method to calculate the rate of change of temperature data, a daily temperature increase of 0.3°C over a certain period is found, indicating a potential heat accumulation trend. A trend dataset is thus generated, containing short-term and long-term change patterns for each parameter, highlighting the dynamic laws of heat accumulation. In one possible implementation, principal component analysis is used to decompose the key patterns in the trend dataset. For example, analysis showing that the combination of temperature and air pressure contributes 80% of the variance is identified as a key feature dataset. Dominant feature components, such as the temperature-air pressure combination, whose contribution rate exceeds a preset threshold of 70%, are further processed. The key feature dataset reflects the core impact of environmental parameters on spontaneous combustion, making it easier to focus on major risk factors.
[0071] Specifically, cluster analysis can group key feature datasets. For example, the K-means algorithm can be used to divide temperature-pressure combinations into high-risk and low-risk groups. The high-risk group includes data points with temperatures above 35°C and large pressure fluctuations, accounting for 60%. Feature grouping datasets reveal different risk patterns through grouping, providing a basis for subsequent stability analysis. For example, stability index datasets can be generated by calculating the variance of each group. The high-risk group has a smaller variance in its temperature-pressure combination, with a stability index of 0.85, higher than the preset threshold of 0.75, indicating that the features of this group are reliable. The stability index dataset lays the foundation for correlation analysis, ensuring that the analysis is based on stable data patterns. In one possible implementation, regression analysis is used to calculate the correlation strength between the dominant group and historical spontaneous combustion records. For example, the analysis shows that the correlation coefficient between the temperature-pressure combination of the high-risk group and spontaneous combustion events is 0.9, generating a correlation strength dataset. The correlation strength dataset highlights the predictive value of key features for spontaneous combustion, providing a basis for selecting highly correlated features. After selecting highly correlated feature components, the final spontaneous combustion prediction result can be determined.
[0072] Based on the temperature-pressure combination with a correlation strength of 0.9, and combined with historical spontaneous combustion thresholds, a high risk of spontaneous combustion is predicted for the current backfill material. The final prediction results provide data support for mine safety management and guide the implementation of prevention and control measures.
[0073] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for analyzing the spontaneous combustion trend of mine backfill bodies based on rapid identification, characterized in that, include: Obtain on-site sample data of mine backfill material, process the sample data using a spectral analysis algorithm to extract mineral proportions, obtain component distribution information, and, based on the component distribution information, use a neural network to simulate the oxidation reaction paths between different minerals to determine the interaction intensity value; Based on the interaction strength value, a subset of sulfide and organic matter ratios is obtained, and high-risk interaction combinations are identified through logical screening rules to obtain risk classification labels; For the aforementioned risk classification label, a flammable characteristic parameter sequence is extracted from historical spontaneous combustion data. From the flammable characteristic parameter sequence, a temperature and gas concentration correlation subsequence is obtained, and the subsequence is optimized to obtain a dynamic change trend index. Based on the dynamic trend indicators, the spontaneous combustion trend type is determined, an early warning level label is obtained, and based on the early warning level label, the mine site environmental parameters are integrated to determine the final spontaneous combustion trend prediction result. Specifically, based on the component distribution information, a neural network is used to simulate the oxidation reaction pathways between different minerals to determine the interaction strength values, including: Chemical reaction data from component distribution information is collected by sensors to obtain raw reaction datasets. Based on the raw reaction datasets, data preprocessing rules are used to clean and standardize the data to obtain standard reaction datasets. If there are outliers in the standard reaction dataset, they are filtered by a preset anomaly detection rule to obtain a filtered reaction dataset. Based on the filtered reaction dataset, a neural network algorithm is used to extract reaction path features to obtain a feature reaction dataset. Based on the characteristic reaction dataset, the oxidation reaction pathways between different minerals are analyzed to obtain a set of reaction pathways. If the path feature values in the set of reaction pathways exceed a preset path feature threshold, the reaction pathways are classified using a clustering analysis algorithm to obtain a classified path dataset. The interaction strength value is calculated based on the classification path dataset.
2. The method for analyzing the spontaneous combustion trend of mine backfill bodies based on rapid identification according to claim 1, characterized in that, Obtaining the component distribution information includes: Acquire field sample data of mine backfill material, collect raw spectral data through sensors to obtain spectral dataset, process the spectral dataset using spectral analysis algorithms, analyze spectral band characteristics, extract mineral proportion characteristics, and determine feature dataset; If the mineral proportion feature value in the feature dataset exceeds the preset feature threshold, the feature dataset is dimensionality reduced by principal component analysis algorithm to obtain a dimensionality-reduced feature set. Based on the dimensionality-reduced feature set, the mineral proportion and additive content are classified by support vector machine algorithm to determine the classification result set. The classification result set is processed by a clustering analysis algorithm to determine the component distribution information and obtain a component distribution dataset. If there are outliers in the component distribution dataset, the outliers are filtered by a preset anomaly detection rule to obtain a filtered component distribution dataset. The component distribution information is generated based on the filtered component distribution dataset.
3. The method for analyzing the spontaneous combustion trend of mine backfill bodies based on rapid identification as described in claim 1, characterized in that, Based on the interaction strength values, a subset of sulfide and organic matter ratios is obtained, and high-risk interaction combinations are identified through logical filtering rules to obtain risk classification labels, including: The ratio subset of sulfides and organic matter is obtained from the interaction intensity values. The ratio subset of organic matter is standardized using data filtering rules to generate a standardized ratio dataset. High-risk interaction combinations are marked by logical judgment rules to obtain a high-risk marked dataset. Based on the high-risk labeled dataset, the high-risk interaction combinations are classified using the K-means clustering algorithm to obtain a classification interaction dataset. Chemical reaction data of high-risk combinations are extracted from the classification interaction dataset, and the reaction features are decomposed using the principal component analysis algorithm to obtain a feature decomposition dataset. Based on the feature decomposition dataset, feature vectors of mineral interaction paths are obtained. The similarity of interaction paths is determined by vector comparison rules to obtain a set of similar paths. If the path similarity in the set of similar paths exceeds a preset similarity threshold, the co-occurrence relationship between paths is extracted by association analysis rules to obtain a co-occurrence relationship dataset. Based on the co-occurrence relationship dataset, the distribution features of high-risk interaction combinations are extracted to obtain risk classification labels.
4. The method for analyzing the spontaneous combustion trend of mine backfill bodies based on rapid identification as described in claim 1, characterized in that, After extracting the flammability characteristic parameter sequence from historical spontaneous combustion data, it includes: The flammable characteristic parameter sequence is normalized using data standardization rules to obtain a standardized parameter dataset. If the parameter value in the standardized parameter dataset exceeds the preset parameter value, it is marked as a high-risk characteristic combination through logical judgment rules to obtain a high-risk characteristic marked dataset. Based on the high-risk characteristic label dataset, the high-risk characteristic combinations are classified using a decision tree algorithm to obtain a classification characteristic dataset. The distribution pattern of the high-risk characteristic combinations is extracted, and the stability of the distribution pattern is determined using statistical analysis methods to obtain stable distribution labels. Based on the stable distribution labels, the associated features of high-risk characteristic combinations are obtained. The similarity between features is calculated by vector comparison rules to obtain a set of similar features. The co-occurrence relationship between features is extracted by association analysis rules to obtain a co-occurrence relationship dataset. Predictors of high-risk trait combinations are extracted from the co-occurrence relation dataset. Regression analysis is used to determine the weights of the predictors, resulting in a weight distribution dataset. The probability of occurrence of high-risk trait combinations is then predicted based on the weight distribution dataset.
5. The method for analyzing the spontaneous combustion trend of mine backfill bodies based on rapid identification as described in claim 1, characterized in that, Based on the aforementioned dynamic trend indicators, the type of spontaneous combustion trend is determined, including: If the rate of increase of the dynamic trend indicator is higher than the preset rate, the spontaneous combustion trend is determined to be accelerating.
6. The method for analyzing the spontaneous combustion trend of mine backfill bodies based on rapid identification as described in claim 5, characterized in that, Obtain warning level labels, including: Time series data is extracted from the dynamic trend indicators to obtain a time series dataset. Based on the time series dataset, the periodic components are decomposed using the Fourier transform algorithm to obtain a periodic component dataset. If the frequency components of the periodic component dataset are higher than the preset frequency threshold, the frequency components are grouped by cluster analysis to obtain a frequency group dataset. Based on the frequency group dataset, the stability index of each group is calculated to obtain a stability index dataset. Based on the stability index dataset, the correlation strength between the dominant group and the temperature sequence is calculated by regression analysis to obtain a correlation strength dataset. Based on the correlation strength dataset, highly correlated features are selected, and a decision tree algorithm is used to construct a warning level prediction model. The prediction model parameters are obtained, and the dynamic trend indicators are monitored in real time through the prediction model parameters to obtain real-time warning level labels.
7. The method for analyzing the spontaneous combustion trend of mine backfill bodies based on rapid identification as described in claim 6, characterized in that, Determine the final spontaneous combustion trend prediction results, including: Temperature, humidity, and air pressure data are acquired from on-site data acquisition equipment. An environmental parameter dataset is generated through data integration methods, and a time series analysis method is used to process the environmental parameter dataset to obtain a trend dataset. Based on the change trend dataset, the main change patterns are decomposed using principal component analysis to obtain a key feature dataset. Dominant feature components are extracted from the key feature dataset. If the contribution rate of the dominant feature components is higher than a preset contribution rate threshold, the key feature dataset is grouped using cluster analysis to obtain a feature group dataset. Based on the feature grouping dataset, the feature stability index of each group is calculated to obtain the stability index dataset. If the stability index of the dominant group in the stability index dataset meets the preset stability condition, the correlation strength between the dominant group and the historical spontaneous combustion records is calculated by regression analysis to obtain the correlation strength dataset. Based on the correlation strength dataset, highly correlated feature components are selected to determine the final spontaneous combustion trend prediction result.