Coal mine filling mining geological risk assessment method based on data mining
Through multi-source sensor data fusion and multivariate regression models, the inaccuracy problem of stratum movement analysis in traditional coal mine backfill mining has been solved, and real-time, accurate monitoring and optimization of stratum movement have been achieved, reducing geological risks and improving the safety and economy of mines.
Patent Information
- Application Number
- CN202510964895.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-14
- Publication Date
- 2025-09-26
AI Technical Summary
Existing technologies for analyzing stratum movement during coal backfill mining rely on traditional mine pressure monitoring methods, resulting in incomplete monitoring data and time lags, making it difficult to achieve accurate and real-time predictions of stratum stability and unable to effectively reduce geological risks.
Multi-source sensor data fusion technology is used to collect stratum movement monitoring data through sensors, perform data cleaning and normalization processing, establish a multivariate regression model, analyze time series characteristics, obtain sensitivity index, adjust filling parameters to adjust the stratum movement prediction model, and achieve real-time optimization.
It has achieved refined monitoring of stratum movement in coal mine backfill mining areas, improved the real-time and accuracy of data, and can capture the dynamic trend of stratum movement, reduce geological risks, and improve the safety and economy of mine mining.
Smart Images

Figure CN120706913A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of geological risk assessment, and in particular to a method for geological risk assessment of coal mine backfill mining based on data mining. Background Art
[0002] Coal mining technology is one of the most challenging research areas in mining engineering. In particular, backfill mining, as a green and efficient mining method, has garnered widespread attention in recent years. The core goal of backfill mining is to support the goaf with fill material, thereby reducing ground movement, lowering surface subsidence, and improving the mine environment. Analysis of ground movement trends is crucial for ensuring the safety of backfill mining. This requires the development of accurate predictive models based on mine pressure theory, geological structure analysis, and data mining techniques to enable real-time monitoring and optimization.
[0003] Current analysis of stratum movement during backfill mining relies primarily on traditional mine pressure monitoring methods and numerical simulation techniques. Mine pressure monitoring typically uses data from surface settlement monitoring points, rock movement sensors, and tunnel surrounding rock stress monitors. While this data can accurately reflect stratum movement trends, it is difficult to achieve accurate and real-time predictions of stratum stability due to the limited number of monitoring points, time lags in data acquisition, and the significant influence of the geological environment on data collection.
[0004] The main reasons for the lack of stratum movement analysis in the current backfill mining process include: incomplete monitoring data. Due to the complex geological environment of coal mine goafs, it is difficult to deploy monitoring points in some areas, resulting in limitations in monitoring data. Therefore, there is an urgent need for a geological risk assessment method for coal mine backfill mining based on data mining. Through multi-source data fusion, deep learning and data mining technology, the laws of stratum movement can be accurately modeled, and a solution for intelligent optimization of the backfill process can be provided to reduce geological risks and improve the safety and economy of mine mining. Summary of the Invention
[0005] In view of the shortcomings of the existing technology, the present invention provides a method for geological risk assessment of coal mine filling mining based on data mining, which solves the problems mentioned in the background technology.
[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for assessing geological risks of coal filling mining based on data mining, comprising the following steps:
[0007] S1, collect the ground movement monitoring data of the coal filling mining area through sensors and coal mine data acquisition equipment, and fit it into the original data set WA;
[0008] S2, performing data cleaning and normalization on the collected original data set WA to obtain the geological data set DW;
[0009] S3, using a multivariate regression model to establish a stratum movement prediction model, analyze the geological data set DW, and obtain the stratum settlement prediction value SP;
[0010] S4. Analyze the time series feature set Fs based on the predicted stratum settlement SP, determine the degree of influence of the features of the time series feature set Fs on stratum movement, and obtain the sensitivity index Im;
[0011] S5. Adjust the filling parameters, including the filling density pf and the mining rate Ve, according to the sensitivity index Im, to obtain the adjusted filling density npf and the adjusted mining rate nVe;
[0012] S6. According to the adjusted filling density npf and the adjusted mining rate nVe, the model parameters of the stratum movement prediction model are adjusted to obtain new model parameters, and the stratum settlement prediction amount SP is recalculated to obtain a new stratum settlement prediction amount nSP.
[0013] Preferably, S1, collecting stratum movement monitoring data including surface subsidence SG, stratum displacement Dr, mine pressure Pm, filling density pf and groundwater level change Hw through sensors and coal mine data acquisition equipment;
[0014] Among them, the surface subsidence SG is collected by deploying GPS high-precision measurement points on the surface of the mining area and using GNSS positioning technology;
[0015] The rock formation displacement Dr is acquired through microseismic sensors deployed in the filling area;
[0016] The rock formation displacement Dr is obtained by the following formula:
[0017]
[0018] Where, X i,t represents the displacement coordinate of the rock layer Xi at time t, X i,t-1 represents the displacement coordinate of the rock layer at time t-1;
[0019] The mine pressure Pm is obtained through the stress sensor installed in the filling body, specifically through the ratio of the mine pressure force to the area of the force-bearing area;
[0020] The filling density pf is obtained by measuring with a high-precision density meter;
[0021] The groundwater level change Hw is obtained through water level sensors and groundwater observation wells deployed in the mining area;
[0022] The groundwater level change Hw is obtained by the following formula:
[0023] Hw=Hwo-Hwt;
[0024] Where Hwo represents the initial position of the groundwater level, and Hwt represents the groundwater level at time t;
[0025] The collected surface settlement SG, rock displacement Dr, mine pressure Pm, filling density pf and groundwater level change Hw are fitted to obtain the original data set WA.
[0026] Preferably, S2 includes S21 and S22;
[0027] S21, clean the original data set WA to obtain the cleaned data set WC;
[0028] Data cleaning includes outlier detection and removal and missing value filling;
[0029] Outlier detection and removal: Outliers are detected and removed by using the boxplot method;
[0030] Missing value imputation fills missing values in the data through linear interpolation;
[0031] S22, normalizing the cleaned dataset WC to eliminate dimensional differences so that data of different dimensions can be calculated on the same scale, thereby obtaining a geological dataset DW;
[0032] The geological dataset DW is obtained by the following formula:
[0033]
[0034] Where DWd represents the d-th data item in the geological dataset DW, WCd represents the d-th data item in the cleaned dataset WC, minWCd represents the valley value of the d-th data item in the cleaned dataset WC, and maxWCd represents the peak value of the d-th data item in the cleaned dataset WC.
[0035] Preferably, S3 includes S31 and S32;
[0036] S31, extracting time series features from the geological data set DW, including formation subsidence rate Vsp, formation gradient change rate ΔDr, formation strain YPm, and groundwater permeability SpH, and forming a time series feature set Fs;
[0037] The formation subsidence rate Vsp is extracted using the sliding window technique and is obtained using the following formula:
[0038]
[0039] Where SG(t) represents the surface subsidence at time t, SG(t-1) represents the surface subsidence at time t-1, and Δt represents the time interval;
[0040] The formation gradient change rate ΔDr is obtained by the following formula:
[0041]
[0042] Where Δz represents the vertical depth interval between measurement points;
[0043] The rock formation strain YPm is obtained by the following formula:
[0044]
[0045] Where, Pm(t) represents the mine pressure at time t, and Pm(t-1) represents the mine pressure at time t-1;
[0046] Groundwater permeability Sp is obtained by the following formula:
[0047]
[0048] In the formula, Co represents the empirical coefficient and maxpf represents the maximum value of the filling density.
[0049] Preferably, S32, based on the acquired time series feature set Fs, a stratum movement prediction model is established by using a multivariate regression model to predict future surface subsidence and obtain a predicted stratum subsidence amount SP;
[0050] The predicted amount of ground settlement SP is obtained by the following formula:
[0051]
[0052] In the formula, M represents the total number of features, Fsj represents the jth feature in the time series feature set Fs, and α j represents the model parameter of the jth feature, and EX represents the error term.
[0053] Preferably, S4 includes S41 and S42;
[0054] S41, analyzing the contribution of the features in the time series feature set Fs to the formation settlement prediction value SP, quantifying the influence of the features on the formation settlement prediction value SP, obtaining the contribution rate Cj, and judging the influence of the features on the formation movement by the contribution rate Cj;
[0055] The contribution rate Cj is obtained by the following formula:
[0056]
[0057] In the formula, Cj represents the contribution rate, specifically the contribution rate of the j-th feature in the time series feature set Fs to the predicted value of formation settlement SP, represents partial derivative;
[0058] The influence of the features on the ground movement is obtained by matching the following methods:
[0059] When 0.5≤contribution ratio Cj<1, this indicates that the feature has an impact on the formation movement;
[0060] When 0.0<contribution rate Cj<0.5, it means that the feature has no effect on stratum movement, indicating that the influence of the feature is weak and its weight can be ignored or reduced;
[0061] Quantify the contribution rate Cj and obtain the characteristic impact factor Wj;
[0062] The characteristic impact factor Wj is obtained by the following formula:
[0063]
[0064] In the formula, Wj represents the feature impact factor, specifically the influence ratio of the jth feature in the time series feature set Fs among all features. represents the sum of contributions of all features.
[0065] Preferably, S42, calculating and obtaining a sensitivity index Im based on the obtained contribution rate Cj and characteristic influencing factor Wj, and comparing it with a preset sensitivity threshold TIm to determine the formation risk status;
[0066] The sensitivity index Im is obtained by the following formula:
[0067]
[0068] The formation risk status is obtained by matching in the following ways:
[0069] When the sensitivity index Im≤sensitivity threshold Tim, it means that there is no risk in the formation, and the current process is maintained and monitoring continues;
[0070] When the sensitivity index Im>sensitivity threshold TIM, it indicates that there is a risk in the formation and the filling parameters should be adjusted.
[0071] Preferably, S5, adjusting the filling density pf and the mining rate Ve according to the obtained sensitivity index Im, to obtain an adjusted filling density npf and an adjusted mining rate nVe;
[0072] When the sensitivity index Im>sensitivity threshold TIm, it means that there is a risk in the formation and the filling density pf is adjusted;
[0073] According to the sensitivity index Im and the sensitivity threshold TIm, the density adjustment amount Δpf is calculated;
[0074] The density adjustment Δpf is obtained by the following formula:
[0075] Δpf=k1*(Im-TIm);
[0076] Where, k1 represents the density adjustment coefficient;
[0077] Adjust the filling density pf using the obtained density adjustment amount Δpf to obtain an adjusted filling density npf;
[0078] The adjusted filling density npf is obtained by the following formula:
[0079] npf=pf+Δpf;
[0080] When the sensitivity index Im>sensitivity threshold TIm, it means that there is a risk in the formation and the mining rate Ve is adjusted;
[0081] Calculate the acquisition rate adjustment ΔVe based on the sensitivity index Im and the sensitivity threshold TIm;
[0082] The rate adjustment ΔVe is obtained by the following formula:
[0083] ΔVe=k2*(Im-TIm);
[0084] Where k2 represents the rate adjustment coefficient;
[0085] Adjust the mining rate Ve using the obtained rate adjustment amount ΔVe to obtain an adjusted mining rate nVe;
[0086] nVe=Ve-ΔVe.
[0087] When the adjusted mining rate nVe is too low, the packing density can be adjusted to improve the formation stability so that the mining rate is not too restricted;
[0088] When the adjusted mining rate nVe is too high, it may increase the risk of formation movement and should be optimized in combination with real-time monitoring data.
[0089] Preferably, S6 includes S61 and S62;
[0090] S61. Adjust the model parameter α of the formation movement prediction model according to the obtained adjusted filling density npf and the adjusted production rate nVe, and calculate and obtain the parameter adjustment amount Δα;
[0091] The parameter adjustment amount Δα is obtained by the following formula:
[0092]
[0093] Where tanh represents the hyperbolic tangent function, λ1 represents the density adjustment weight factor, and λ2 represents the rate adjustment weight factor;
[0094] The new model parameter nα is calculated by the parameter adjustment amount Δα and the model parameter α;
[0095] The new model parameter nα is obtained by the following formula:
[0096] nα=Δα+α.
[0097] Preferably, S62, bringing the obtained new model parameter nα into the stratum movement prediction model, re-predicting the stratum settlement prediction amount SP, and obtaining a new stratum settlement prediction amount nSP;
[0098] The new stratum settlement prediction nSP is obtained by the following formula:
[0099]
[0100] The present invention provides a method for assessing geological risks of coal mine backfill mining based on data mining, which has the following beneficial effects:
[0101] (1) Using multi-source sensor data fusion technology, we conduct detailed monitoring of stratum movement in the coal mine backfill mining area, including surface subsidence SG, rock displacement Dr, mine pressure Pm, backfill density pf, and groundwater level change Hw, and fit them into a high-precision raw data set WA. Data is collected through multiple monitoring methods to improve data coverage and ensure comprehensive monitoring of the mining area. Combined with time series modeling, the data is not only real-time but also captures the dynamic trends of stratum movement, avoiding the prediction errors caused by traditional methods that rely solely on static data.
[0102] Through data cleaning and normalization, noise and outliers are eliminated, and a standardized geological dataset (DW) is constructed to ensure the stability of subsequent modeling calculations. Removing outliers and filling missing data improves data consistency, making it more suitable for regression model analysis. Normalization unifies data from different physical units, improving the accuracy of multivariate model calculations.
[0103] (2) Using sliding window technology, gradient calculation, and permeability analysis methods, we extract time series features of key variables such as formation subsidence rate Vsp, formation gradient change rate ΔDr, rock strain YPm, and groundwater permeability SpH, and construct a time series feature set Fs. Traditional methods usually analyze data based on a single time point, while this method extracts subsidence rate using sliding window technology, allowing the prediction model to consider the impact of historical data on future subsidence trends and improve the rationality of time series modeling. Time series features such as formation subsidence rate Vsp, formation gradient change rate ΔDr, rock strain YPm, and groundwater permeability SpH can respectively describe the changing trends of surface, underground rock formations, mine pressure, and hydrological conditions, making the prediction model more comprehensive and accurate.
[0104] (3) The characteristic influence factor Wj calculation method is used to normalize the contribution rate of each characteristic to ensure that the influence of each characteristic in the model is balanced. In multivariate analysis, some characteristics may dominate the calculation due to different dimensions or numerical ranges. For example, if the mine pressure data value is large and the settlement rate value is small, if normalization is not performed, the contribution of the mine pressure may be incorrectly amplified. This method uses characteristic influence factor Wj normalization to ensure that all characteristics have consistent influence weights on a relative scale.
[0105] Compared to traditional single-feature analysis, this method simultaneously considers the interactive effects of multiple features, such as the relationship between groundwater permeability and packing density, improving prediction accuracy. Based on the analysis of characteristic influencing factors, packing parameters can be adjusted more precisely. For example, when the contribution of groundwater permeability (SpH) is high, it indicates that the packing density (ρf) needs to be appropriately increased to reduce the impact of permeability on formation stability.
[0106] (4) The updated new model parameters nα are used to recalculate the predicted stratum settlement SP to ensure that the prediction results can reflect the optimized filling process and mining strategy in real time. Traditional prediction methods are usually based on static models trained with historical data, while this method can adjust the model in real time according to the latest process parameters so that it always adapts to the latest working conditions of the mining area. When the mining rate Ve decreases, the release rate of the mine pressure will also change. If the old parameters are still used for prediction, the results may have large errors. However, this method can adjust the model parameters according to the latest process to ensure more accurate predictions. Since this method can continuously update the prediction model, mine managers can conduct safety assessments based on the latest prediction results, identify potential risks in advance, formulate scientific and reasonable mining plans, and improve mine safety. BRIEF DESCRIPTION OF THE DRAWINGS
[0107] Figure 1 This is a flowchart diagram of a method for geological risk assessment of coal filling mining based on data mining according to the present invention;
[0108] Figure 2 This is a schematic diagram of the model parameter acquisition process of the present invention;
[0109] Figure 3 This is a diagram of the stratum temporal characteristics and settlement prediction trends of the present invention. DETAILED DESCRIPTION
[0110] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0111] Example 1
[0112] The present invention provides a method for assessing geological risks of coal filling mining based on data mining. Figure 1-Figure 3 , including the following steps:
[0113] S1, collect the ground movement monitoring data of the coal filling mining area through sensors and coal mine data acquisition equipment, and fit it into the original data set WA;
[0114] S2, performing data cleaning and normalization on the collected original data set WA to obtain the geological data set DW;
[0115] S3, using a multivariate regression model to establish a stratum movement prediction model, analyze the geological data set DW, and obtain the stratum settlement prediction value SP;
[0116] S4. Analyze the time series feature set Fs based on the predicted stratum settlement SP, determine the degree of influence of the features of the time series feature set Fs on stratum movement, and obtain the sensitivity index Im;
[0117] S5. Adjust the filling parameters, including the filling density pf and the mining rate Ve, according to the sensitivity index Im, to obtain the adjusted filling density npf and the adjusted mining rate nVe;
[0118] S6. According to the adjusted filling density npf and the adjusted mining rate nVe, the model parameters of the stratum movement prediction model are adjusted to obtain new model parameters, and the stratum settlement prediction amount SP is recalculated to obtain a new stratum settlement prediction amount nSP.
[0119] In this example, multi-source sensor data fusion technology is used to perform detailed monitoring of stratum movement in coal mine backfill mining areas. This includes surface subsidence SG, rock displacement Dr, mine pressure Pm, backfill density pf, and groundwater level change Hw, and then fits these data into a high-precision raw dataset WA. Data collection through multiple monitoring methods improves data coverage and ensures comprehensive monitoring of the mining area. Combined with time series modeling, this data is not only real-time but also captures dynamic trends in stratum movement, avoiding the prediction errors caused by traditional methods that rely solely on static data.
[0120] Through data cleaning and normalization, noise and outliers are eliminated, and a standardized geological dataset (DW) is constructed to ensure the stability of subsequent modeling calculations. Removing outliers and filling missing data improves data consistency, making it more suitable for regression model analysis. Normalization unifies data from different physical units, improving the accuracy of multivariate model calculations.
[0121] A multivariate regression model was used to analyze the geological dataset DW, establishing a stratum subsidence prediction model and obtaining the stratum subsidence prediction SP. This model accurately captures the nonlinear characteristics of stratum subsidence. Compared to traditional methods such as finite element analysis, this data-driven approach avoids cumbersome parameter assumptions and improves prediction flexibility. Combining rock pressure theory with data mining techniques, the regression model is trained using both historical and real-time data, resulting in predictions that are more accurate than actual conditions.
[0122] By analyzing the time series characteristics, the sensitivity index Im of the characteristics to stratum movement is calculated, and the influence of filling density pf and mining rate Ve on stratum subsidence is judged.
[0123] By calculating the contribution of each characteristic variable to the predicted settlement value (SP), we avoid relying solely on empirical judgment in traditional methods and instead quantify the effects of various factors through data analysis. This allows us to determine which parameter adjustments have the most significant impact on settlement, thereby prioritizing the optimization of key variables and improving filling results.
[0124] Example 2
[0125] This embodiment is explained in Example 1, please refer to Figure 1 ,Specifically: S1, collect stratum movement monitoring data through sensors and coal mine data ,acquisition equipment, including surface settlement SG, rock stratum displacement Dr, ,mine pressure Pm, filling density pf and groundwater level change Hw;
[0126] Among them, the surface subsidence SG is collected by deploying GPS high-precision measurement points on the surface of the mining area and using GNSS positioning technology;
[0127] The rock formation displacement Dr is acquired through microseismic sensors deployed in the filling area;
[0128] The rock formation displacement Dr is obtained by the following formula:
[0129]
[0130] Where, X i,t represents the displacement coordinate of the rock layer Xi at time t, X i,t-1 represents the displacement coordinate of the rock layer at time t-1;
[0131] The mine pressure Pm is obtained through the stress sensor installed in the filling body, specifically through the ratio of the mine pressure force to the area of the force-bearing area;
[0132] The filling density pf is obtained by measuring with a high-precision density meter;
[0133] The groundwater level change Hw is obtained through water level sensors and groundwater observation wells deployed in the mining area;
[0134] The groundwater level change Hw is obtained by the following formula:
[0135] Hw=Hwo-Hwt;
[0136] Where Hwo represents the initial position of the groundwater level, and Hwt represents the groundwater level at time t;
[0137] The collected surface settlement SG, rock displacement Dr, mine pressure Pm, filling density pf and groundwater level change Hw are fitted to obtain the original data set WA.
[0138] S2 includes S21 and S22;
[0139] S21, clean the original data set WA to obtain the cleaned data set WC;
[0140] Data cleaning includes outlier detection and removal and missing value filling;
[0141] Outlier detection and removal: Outliers are detected and removed by using the boxplot method;
[0142] Missing value imputation fills missing values in the data through linear interpolation;
[0143] S22, normalizing the cleaned dataset WC to eliminate dimensional differences so that data of different dimensions can be calculated on the same scale, thereby obtaining a geological dataset DW;
[0144] The geological dataset DW is obtained by the following formula:
[0145]
[0146] Where DWd represents the d-th data item in the geological dataset DW, WCd represents the d-th data item in the cleaned dataset WC, minWCd represents the valley value of the d-th data item in the cleaned dataset WC, and maxWCd represents the peak value of the d-th data item in the cleaned dataset WC.
[0147] This embodiment utilizes a variety of high-precision sensor equipment, combining high-precision GPS measurement points, GNSS positioning technology, microseismic sensors, stress sensors, high-precision densitometers, and water level sensors to achieve comprehensive, multi-dimensional ground movement monitoring. Different sensors provide precise measurements for different monitoring indicators, ensuring data integrity. This improves measurement accuracy and reduces human error: GPS and GNSS technologies can provide high-precision measurements at the sub-meter or centimeter level, while microseismic sensors can monitor minute rock formation displacements, enhancing measurement sensitivity.
[0148] Compared with traditional manual inspections or single-sensor monitoring methods, this method can collect surface subsidence, rock displacement, mine pressure, filling density and groundwater level changes in real time, providing more accurate basic data for subsequent analysis.
[0149] Data cleaning methods such as outlier detection and elimination, and missing value filling are used to eliminate abnormal data, fill missing data, and construct a high-quality cleaned data set WC. Boxplots are used to identify and eliminate outliers, avoiding interference from factors such as equipment failure and data noise on the analysis results, making the data more consistent with actual conditions.
[0150] Using linear interpolation to fill missing data prevents data analysis from being biased by missing individual data points, improving the stability of the prediction model. Compared to simply deleting missing values, this method uses interpolation to preserve the time series trend of the data as much as possible, improving the continuity and usability of the data.
[0151] Normalization converts data of different dimensions to the same scale, eliminating unit constraints. This allows different physical quantities to be calculated and compared within the same mathematical model, resulting in a standardized geological dataset (DW). Unnormalized data can cause certain features to dominate during model training, affecting prediction accuracy. However, normalization balances the influence of all variables, resulting in faster model convergence and more stable training results.
[0152] Normalization reduces the data's numerical range, improving computational accuracy while also reducing computational complexity, enabling more efficient computations for subsequent multivariate regression analysis and deep learning methods. Through optimized data collection, cleaning, and normalization, this embodiment generates higher-quality geological datasets (DWs), providing more reliable input data for subsequent formation movement prediction and filling parameter optimization.
[0153] Improved data quality enables multivariate regression models to be trained based on more accurate data, avoiding prediction biases caused by data anomalies or noise interference, and improving prediction accuracy. Accurate data can better reflect the geological conditions of the mining area and provide precise data support for subsequent decisions such as filling density optimization and mining rate adjustment. Reduced ground movement risk and improved coal mine safety: More accurate predictions can help managers identify potential areas of ground instability in advance, implement appropriate filling and mining measures, reduce the occurrence of geological disasters such as ground collapse and roadway deformation, and improve mine safety management.
[0154] Example 3
[0155] This embodiment is explained in Example 2, please refer to Figure 2 and Figure 3 ,Specifically: S3 includes S31 and S32;
[0156] S31, extracting time series features from the geological data set DW, including formation subsidence rate Vsp, formation gradient change rate ΔDr, formation strain YPm, and groundwater permeability SpH, and forming a time series feature set Fs;
[0157] The formation subsidence rate Vsp is extracted using the sliding window technique and is obtained using the following formula:
[0158]
[0159] Where SG(t) represents the surface subsidence at time t, SG(t-1) represents the surface subsidence at time t-1, and Δt represents the time interval;
[0160] The formation gradient change rate ΔDr is obtained by the following formula:
[0161]
[0162] Where Δz represents the vertical depth interval between measurement points;
[0163] The rock formation strain YPm is obtained by the following formula:
[0164]
[0165] Where, Pm(t) represents the mine pressure at time t, and Pm(t-1) represents the mine pressure at time t-1;
[0166] Groundwater permeability Sp is obtained by the following formula:
[0167]
[0168] In the formula, Co represents the empirical coefficient and maxpf represents the maximum value of the filling density.
[0169] S32. Based on the acquired time series feature set Fs, a stratum movement prediction model is established by using a multivariate regression model to predict the future surface subsidence and obtain a predicted stratum subsidence amount SP, as shown in Table 1.
[0170] The predicted amount of ground settlement SP is obtained by the following formula:
[0171]
[0172] In the formula, M represents the total number of features, Fsj represents the jth feature in the time series feature set Fs, and α j represents the model parameter of the jth feature, and EX represents the error term.
[0173] Table 1 Ground movement prediction data table:
[0174]
[0175] In this example, a sliding window technique, gradient calculation, and permeability analysis methods are used to extract time series features from key variables such as formation subsidence rate Vsp, formation gradient change rate ΔDr, formation strain YPm, and groundwater permeability SpH, thereby constructing a time series feature set Fs. Traditional methods typically analyze data based on a single point in time. However, this method uses sliding window technology to extract subsidence rates, allowing the prediction model to consider the impact of historical data on future subsidence trends, thereby improving the rationality of time series modeling.
[0176] Time series characteristics such as formation subsidence rate Vsp, formation gradient change rate ΔDr, formation strain YPm, and groundwater permeability SpH can respectively describe the changing trends of surface and subsurface rock formations, mine pressure, and hydrological conditions, making the prediction model more comprehensive and accurate. Compared with prediction methods that rely solely on formation subsidence data, this method comprehensively considers filling density, groundwater permeability, and formation stress state, making the prediction results more practical.
[0177] By constructing a multivariate regression model, we predict future ground subsidence predictions (SP) based on the time series feature set Fs, and incorporating an error correction mechanism to enhance the model's stability. This overcomes the limitations of traditional empirical formulas and improves the model's adaptability: Traditional ground subsidence prediction methods are typically based on empirical formulas. This method, however, employs a data-driven regression modeling approach that dynamically adapts to the geological conditions of different mining areas, improving prediction accuracy.
[0178] Introducing an error term, EX, into the regression model corrects for the deviation between the model's predicted value and the true value, improving prediction reliability. Due to the complex mining environment, data collection may be subject to influences such as measurement errors and environmental noise. Introducing the error term automatically corrects the model's predicted value, improving model robustness.
[0179] Traditional methods can lead to significant deviations in long-term predictions due to accumulated errors. However, this method uses an error correction mechanism to continuously adjust the model's predictions, bringing them closer to actual subsidence trends. Accurately predicting stratum subsidence trends can provide scientific decision-making support for filling processes and mine safety management.
[0180] Accurately predicting ground subsidence allows for adjustments to fill density and material ratios based on the predicted results, making the filling process more precise, reducing unnecessary material waste, and improving economic efficiency. Accurately predicting ground subsidence changes allows for proactive preventative measures to avoid problems such as ground collapse, roadway deformation, and support structure failure caused by rapid ground movement, thereby improving mine safety management.
[0181] Example 4
[0182] This embodiment is explained in Example 3, please refer to Figure 1 , specifically: S4 includes S41 and S42;
[0183] S41, analyzing the contribution of the features in the time series feature set Fs to the formation settlement prediction value SP, quantifying the influence of the features on the formation settlement prediction value SP, obtaining the contribution rate Cj, and judging the influence of the features on the formation movement by the contribution rate Cj;
[0184] The contribution rate Cj is obtained by the following formula:
[0185]
[0186] In the formula, Cj represents the contribution rate, specifically the contribution rate of the j-th feature in the time series feature set Fs to the predicted value of formation settlement SP, represents partial derivative;
[0187] The influence of the features on the ground movement is obtained by matching the following methods:
[0188] When 0.5≤contribution ratio Cj<1, this indicates that the feature has an impact on the formation movement;
[0189] When 0.0<contribution rate Cj<0.5, it means that the feature has no effect on stratum movement;
[0190] Quantify the contribution rate Cj and obtain the characteristic impact factor Wj;
[0191] The characteristic impact factor Wj is obtained by the following formula:
[0192]
[0193] In the formula, Wj represents the feature impact factor, specifically the influence ratio of the jth feature in the time series feature set Fs among all features. represents the sum of contributions of all features.
[0194] S42. Calculate the sensitivity index Im based on the obtained contribution rate Cj and characteristic influencing factor Wj, and compare it with the preset sensitivity threshold TIm to determine the formation risk status;
[0195] The sensitivity index Im is obtained by the following formula:
[0196]
[0197] The formation risk status is obtained by matching in the following ways:
[0198] When the sensitivity index Im≤sensitivity threshold Tim, it means that there is no risk in the formation, and the current process is maintained and monitoring continues;
[0199] When the sensitivity index Im>sensitivity threshold TIM, it indicates that there is a risk in the formation and the filling parameters should be adjusted.
[0200] In this example, partial derivatives are used to calculate the feature contribution rate Cj, quantifying the influence of each feature in the time series feature set Fs on the predicted stratum subsidence SP and ensuring the interpretability of the model. Different features have varying degrees of influence on stratum movement. For example, stratum subsidence rate, rock strain, and groundwater permeability may be primary factors, while some features have less influence on subsidence. Contribution rate analysis can scientifically quantify the importance of these features and avoid the uncertainty of human judgment.
[0201] When the contribution rate Cj is less than 0.5, it indicates that the feature has little effect on the prediction of stratum subsidence. Its weight in the model can be appropriately reduced, or even removed in some cases to reduce redundant calculations and improve the efficiency of the prediction model.
[0202] The characteristic impact factor (Wj) calculation method is used to normalize the contribution of each characteristic, ensuring a balanced impact within the model. In multivariate analysis, some characteristics may dominate the calculation due to differences in scale or numerical range. For example, if the rock pressure data is large while the settlement rate is small, the contribution of the rock pressure may be incorrectly exaggerated without normalization. This method uses the characteristic impact factor (Wj) to normalize the data, ensuring that all characteristics have consistent influence weights on a relative scale.
[0203] Compared to traditional single-feature analysis, this method simultaneously considers the interactive effects of multiple features, such as the relationship between groundwater permeability and packing density, improving prediction accuracy. Based on the analysis of characteristic influencing factors, packing parameters can be adjusted more precisely. For example, when the contribution of groundwater permeability (SpH) is high, it indicates that the packing density (ρf) needs to be appropriately increased to reduce the impact of permeability on formation stability.
[0204] Using the sensitivity index (Im) calculation method, we comprehensively assess the overall impact of various characteristics on formation stability and compare it with a preset sensitivity threshold (TIm) to accurately determine the formation's risk status. Traditional formation stability analysis relies primarily on the empirical judgment of geological engineers and lacks a unified quantitative standard. However, this method, by calculating the sensitivity index (Im), provides a clear numerical judgment standard, ensuring objectivity and consistency in risk assessment.
[0205] Optimizing the timing of filling parameter adjustments: When the sensitivity index Im ≤ the sensitivity threshold Tim, the current filling process can be maintained. However, when the sensitivity index Im > the sensitivity threshold Tim, timely filling parameter adjustments are necessary to prevent the risk of collapse caused by increased formation movement. This method regularly calculates the sensitivity index and dynamically assesses formation stability. By comprehensively calculating the contribution rate Cj, the characteristic influencing factor Wj, and the sensitivity index Im, this method provides a quantitative basis for optimizing filling density and production rate, improving the scientific and economic efficiency of the filling process.
[0206] When the sensitivity index Im is too high, it indicates that the current filling density may be insufficient to effectively support the formation and should be increased to improve formation stability. If Im exceeds a threshold, the mining rate needs to be reduced to minimize disturbance to the formation, reduce the risk of formation movement, and improve the long-term stability of the mining area. Compared to traditional methods that use fixed filling parameters, this method can precisely optimize filling parameters while maintaining formation stability, reducing unnecessary filling material consumption and improving economic efficiency.
[0207] Example 5
[0208] This embodiment is explained in Example 4. Please refer to Figure 1 Specifically: S5, adjusting the filling density pf and the mining rate Ve according to the obtained sensitivity index Im, to obtain an adjusted filling density npf and an adjusted mining rate nVe;
[0209] When the sensitivity index Im>sensitivity threshold TIm, it means that there is a risk in the formation and the filling density pf is adjusted;
[0210] According to the sensitivity index Im and the sensitivity threshold TIm, the density adjustment amount Δpf is calculated;
[0211] The density adjustment Δpf is obtained by the following formula:
[0212] Δpf=k1*(Im-TIm);
[0213] Where, k1 represents the density adjustment coefficient;
[0214] Adjust the filling density pf using the obtained density adjustment amount Δpf to obtain an adjusted filling density npf;
[0215] The adjusted filling density npf is obtained by the following formula:
[0216] npf=pf+Δpf;
[0217] When the sensitivity index Im>sensitivity threshold TIm, it means that there is a risk in the formation and the mining rate Ve is adjusted;
[0218] Calculate the acquisition rate adjustment ΔVe based on the sensitivity index Im and the sensitivity threshold TIm;
[0219] The rate adjustment ΔVe is obtained by the following formula:
[0220] ΔVe=k2*(Im-TIm);
[0221] Where k2 represents the rate adjustment coefficient;
[0222] Adjust the mining rate Ve using the obtained rate adjustment amount ΔVe to obtain an adjusted mining rate nVe;
[0223] nVe=Ve-ΔVe.
[0224] S6 includes S61 and S62;
[0225] S61. Adjust the model parameter α of the formation movement prediction model according to the obtained adjusted filling density npf and the adjusted production rate nVe, and calculate and obtain the parameter adjustment amount Δα;
[0226] The parameter adjustment amount Δα is obtained by the following formula:
[0227]
[0228] Where tanh represents the hyperbolic tangent function, λ1 represents the density adjustment weight factor, and λ2 represents the rate adjustment weight factor;
[0229] The new model parameter nα is calculated by the parameter adjustment amount Δα and the model parameter α;
[0230] The new model parameter nα is obtained by the following formula:
[0231] nα=Δα+α.
[0232] S62, bringing the obtained new model parameter nα into the stratum movement prediction model, re-predicting the stratum settlement prediction value SP, and obtaining a new stratum settlement prediction value nSP;
[0233] The new stratum settlement prediction nSP is obtained by the following formula:
[0234]
[0235] In this example, while traditional mining strategies typically rely on a fixed mining rate, this method automatically reduces the mining rate when the formation is unstable, mitigating the impact of varying rock pressure and minimizing the risk of tunnel deformation and surface subsidence. By calculating the rate adjustment ΔVe, the mining rate is ensured to meet production requirements while minimizing disturbance to the formation. Both filling density and mining rate adjustments are calculated based on sensitivity indices, preventing overfilling or excessive mining, improving fill material utilization, and reducing production costs.
[0236] The hyperbolic tangent function is used to adjust parameters and optimize the parameter α of the ground movement prediction model, enabling the model to more accurately adapt to the dynamic changes in filling and mining conditions. Traditional model parameter adjustment typically uses linear correction, while this method uses the tanh function for nonlinear parameter optimization, making the adjustment process smoother and avoiding large parameter changes that can lead to unstable predictions.
[0237] When the filling density and extraction rate undergo minor adjustments, the parameter changes are also minimal, ensuring model stability. However, when the risk of formation subsidence is high and requires more significant adjustments, the parameter changes are also larger, ensuring that the model can quickly adapt to the new working conditions. By adjusting the density and rate weighting factors, the model can automatically determine the impact of filling density and extraction rate on formation subsidence and adjust the parameters accordingly, ensuring higher prediction accuracy.
[0238] The updated model parameters nα are used to recalculate the predicted ground subsidence SP, ensuring that the prediction results reflect the optimized filling process and mining strategy in real time. Traditional prediction methods are typically based on static models trained with historical data. This method, however, can adjust the model in real time based on the latest process parameters, ensuring that it always adapts to the latest operating conditions in the mine. When the mining rate Ve decreases, the rate of release of the rock pressure also changes. If the old parameters are still used for prediction, the results may show large errors. However, this method can adjust the model parameters based on the latest process to ensure more accurate predictions. Because this method can continuously update the prediction model, mine managers can conduct safety assessments based on the latest prediction results, identify potential risks in advance, formulate scientific and reasonable mining plans, and improve mine safety.
[0239] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.
Claims
1. A method for geological risk assessment of coal filling mining based on data mining, characterized by: The following steps are involved: S1, collect the ground movement monitoring data of the coal filling mining area through sensors and coal mine data acquisition equipment, and fit it into the original data set WA; S2, performing data cleaning and normalization on the collected original data set WA to obtain the geological data set DW; S3, using a multivariate regression model to establish a stratum movement prediction model, analyze the geological data set DW, and obtain the stratum settlement prediction value SP; S4. Analyze the time series feature set Fs based on the predicted stratum settlement SP, determine the degree of influence of the features of the time series feature set Fs on stratum movement, and obtain the sensitivity index Im; S5. Adjust the filling parameters, including the filling density pf and the mining rate Ve, according to the sensitivity index Im, to obtain the adjusted filling density npf and the adjusted mining rate nVe; S6. According to the adjusted filling density npf and the adjusted mining rate nVe, the model parameters of the stratum movement prediction model are adjusted to obtain new model parameters, and the stratum settlement prediction amount SP is recalculated to obtain a new stratum settlement prediction amount nSP.
2. The method for geological risk assessment of coal mine backfill mining based on data mining according to claim 1, characterized in that: S1. Collect ground movement monitoring data, including surface subsidence SG, rock displacement Dr, mine pressure Pm, filling density pf, and groundwater level change Hw, through sensors and coal mine data acquisition equipment; Among them, the surface subsidence SG is collected by deploying GPS high-precision measurement points on the surface of the mining area and using GNSS positioning technology; The rock formation displacement Dr is acquired through microseismic sensors deployed in the filling area; The rock formation displacement Dr is obtained by the following formula: Where, X i,t represents the displacement coordinate of the rock layer Xi at time t, X i,t-1 represents the displacement coordinate of the rock layer at time t-1; The mine pressure Pm is obtained through the stress sensor installed in the filling body, specifically through the ratio of the mine pressure force to the area of the force-bearing area; The filling density pf is obtained by measuring with a high-precision density meter; The groundwater level change Hw is obtained through water level sensors and groundwater observation wells deployed in the mining area; The groundwater level change Hw is obtained by the following formula: Hw=Hwo-Hwt; Where Hwo represents the initial position of the groundwater level, and Hwt represents the groundwater level at time t; The collected surface settlement SG, rock displacement Dr, mine pressure Pm, filling density pf and groundwater level change Hw are fitted to obtain the original data set WA.
3. The method for geological risk assessment of coal filling mining based on data mining according to claim 1, characterized in that: S2 includes S21 and S22; S21, clean the original data set WA to obtain the cleaned data set WC; Data cleaning includes outlier detection and removal and missing value filling; Outlier detection and removal: Outliers are detected and removed by using the boxplot method; Missing value imputation fills missing values in the data through linear interpolation; S22, normalizing the cleaned dataset WC to eliminate dimensional differences so that data of different dimensions can be calculated on the same scale, thereby obtaining a geological dataset DW; The geological dataset DW is obtained by the following formula: Where DWd represents the d-th data item in the geological dataset DW, WCd represents the d-th data item in the cleaned dataset WC, minWCd represents the valley value of the d-th data item in the cleaned dataset WC, and maxWCd represents the peak value of the d-th data item in the cleaned dataset WC.
4. The method for geological risk assessment of coal filling mining based on data mining according to claim 3, characterized in that: S3 includes S31 and S32; S31, extracting time series features from the geological data set DW, including formation subsidence rate Vsp, formation gradient change rate ΔDr, formation strain YPm, and groundwater permeability SpH, and forming a time series feature set Fs; The formation subsidence rate Vsp is extracted using the sliding window technique and is obtained using the following formula: Where SG(t) represents the surface subsidence at time t, SG(t-1) represents the surface subsidence at time t-1, and Δt represents the time interval; The formation gradient change rate ΔDr is obtained by the following formula: Where Δz represents the vertical depth interval between measurement points; The rock formation strain YPm is obtained by the following formula: Where, Pm(t) represents the mine pressure at time t, and Pm(t-1) represents the mine pressure at time t-1; Groundwater permeability Sp is obtained by the following formula: In the formula, Co represents the empirical coefficient and maxpf represents the maximum value of the filling density.
5. The method for geological risk assessment of coal filling mining based on data mining according to claim 4, characterized in that: S32, based on the acquired time series feature set Fs, a stratum movement prediction model is established by using a multivariate regression model to predict future surface subsidence and obtain a predicted stratum subsidence amount SP; The predicted amount of ground settlement SP is obtained by the following formula: In the formula, M represents the total number of features, Fsj represents the jth feature in the time series feature set Fs, and α j represents the model parameter of the jth feature, and EX represents the error term.
6. A method for assessing geological risks of coal filling mining based on data mining according to claim 5, wherein Characterized in that: S4 includes S41 and S42; S41, analyzing the contribution of the features in the time series feature set Fs to the formation settlement prediction value SP, quantifying the influence of the features on the formation settlement prediction value SP, obtaining the contribution rate Cj, and judging the influence of the features on the formation movement by the contribution rate Cj; The contribution rate Cj is obtained by the following formula: In the formula, Cj represents the contribution rate, specifically the contribution rate of the j-th feature in the time series feature set Fs to the predicted value of formation settlement SP, represents partial derivative; The influence of the features on the ground movement is obtained by matching the following methods: When 0.5≤contribution ratio Cj<1, this indicates that the feature has an impact on the formation movement; When 0.0<contribution rate Cj<0.5, it means that the feature has no effect on stratum movement; Quantify the contribution rate Cj and obtain the characteristic impact factor Wj; The characteristic impact factor Wj is obtained by the following formula: In the formula, Wj represents the feature impact factor, specifically the influence ratio of the jth feature in the time series feature set Fs among all features. represents the sum of contributions of all features.
7. The method for geological risk assessment of coal filling mining based on data mining according to claim 6, characterized in that: S42. Calculate the sensitivity index Im based on the obtained contribution rate Cj and characteristic influencing factor Wj, and compare it with the preset sensitivity threshold TIm to determine the formation risk status; The sensitivity index Im is obtained by the following formula: The formation risk status is obtained by matching in the following ways: When the sensitivity index Im≤sensitivity threshold Tim, it means that there is no risk in the formation, and the current process is maintained and monitoring continues; When the sensitivity index Im>sensitivity threshold TIM, it indicates that there is a risk in the formation and the filling parameters should be adjusted.
8. The method for geological risk assessment of coal filling mining based on data mining according to claim 7, characterized in that: S5. Adjust the filling density pf and the mining rate Ve according to the obtained sensitivity index Im to obtain an adjusted filling density npf and an adjusted mining rate nVe; When the sensitivity index Im>sensitivity threshold TIm, it means that there is a risk in the formation and the filling density pf is adjusted; According to the sensitivity index Im and the sensitivity threshold TIm, the density adjustment amount Δpf is calculated; The density adjustment Δpf is obtained by the following formula: Δpf=k1*(Im-TIm); Where, k1 represents the density adjustment coefficient; Adjust the filling density pf using the obtained density adjustment amount Δpf to obtain an adjusted filling density npf; The adjusted filling density npf is obtained by the following formula: npf=pf+Δpf; When the sensitivity index Im>sensitivity threshold TIm, it means that there is a risk in the formation and the mining rate Ve is adjusted; Calculate the acquisition rate adjustment ΔVe based on the sensitivity index Im and the sensitivity threshold TIm; The rate adjustment ΔVe is obtained by the following formula: ΔVe=k2*(Im-TIm); Where k2 represents the rate adjustment coefficient; Adjust the mining rate Ve using the obtained rate adjustment amount ΔVe to obtain an adjusted mining rate nVe; nVe=Ve-ΔVe.
9. The method for geological risk assessment of coal filling mining based on data mining according to claim 8, characterized in that: S6 includes S61 and S62; S61. Adjust the model parameter α of the formation movement prediction model according to the obtained adjusted filling density npf and the adjusted production rate nVe, and calculate and obtain the parameter adjustment amount Δα; The parameter adjustment amount Δα is obtained by the following formula: Where tanh represents the hyperbolic tangent function, λ1 represents the density adjustment weight factor, and λ2 represents the rate adjustment weight factor; The new model parameter nα is calculated by the parameter adjustment amount Δα and the model parameter α; The new model parameter nα is obtained by the following formula: nα=Δα+α.
10. The method for geological risk assessment of coal filling mining based on data mining according to claim 9, characterized in that: S62, bringing the obtained new model parameter nα into the stratum movement prediction model, re-predicting the stratum settlement prediction value SP, and obtaining a new stratum settlement prediction value nSP; The new stratum settlement prediction nSP is obtained by the following formula:
Citation Information
Patent Citations
Land subsidence partitioning method based on multivariate influence factors
CN110362867A
Coal mine filling mining geological risk assessment method based on data mining
CN118822261A
Geological disaster deformation monitoring method based on remote sensing data
CN119414382A
Field geological exploration monitoring system based on remote sensing data
CN119649587A
Coal mine pressure risk assessment method based on complex system dynamics
CN119671282A
Cited By
Mining ground subsidence dynamic prediction method based on geological trigger rebalance mechanism
CN121189197A
Mining ground subsidence dynamic prediction method based on geological trigger rebalancing mechanism
CN121189197B