Comprehensive evaluation method and system for rock burst occurrence possibility of deep mine engineering

Abstract

The present application relates to the technical field of mine engineering, in particular to a comprehensive evaluation method and system for the possibility of rock burst in deep mine engineering, which comprises: collecting rock mechanics test data, engineering environment data and time series disturbance record data, and organizing them into state characteristic vectors; calculating strength ratio indexes and the like based on the rock mechanics test data, and dynamically weighting to obtain a rock intrinsic tendency index; calculating a static excitation basic component based on the engineering environment data, extracting a cumulative disturbance intensity index based on the time series disturbance record data, and adaptively calculating a fusion weight to obtain an environmental excitation index; calculating a suppression energy consumption correction coefficient, constructing the rock intrinsic tendency index, the environmental excitation index and the suppression energy consumption correction coefficient into a comprehensive characteristic vector, inputting the comprehensive characteristic vector into an integrated learning model, and outputting a rock burst possibility grade label and a characteristic attribution output. The present application effectively reduces the systematic underestimation risk of the possibility of rock burst induced by periodic blasting disturbance.

Description

Technical Field

[0001] This invention relates to the field of mining engineering technology, and in particular to a comprehensive evaluation method and system for the probability of rock bursts in deep mining engineering. Background Technology

[0002] As the mining depth of deep metal mines continues to increase, the level of original geostress borne by the surrounding rock increases significantly, leading to increasingly severe dynamic responses in the rock mass induced by excavation unloading. Rockburst, a typical dynamic hazard in deep underground engineering, can cause spalling of the surrounding rock in tunnels and even large-scale rock mass ejection and structural damage, posing a serious threat to the safety of workers and the integrity of equipment. Meanwhile, deep metal mines commonly employ staged controlled blasting to advance the mining area, and the superimposed effect of blasting vibration and high static geostress further exacerbates the complexity of the dynamic response of the surrounding rock.

[0003] Existing rockburst evaluation methods have evolved along two main technical routes: one is based on uniaxial or triaxial compression tests to obtain mechanical parameters such as rock brittleness index, elastic energy index, and impact energy index, and then classifies the rock's rockburst tendency using fixed criterion thresholds; the other is to introduce engineering environmental parameters such as in-situ stress ratio and burial depth coefficient on top of single-index criteria, and use artificial intelligence models such as support vector machines, random forests, or BP neural networks to map multidimensional input indicators to rockburst occurrence probability level labels. Building on this, some methods further calculate three types of indicators—rockburst intrinsic tendency index, engineering environmental excitation index, and suppression energy consumption correction coefficient—independently before inputting them into a fusion model, aiming to achieve independent separation of each evaluation dimension.

[0004] Chinese patent document CN117195594B discloses a method, device, electronic device, and storage medium for assessing the rockburst level of a tunnel. The method includes: acquiring strain data collected at each sampling point on the surface of the tunnel surrounding rock at a target sampling time; determining the deformation inducing factor corresponding to each sampling point based on the strain data collected at each sampling point; constructing a strain domain matrix corresponding to the tunnel surrounding rock surface based on the deformation inducing factor corresponding to each sampling point; determining the maximum shear stress on the tunnel surrounding rock surface based on the strain domain matrix; and determining the rockburst level corresponding to any monitoring area on the tunnel surrounding rock surface at the target sampling time based on the maximum shear stress and the corresponding static rock parameters of the monitoring area. This method, to a certain extent, combines static mechanical parameters with dynamic stress conditions. However, the dynamic conditions mentioned in this scheme are essentially instantaneous state quantities at the current acquisition moment, that is, the maximum shear stress value of the surrounding rock of the tunnel wall at a certain moment. They do not track the time series characteristics of periodic blasting disturbances: if two time nodes have similar static mechanical parameters and current shear stress values, they will obtain almost the same rockburst level assessment results, and cannot distinguish the fundamental difference that one node has experienced dozens of blasting vibrations and the internal crack network of the rock mass has undergone substantial expansion.

[0005] However, mining operations in deep metal mines are characterized by significant periodic blasting disturbances. During stope advancement, repeated blasting vibrations from adjacent mining areas continuously act on the surrounding rock of the assessment area in the form of periodic stress waves. Each blast induces microcrack propagation within the rock mass and consumes some of the elastic stored energy, but the increase in cracks caused by a single blast is often difficult to detect directly under conventional monitoring accuracy. Existing methods treat the excitation parameters collected at any given time as independent static inputs, without tracking the time-series characteristics of periodic blasting disturbances: if two time points have similar rock intrinsic tendency indices and energy consumption suppression correction coefficients, they will obtain almost identical rockburst probability ratings, failing to distinguish the fundamental difference that one point has undergone dozens of cumulative blasting vibrations and that the internal crack network of the rock mass has substantially expanded. Fatigue damage of the rock mass under periodic blasting disturbances exhibits path dependence; the cumulative weakening effect of previous disturbances on the rock mass structure systematically lowers the energy excitation threshold required for subsequent rockbursts. The static input model is not able to perceive this cumulative weakening trajectory, which leads to a systematic underestimation of the probability of rockburst induced by periodic blasting disturbance, especially in deep, high-stress sections where blasting is frequent in adjacent mining areas. Summary of the Invention

[0006] To address the problem that existing fusion evaluation models treat the excitation effect of the engineering environment as a static input and fail to perceive the cumulative weakening trajectory of surrounding rock fatigue damage under periodic blasting disturbances, thus leading to a systematic underestimation of the probability of rockburst occurrence, this invention provides a comprehensive evaluation method and system for the probability of rockburst occurrence in deep mining engineering.

[0007] In a first aspect, the present invention provides a comprehensive evaluation method for the probability of rock bursts in deep mining engineering, comprising: S1. Collect rock mechanics test data, engineering environment data, and time-series disturbance record data of the engineering area to be evaluated, and organize these data into a state feature vector. S2. Calculate the strength ratio index, elastic energy index, and brittle mutation index based on the rock mechanics test data in the state feature vector. Then, dynamically weight and fuse these indices based on the Pearson correlation coefficient to obtain the rock intrinsic tendency index. S3. Calculate the static excitation basis component based on the engineering environment data in the state feature vector. Extract the cumulative disturbance intensity index based on the time-series disturbance record data in the state feature vector. Adaptively calculate the fusion weights based on the static excitation basis component and the cumulative disturbance intensity index to obtain the environmental excitation index. S4. Calculate the suppression energy consumption correction coefficient based on the state feature vector. Construct a comprehensive feature vector from the rock intrinsic tendency index, the environmental excitation index, and the suppression energy consumption correction coefficient. Input the comprehensive feature vector into a pre-trained ensemble learning model to output a rock eruption probability level label and feature attribution output.

[0008] This invention quantifies the fatigue damage path of rock mass under periodic blasting disturbance by performing time-decay weighted accumulation of peak particle velocities from successive blastings within a complete mining cycle, expressing it in the form of an environmental excitation index. This index is then used as an independent feature dimension in the final comprehensive evaluation. This fundamentally solves the problem of static input models being unaware of the cumulative weakening trajectory and effectively reduces the risk of systematically underestimating the probability of rock bursts induced by periodic blasting disturbances.

[0009] Preferably, rock mechanics test data, engineering environment data, and time-series disturbance record data of the engineering area to be evaluated are collected, and these data are organized into a state feature vector. This includes: synchronously collecting rock mechanics test data and engineering environment data of the engineering area to be evaluated, and completing spatial registration under a unified engineering coordinate system; extracting the successive blasting vibration monitoring records within a complete mining cycle preceding the evaluation time as time-series disturbance record data, using linear interpolation to fill in missing records, and truncating the data with an upper limit of three times the standard deviation of the statistical distribution; and organizing the rock mechanics test data, engineering environment data, and time-series disturbance record data according to a dual index of mining mileage station number and time to finally obtain the state feature vector.

[0010] Preferably, the strength ratio index, elastic energy index, and brittle abrupt change index are calculated based on the rock mechanics test data in the state feature vector, including: extracting the uniaxial compressive strength and splitting tensile strength from the rock mechanics test data and calculating their ratio as the strength ratio index; extracting the release elastic strain energy density and dissipated energy density from the rock mechanics test data and calculating their ratio as the elastic energy index; extracting the absolute value of the slope from the peak point to the residual strength segment and the slope of the elastic segment before the peak in the post-peak stress-strain curve from the rock mechanics test data and calculating their ratio as the brittle abrupt change index, thus finally obtaining the strength ratio index, elastic energy index, and brittle abrupt change index.

[0011] By calculating the strength ratio index, elastic energy index, and brittle abrupt change index respectively, the intrinsic mechanical properties of rocks were quantified from three complementary dimensions: rock strength characteristics, energy accumulation and release capacity, and degree of failure abrupt change. This provides a complete and independent sub-index basis for subsequent dynamic weighted fusion, ensuring the comprehensiveness and objectivity of rock material tendency characterization.

[0012] Preferably, the rock intrinsic tendency index is obtained by dynamically weighting and fusing the strength ratio index, elastic energy index, and brittle mutation index based on the Pearson correlation coefficient. This includes: normalizing the range of the strength ratio index, elastic energy index, and brittle mutation index using their respective grading thresholds; dynamically allocating the fusion weights of the normalized indicators based on the proportion of the absolute values ​​of the Pearson correlation coefficients between each sub-indicator in the rockburst case database and the known rockburst occurrence probability level labels in the sum of their absolute values; and finally, weighting and summing the normalized indicators according to the fusion weights to obtain the rock intrinsic tendency index.

[0013] By dynamically weighting and fusing complementary intrinsic mechanical sub-indices into a single intrinsic tendency index of rock in a data-driven manner, the subjective errors introduced by fixed weights based on human experience are avoided, and the properties of rock materials are strictly separated from the excitation factors of the engineering environment. This provides an intrinsic tendency representation with simple information content and independent updates for the subsequent construction of multi-dimensional feature vectors.

[0014] Preferably, the static excitation basic component is calculated based on the engineering environment data in the state feature vector, and the cumulative disturbance intensity index is extracted based on the time-series disturbance record data in the state feature vector. This includes: extracting the geostress excitation sub-index, unloading rate sub-index, free surface sub-index, and rock mass integrity sub-index from the engineering environment data, performing range normalization, and then summing them with equal weights to obtain the static excitation basic component; extracting the peak particle velocity generated at the section to be evaluated by each blast in the time-series disturbance record data, and performing weighted accumulation based on the time decay weight determined according to the elapsed duration to obtain the cumulative disturbance intensity index, and finally obtaining the static excitation basic component and the cumulative disturbance intensity index.

[0015] By constructing a static excitation base component and a cumulative disturbance intensity index, the static component integrates four sub-indices: ground stress, unloading rate, free face, and rock mass integrity. The dynamic component uses time decay weighted summation of peak vibration velocities from each blast, thus achieving a quantitative expression of the cumulative effect of periodic blasting disturbance. This enables the model to perceive the path dependence of rock mass fatigue damage and provides mutually independent and physically meaningful excitation components for subsequent adaptive fusion. As a result, the accuracy of the evaluation of blasting disturbance-induced rockburst is effectively improved, and the risk of systematic underestimation is reduced.

[0016] Preferably, the environmental excitation index is obtained by adaptively calculating the fusion weight based on the static excitation base component and the cumulative disturbance intensity index, including: calculating the ratio of the static excitation base component to the sum of the static excitation base component and the cumulative disturbance intensity index as the adaptive fusion weight; weighting the static excitation base component using the fusion weight, and weighting the cumulative disturbance intensity index using the difference between 1 and the fusion weight; and summing the weighted static excitation base component and the weighted cumulative disturbance intensity index to finally obtain the environmental excitation index.

[0017] This method enables the environmental excitation index to integrate the static excitation component and the cumulative disturbance component through adaptive weighting. The weights are automatically adjusted according to the real-time relative magnitude of the two components, eliminating the need for manual setting of fixed weights. This allows the evaluation model to maintain a reasonable response to the dominant excitation factor under different disturbance intensities, overcoming the problem of distortion in the response to the dominant factor under extreme conditions when using fixed weights.

[0018] Preferably, the energy consumption suppression correction coefficient is calculated based on the state feature vector, including: extracting the structural surface energy consumption sub-indices, support constraint sub-indices, and pressure relief sub-indices from the state feature vector and normalizing their ranges; dynamically allocating weights based on the ratio of the absolute value of the Pearson correlation coefficient between each sub-indice and the sample where rockburst did not occur; and performing a weighted summation of the normalized sub-indices according to the allocated weights to finally obtain the energy consumption suppression correction coefficient.

[0019] Preferably, the rock intrinsic tendency index, the environmental excitation index, and the suppression energy consumption correction coefficient are constructed into a comprehensive feature vector, including: splicing and combining the rock intrinsic tendency index, the environmental excitation index, and the suppression energy consumption correction coefficient to finally obtain the comprehensive feature vector.

[0020] Preferably, the comprehensive feature vector is input into a pre-trained ensemble learning model to output a rock eruption probability level label and feature attribution output, including: inputting the comprehensive feature vector into an ensemble learning model trained with hierarchical cross-validation; outputting a rock eruption probability level label through the first layer of the ensemble learning model; and outputting the positive and negative contributions of the rock intrinsic tendency index, environmental excitation index, and suppression energy consumption correction coefficient to the prediction results through the second layer of the ensemble learning model based on SHAP values, finally obtaining the feature attribution output.

[0021] By outputting SHAP values, the propensity contribution, environmental stimulus contribution, and inhibition correction contribution corresponding to the evaluation results can all be presented independently. Engineers can directly judge the source of risk and deploy targeted countermeasures based on the contribution of each dimension. Compared with the traditional integrated model with black box output, it has obvious practical advantages in engineering.

[0022] Secondly, the present invention provides a comprehensive evaluation system for the probability of rock bursts in deep mining engineering, employing the following technical solution: A comprehensive evaluation system for the probability of rock bursts in deep mining engineering includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the comprehensive evaluation method for the probability of rock bursts in deep mining engineering as described above.

[0023] The present invention has the following technical effects: This invention integrates static stress excitation and time-series cumulative blasting disturbance into a single environmental excitation index by performing time-decrease weighted accumulation of peak particle velocities from successive blasting blasts throughout a complete mining cycle. This index simultaneously reflects the current static stress level and the cumulative weakening effect of periodic blasting disturbances, fundamentally compensating for the deficiency of static input models in not being aware of cumulative weakening trajectories. It effectively reduces the risk of systematically underestimating the probability of rock bursts induced by periodic blasting disturbances and improves the evaluation accuracy of deep, high-stress sections with frequent blasting in adjacent mining areas.

[0024] Meanwhile, the three independent dimensions of rock intrinsic tendency index, environmental excitation index and energy suppression correction coefficient are constructed into a three-dimensional comprehensive feature vector, which is input into the integrated learning model and coupled with feature attribution output. This enables the comprehensive evaluation results to have both the prediction accuracy of multi-dimensional information fusion and the engineering interpretability of the contribution of each dimension to be traceable independently, providing a reliable basis for on-site engineers to deploy disaster prevention and mitigation measures in a targeted manner. Attached Figure Description

[0025] Figure 1 This is a flowchart of the comprehensive evaluation method for the probability of rock bursts in deep mining engineering according to the present invention.

[0026] Figure 2 This is a comparison chart of the fusion weights of various sub-indicators in an embodiment of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] This invention discloses a comprehensive evaluation method for the probability of rock bursts in deep mining engineering, referring to... Figure 1 This includes the following steps: S1. Multi-source data acquisition and state feature vector construction.

[0029] In an optional embodiment, three types of data are simultaneously collected for the engineering area to be evaluated and spatially registered in a unified engineering coordinate system. The first type is rock mechanics test data, including uniaxial compressive strength. Splitting tensile strength Releaseable elastic strain energy density (i.e., the releasable elastic strain energy accumulated before peak intensity) and dissipated energy density (i.e., the energy density consumed during post-peak failure, including crack propagation, friction, and plastic deformation), and the ratio of the absolute value of the slope of the post-peak segment of the full stress-strain curve to the elastic modulus before the peak, serve as the basis for calculating brittle characteristics. This type of data comes from indoor compression tests of standard rock core samples, reflects the intrinsic mechanical properties of rock materials, is independent of the engineering environment, and does not change with the progress of mining.

[0030] The second category is engineering environmental data, including the maximum principal stress in the area to be evaluated. The rock mass integrity coefficient is obtained from in-situ stress relief tests or hydraulic fracturing measurements; the distance from the current excavation face to the section to be evaluated and the exposed area of ​​the free face are obtained from on-site measurements; and the longitudinal wave velocity of the rock mass at the site The longitudinal wave velocity of the intact rock core was determined by a combination of acoustic wave testing and fracture statistics. The data was obtained through indoor acoustic testing; the current support type and density, and the ultimate bearing capacity of a single anchor bolt; and the coverage area and borehole diameter parameters of the pressure relief boreholes that have been implemented.

[0031] Furthermore, the third category is time-series disturbance record data, which is a key input that distinguishes this method from existing methods. For the region to be evaluated, the evaluation time is extracted. Tracing back to a complete mining cycle The internal sequential blasting vibration monitoring records include the peak particle velocity generated at the section to be evaluated for each blast. ,in , This represents the number of blasts within a cycle. The value is determined based on the actual mining rhythm of the mine, and usually corresponds to a complete blasting-mining-support cycle. For example, The value can be between 3 and 7 days. Since blasting vibration monitoring systems are already standard equipment in deep mines, It can be read directly by velocity sensors deployed near the section to be evaluated, without the need to establish an additional independent monitoring system.

[0032] To address the issue of missing data, linear interpolation was used to supplement the vibration records missing at certain moments, ensuring the integrity of the temporal disturbance sequence. To address the issue of outliers, historical data from similar operating conditions in the same blasting area were used. The statistical distribution is truncated to three standard deviations to prevent a single abnormal reading from excessively inflating the calculation results of the cumulative disturbance intensity.

[0033] The three types of data are organized into a state feature vector using a dual index of mining mileage station number and time, which is then used by subsequent calculation modules.

[0034] In this way, by simultaneously incorporating three types of information—the intrinsic mechanical properties of the rock, the current engineering environment, and the temporal disturbance records within the complete mining cycle—into a unified feature framework, a data foundation is provided for subsequent calculation modules, ensuring that the information content is independent and that the spatial and temporal positioning is clearly defined. This avoids feature contamination caused by mixed data sources or time alignment deviations.

[0035] S2. Dynamically weighted fusion calculation of rock intrinsic tendency index.

[0036] In an optional embodiment, three sub-indices are calculated based on the rock mechanics test data in the acquired state feature vector. The strength ratio index is the ratio of uniaxial compressive strength to splitting tensile strength:

[0037] in, Indicates the strength ratio index, Indicates uniaxial compressive strength. This indicates the splitting tensile strength.

[0038] This ratio reflects the difference in the ability of rocks to resist tensile failure during the transition from compressive to tensile stress. A higher value indicates that the rock is more prone to abrupt brittle fracture under compression conditions. Elastic energy index Defined as the release elastic strain energy density With dissipation energy density The ratio:

[0039] in, Indicates elastic energy index, This represents the density of releasable elastic strain energy. This represents the dissipated energy density.

[0040] This value reflects the rock's ability to accumulate elastic energy relative to its dissipated energy before failure. A larger value indicates a stronger energy storage capacity and a lower proportion of energy consumed through plastic deformation after the peak, thus increasing the rock's tendency to trigger a rockburst through a sudden release of elastic energy upon failure. Conversely, a smaller value indicates a larger proportion of energy dissipated during loading, with post-peak failure tending towards ductility, thus reducing the probability of explosive failure.

[0041] Next, the brittle mutation index Defined as the absolute value of the slope from the peak point to the residual strength segment in the post-peak stress-strain curve. Compared with the slope of the elastic segment before the peak (elastic modulus) The ratio:

[0042] in, Indicators of brittle mutation This represents the absolute value of the slope of the stress-strain curve from the peak point to the residual strength segment. This represents the slope of the elastic segment before the peak.

[0043] in The absolute value of the average slope within the deformation range from peak strength to residual strength is taken, and the unit is the same as that of the elastic modulus. The larger the value, the steeper the stress drop after the peak, and the more significant the brittle failure characteristics of the rock relative to the loading stiffness before the peak. The smaller the value, the more gradually the post-peak stress decreases, indicating that the rock has a certain ductile buffering capacity.

[0044] Furthermore, regarding , , The three sub-indicators are respectively normalized in range using their corresponding grading thresholds in the rockburst case database, and mapped to... The interval is used to obtain the normalized value. , , The fusion weights of the three normalized sub-indicators are not assigned manually based on experience, but are dynamically allocated based on the ratio of the absolute value of the Pearson correlation coefficient between each sub-indicator and the known rock eruption probability level label in the database. The weights of each sub-indicator are:

[0045] in, This represents the dynamic fusion weight of the j-th sub-indicator. denoted as the Pearson correlation coefficient between the j-th sub-index and the known rock eruption probability level label.

[0046] in For the first The Pearson correlation coefficients between each sub-indicator and the numerical code for rockburst severity were derived statistically from the database samples. The engineering implications of this weighting method are: sub-indicators with a stronger linear correlation to rockburst severity in the current database are automatically assigned higher weights, continuously self-correcting as the database sample size increases, without manual intervention. The intrinsic rock tendency index is:

[0047] in, Indicates the intrinsic dip index of rocks. These represent the strength ratio index, elastic energy index, and brittle abrupt change index, respectively, after range normalization. These represent the dynamic fusion weights of the corresponding indicators.

[0048] The range of values ​​is It only characterizes the intrinsic rockburst tendency of rock materials and is not used as the sole basis for judging the rockburst level in the field. It needs to be used in conjunction with the subsequent environmental excitation index and the suppression energy consumption correction coefficient to form a complete evaluation.

[0049] It should be noted that there are many other sub-indicators in this field that can reflect the tendency of rocks to brittle failure. Those skilled in the art can select other mechanical parameters such as the impact energy index to replace or supplement the above three sub-indicators according to the actual lithological characteristics and existing test conditions. As long as the normalization mapping range is kept consistent and the weights are dynamically allocated through the correlation coefficient, the overall applicability of this method will not be affected.

[0050] Thus, by dynamically weighting and fusing three complementary intrinsic mechanical sub-indices into a single intrinsic tendency index of rock in a data-driven manner, the subjective errors introduced by fixed weights based on human experience are avoided, and the properties of rock materials are strictly separated from the excitation factors of the engineering environment. This provides an intrinsic tendency representation with simple information content and independent updates for the subsequent construction of multidimensional feature vectors.

[0051] S3. Adaptive calculation of fusion weights yields the environmental excitation index.

[0052] In an optional embodiment, the environmental excitation index is calculated by constructing two components separately and then adaptively fusing them to integrate the current static stress level with the accumulated blasting disturbance during the mining cycle, thus forming a unified environmental excitation index. .

[0053] Calculate the static excitation fundamental components based on engineering environment data in the state feature vector. It encompasses four sub-indicators: the geostress excitation sub-indicator is the ratio of the current maximum principal stress to the uniaxial compressive strength of the rock. The index reflects the stress concentration of the surrounding rock; the unloading rate sub-index is the ratio of the current excavation face advance speed to the elastic longitudinal wave velocity of the rock mass, reflecting the relative relationship between the unloading rate and the elastic response rate of the rock mass; the free face sub-index is the ratio of the current exposed area of ​​the free face to the cross-sectional area of ​​the roadway; the rock mass integrity sub-index is... The complement of the integrity coefficient characterizes the amplification effect of the degree of joint and fracture development in the rock mass on stress concentration. After range normalization of the four sub-indices, they are summed with equal weights to obtain the statically excited foundation component. .

[0054] Then, the cumulative disturbance intensity index is extracted based on the temporal disturbance record data in the state feature vector. At the evaluation time... The complete mining cycle traced back to the beginning Inside, the first The peak particle velocity generated at the evaluation surface by the secondary blast This is considered a characterization of the dynamic impact intensity of the disturbance on the rock mass. For all [parts / items] within the period... For each disturbance event, a time-decay-weighted cumulative disturbance strength index is defined:

[0055] in, This represents a time-decay-weighted index of cumulative disturbance strength. This represents the peak particle velocity produced by the k-th blast at the section to be evaluated. This represents the time decay weight, and n represents the total number of blasts within the evaluation period.

[0056] Among them, the time decay weight Assign values ​​as follows:

[0057] in, Indicates the complete mining cycle. For the first The elapsed time since the evaluation, in units of... The same applies. Due to creep relaxation between the two disturbances, the crack propagation caused by the early disturbance will partially heal itself in the subsequent time due to local stress redistribution, and its impact on the current crack state will gradually decrease. At the moment the blast occurs ( The value is 1, at the starting point of the trace window ( The value approaches 0, with recent disturbances receiving high weight and long-term disturbances receiving low weight, which aligns with the aforementioned physical laws. Range normalization is performed within the statistical sample of all historical evaluation windows in the current mining section to obtain the normalized cumulative disturbance intensity index. The upper and lower bounds used for normalization are derived from the statistical interval of measured data in the same mining stage of the mine, and do not rely on human experience to determine the values.

[0058] Next, based on the static excitation fundamental component and the normalized cumulative disturbance intensity index, the fusion weight is adaptively calculated, and the static component and the time-series cumulative component are adaptively fused to obtain the environmental excitation index. :

[0059] in, This represents the environmental stimulation index obtained from the final fusion. Represents the static excitation fundamental components, This represents the cumulative disturbance intensity index after range normalization.

[0060] Among them, the fusion weight Instead of taking fixed values, the calculation is adaptively based on the relative magnitudes of the two components:

[0061] This relationship automatically assigns higher fusion weights to components with relatively larger values ​​and lower fusion weights to components with relatively smaller values. When the cumulative perturbation effect is significantly enhanced... Increase It decreases accordingly. Item in The proportion of [something] automatically increases; when static stress excitation significantly dominates [something], [something] automatically increases. Approaching 1, The item weights automatically increase. If the value is lower than the historical minimum value of the evaluation section, the current evaluation will be marked as data abnormal and the calculation will be terminated to ensure that the denominator is valid.

[0062] Thus, by fusing static stress excitation and time-series cumulative blasting disturbance into a single environmental excitation index, which simultaneously reflects the current static stress level and the cumulative weakening effect of periodic blasting disturbance, and with the fusion weights automatically adjusted according to the real-time relative magnitudes of the two components, the environmental excitation index can respond reasonably under three working conditions: high static stress dominance, high cumulative disturbance dominance, and both dominance. This overcomes the problem of distorted response to dominant factors under extreme working conditions when using fixed-weight fusion.

[0063] S4. Construct a comprehensive feature vector and output the feature attribution.

[0064] In an optional embodiment, the energy consumption suppression correction coefficient is calculated based on the state feature vector. It represents the comprehensive weakening effect of on-site rock mass structure, support system and pressure relief measures on the process of rockburst energy accumulation and sudden release, and is obtained by weighted fusion of three sub-indicators.

[0065] The energy dissipation sub-index of the structural surface is calculated as the square of the ratio of the measured acoustic velocity to the acoustic velocity in the intact rock core. The elastic wave energy transmittance characterizes fractured rock masses. Since the elastic wave energy density is proportional to the square of the particle velocity, under the same amplitude conditions in the same medium, the energy ratio of a fractured rock mass to that of a intact rock core is equal to the ratio of the squares of the corresponding particle velocities. Therefore, this directly corresponds to the energy transmittance when elastic waves pass through structural surfaces. The lower this value, the stronger the energy dissipation of the structural surface, and the less elastic wave energy accumulates in the rock mass and can be suddenly released. For the on-site measured P-wave velocity of the rock mass, The P-wave velocity is from intact indoor rock cores. The support constraint sub-index is taken as the resistance per unit area of ​​support. The ratio of the support resistance per unit area to the ultimate bearing capacity of a single anchor bolt reflects the restraining ability of the support system to release the deformation of the surrounding rock. The pressure relief sub-index is the coverage area ratio of the pressure relief borehole within a range of 5 times the borehole diameter in front of the assessment face, reflecting the degree of reduction of energy accumulation by active pressure relief measures.

[0066] After normalizing the ranges of the three sub-indicators, the weights are dynamically allocated based on the ratio of the absolute value of the Pearson correlation coefficient between each sub-indicator and the samples where rockburst did not occur. The calculation formula is as follows: ,in For the first The Pearson correlation coefficients between individual indicators and the "no rockburst" tag were derived from database samples. A weighted summation was then performed. The larger the value, the stronger the energy-suppressing effect at the site, and the higher the energy threshold required to trigger a rockburst.

[0067] Furthermore, after obtaining , , After considering the three indicators, the intrinsic rock tendency index, environmental excitation index, and energy consumption suppression correction coefficient are used to construct a comprehensive feature vector:

[0068] in The static excitation fundamental components have been fused using adaptive weights. With normalized cumulative disturbance intensity index This characterizes the comprehensive initiation effect of the current engineering environment on rockburst, and is related to the intrinsic rock tendency index. Energy consumption suppression correction coefficient The three pieces of information do not overlap and together constitute the three-dimensional feature space for evaluation. The comprehensive feature vector is input into a pre-trained ensemble learning model, and the fusion evaluation model employs ensemble learning models such as gradient boosting trees or XGBoost, based on historical samples from the rockburst case database. The model is trained using vectors as input and rockburst probability levels (none, slight, moderate, and severe) as labels. In addition to regular samples of rockbursts that have occurred and those that haven't, the database specifically includes samples with inconsistent propensity assessments and field results, as well as samples with complete temporal perturbation records. This ensures the model can learn the moderating effect of cumulative perturbation on level discrimination. Hierarchical cross-validation is used for model training, with the weighted harmonic mean of recall rates for each level in the confusion matrix as the optimization objective. The focus is on improving the ability to identify moderate and severe levels, preventing the model from degenerating into a conservative predictor that tends to output low-risk levels due to sample class imbalance.

[0069] The ensemble learning model directly outputs a rock eruption probability level label, and also provides a value based on the SHAP value. , , The positive and negative contributions of the three features to the current prediction result are used as feature attribution outputs. Taking a certain evaluation as an example, if... If the SHAP value is the largest and positive among the three characteristics, it indicates that the dominant reason for the current rating improvement is the stronger excitation effect of the engineering environment. This can be further traced back to a higher static stress level or stronger cumulative blasting disturbance within the current mining cycle. Engineers can prioritize targeted measures such as deploying more dense pressure-relief boreholes or suspending blasting in adjacent mining areas, rather than simply strengthening the full-section support. The evaluation results are archived in the database according to the spatial coordinates and timestamps of the evaluated section, supporting the visualization output of the rating evolution trend along the mining advance direction, as well as the relationship between different mining nodes. and its components and The comparative display makes it easier for engineers to intuitively judge the evolution of environmental stimuli and their dominant factors.

[0070] In this way, by inputting the three-dimensional feature vector into the integrated learning model and matching it with SHAP feature attribution output, the comprehensive evaluation results can simultaneously possess the predictive accuracy of multi-dimensional information fusion and the engineering interpretability of independently traceable contribution of each dimension. Engineers can directly judge the source of risk based on the attribution results and deploy targeted countermeasures, which has obvious engineering practical advantages compared with the traditional integrated model with black box output.

[0071] The technical effects of this invention can also be illustrated in conjunction with the accompanying drawings. Figure 2 This is a comparison chart of the fusion weights of various sub-indicators in this invention embodiment. The horizontal axis represents the indicator dimension, and the vertical axis represents the fusion weight. The chart shows six evaluation sub-indicator dimensions, divided into two groups: the first three (strength ratio, elastic energy, and brittle mutation) belong to the intrinsic tendency sub-indicator category, and the latter three (structural surface energy dissipation, support constraint, and pressure relief) belong to the energy dissipation suppression correction sub-indicator category. As can be observed from the chart, the weights of each dimension in existing methods tend to be evenly distributed, while the method of this invention exhibits differentiated weight allocations across dimensions. This indicates that the method of this invention can adaptively fuse weights based on the actual differences in the contribution of different indicators to rockburst assessment, thereby improving the scientific rigor and accuracy of the assessment results.

[0072] This invention also discloses a comprehensive evaluation system for the probability of rock bursts in deep mining engineering, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the comprehensive evaluation method for the probability of rock bursts in deep mining engineering according to this invention is implemented.

[0073] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A comprehensive evaluation method for the probability of rock bursts in deep mining engineering, characterized in that, include: S1. Collect rock mechanics test data, engineering environment data, and time-series disturbance record data of the engineering area to be evaluated, and organize these data into a state feature vector. S2. Calculate the strength ratio index, elastic energy index, and brittleness mutation index based on the rock mechanics test data in the state feature vector. Then, dynamically weight and fuse these indices using the Pearson correlation coefficient to obtain the rock intrinsic tendency index. Extract the uniaxial compressive strength and splitting tensile strength from the rock mechanics test data, and calculate their ratio as the strength ratio index. Extract the peak value from the post-peak stress-strain curve of the rock mechanics test data. S3. Calculate the ratio of the absolute value of the slope of the residual strength segment to the slope of the elastic segment before the peak as a brittle abrupt change index; S4. Calculate the static excitation basic component based on the engineering environment data in the state feature vector, and extract the cumulative disturbance intensity index based on the time-series disturbance record data in the state feature vector. Adaptively calculate the fusion weight based on the static excitation basic component and the cumulative disturbance intensity index to obtain the environmental excitation index; S5. Calculate the suppression energy consumption correction coefficient based on the state feature vector, construct a comprehensive feature vector from the rock intrinsic tendency index, the environmental excitation index, and the suppression energy consumption correction coefficient, and input the comprehensive feature vector into a pre-trained ensemble learning model to output the rock eruption probability level label and feature attribution output; The static excitation foundation component is calculated based on the engineering environment data in the state feature vector, and the cumulative disturbance intensity index is extracted based on the time-series disturbance record data in the state feature vector. This includes: extracting the geostress excitation sub-index, unloading rate sub-index, free surface sub-index, and rock mass integrity sub-index from the engineering environment data, performing range normalization, and then summing them with equal weights to obtain the static excitation foundation component; extracting the peak particle velocity generated at the section to be evaluated by each blast in the time-series disturbance record data, and performing weighted accumulation based on the time decay weight determined according to the elapsed time to obtain the cumulative disturbance intensity index. Finally, the static excitation foundation component and the cumulative disturbance intensity index are obtained. The energy consumption suppression correction coefficient is calculated based on the state feature vector, including: extracting the structural surface energy consumption sub-indices, support constraint sub-indices, and pressure relief sub-indices from the state feature vector and normalizing their ranges; dynamically assigning weights based on the ratio of the absolute value of the Pearson correlation coefficient between each sub-indice and the sample where rockburst did not occur; and performing a weighted summation of the normalized sub-indices according to the assigned weights to finally obtain the energy consumption suppression correction coefficient.

2. The comprehensive evaluation method for the probability of rock bursts in deep mining engineering according to claim 1, characterized in that, Rock mechanics test data, engineering environment data, and time-series disturbance record data of the project area to be evaluated were collected and organized into a state feature vector. This process included: synchronously collecting rock mechanics test data and engineering environment data of the project area to be evaluated and completing spatial registration under a unified engineering coordinate system; extracting successive blasting vibration monitoring records from a complete mining cycle preceding the evaluation time as time-series disturbance record data, using linear interpolation to fill in missing records, and truncating the data with an upper limit of three times the standard deviation of the statistical distribution; and organizing the rock mechanics test data, engineering environment data, and time-series disturbance record data according to a dual index of mining mileage station number and time to finally obtain the state feature vector.

3. The comprehensive evaluation method for the probability of rock bursts in deep mining engineering according to claim 1, characterized in that, Based on the rock mechanics test data in the state feature vector, the strength ratio index, elastic energy index, and brittle mutation index are calculated respectively. This includes: extracting the release elastic strain energy density and dissipated energy density from the rock mechanics test data, calculating the ratio of the two as the elastic energy index; and finally obtaining the strength ratio index, elastic energy index, and brittle mutation index.

4. The comprehensive evaluation method for the probability of rock bursts in deep mining engineering according to claim 1, characterized in that, The rock intrinsic tendency index is obtained by dynamically weighting and fusing the strength ratio, elastic energy, and brittle mutation indices based on the Pearson correlation coefficient. This process includes: normalizing the range of the strength ratio, elastic energy, and brittle mutation indices using their respective grading thresholds; dynamically allocating the fusion weights of the normalized indices based on the proportion of the absolute values ​​of the Pearson correlation coefficients between each sub-indice and the known rockburst probability level labels in the total absolute value of the index; and finally, weighting and summing the normalized indices according to the fusion weights to obtain the rock intrinsic tendency index.

5. The comprehensive evaluation method for the probability of rock bursts in deep mining engineering according to claim 1, characterized in that, The environmental excitation index is obtained by adaptively calculating the fusion weight based on the static excitation base component and the cumulative disturbance intensity index. This includes: calculating the ratio of the static excitation base component to the sum of the static excitation base component and the cumulative disturbance intensity index as the adaptive fusion weight; weighting the static excitation base component using the fusion weight, and weighting the cumulative disturbance intensity index using the difference between 1 and the fusion weight; and summing the weighted static excitation base component and the weighted cumulative disturbance intensity index to obtain the final environmental excitation index.

6. The comprehensive evaluation method for the probability of rock bursts in deep mining engineering according to claim 1, characterized in that, The intrinsic rock tendency index, environmental excitation index, and suppression energy consumption correction coefficient are used to construct a comprehensive feature vector. This involves concatenating and combining the intrinsic rock tendency index, environmental excitation index, and suppression energy consumption correction coefficient to obtain the comprehensive feature vector.

7. The comprehensive evaluation method for the probability of rock bursts in deep mining engineering according to claim 1, characterized in that, The comprehensive feature vector is input into a pre-trained ensemble learning model, which outputs a rock eruption probability level label and feature attribution output. This includes: inputting the comprehensive feature vector into an ensemble learning model trained with hierarchical cross-validation; outputting a rock eruption probability level label through the first layer of the ensemble learning model; and outputting the positive and negative contributions of the rock intrinsic tendency index, environmental excitation index, and suppression energy consumption correction coefficient to the prediction results through the second layer of the ensemble learning model based on SHAP values, ultimately obtaining the feature attribution output.

8. A comprehensive evaluation system for the probability of rock bursts in deep mining engineering, characterized in that, include: The processor and memory, the memory storing computer program instructions, when executed by the processor, implement the comprehensive evaluation method for the probability of rock bursts in deep mining engineering according to any one of claims 1-7.

Images