Acoustic emission b value and fractal dimension cross early warning coal rock damage short-term and imminent forecasting method

Abstract

The invention discloses a coal rock damage short-term and imminent forecasting method for cross early warning of an acoustic emission b value and fractal dimension, which comprises the following steps of: firstly, collecting acoustic emission monitoring data of coal rocks with different lithology characters, and constructing an acoustic emission monitoring database; secondly, respectively establishing a time sequence prediction model, a space prediction model and an energy prediction model based on coal rock acoustic emission monitoring data of an acoustic emission monitoring database, and combining the three models to obtain a coal rock damage short-term and imminent prediction system; and finally, collecting coal rock acoustic emission monitoring data to a coal rock damage short-term and imminent forecasting system, and outputting to obtain damage time, damage position and damage energy. Wherein the input of the coal rock damage short-term and imminent forecasting system is coal rock acoustic emission monitoring data with different lithology characters, and the output is damage time, damage position and damage energy. According to the invention, the difficult problem of accurate short-term and imminent forecasting of the instability failure time, the fracture dominant position and the instability release energy in the loading process of the coal rock sample is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of coal rock dynamic disaster prediction, and particularly relates to a coal rock damage short-term and impending disaster prediction method based on acoustic emission b value and fractal dimension cross early warning. BACKGROUND

[0002] Rock burst is a typical dynamic disaster in coal mining, and its essence is the macroscopic manifestation of coal rock mass rupture and instability under stress. A large number of studies have shown that acoustic emission (AE) or microseismic (MS) technology can be used to realize continuous monitoring of coal rock rupture signals, thereby effectively characterizing the damage evolution process inside the material, and providing theoretical basis and technical support for the monitoring and prevention of rock burst. In acoustic emission monitoring, b value and fractal dimension are two important early warning indicators: b value can reflect the size distribution characteristics of internal micro-fracture events of coal rock, and fractal dimension is used to describe the spatial aggregation degree of rupture events. These indicators have been widely used in the field of rock mechanics to analyze the acoustic emission characteristics of coal rock, and significant progress has been made in laboratory tests, for example, the decrease of b value and fractal dimension is usually observed before rock failure and instability.

[0003] However, although b value and fractal dimension have been widely used in coal rock dynamic disaster prediction, the existing technology still has limited discussion on the scientific meaning and spatiotemporal evolution common law of the two. Especially, there is a lack of effective correlation from the specimen scale to the engineering scale, which makes it difficult to achieve accurate short-term and impending prediction of coal rock failure. At present, the prediction of instability and failure of coal rock samples at the laboratory scale mainly relies on the analysis of a single parameter or simple combination of acoustic emission signals. Common technical solutions include: real-time monitoring of the sudden increase of acoustic emission event rate, cumulative event number or energy rate, and setting an empirical threshold for early warning; or separately analyzing the time series of b value, and taking its downward trend as a precursor of main rupture. In addition, some studies attempt to qualitatively observe the crack evolution process through the positioning results of acoustic emission events.

[0004] Although these methods can provide early warning information to some extent, they have significant limitations. First, the acoustic emission event rate, energy rate and other parameters are easily disturbed by experimental loading noise, internal micro-crack stage adjustment and material heterogeneity, resulting in high false alarm rate and poor reliability of the early warning signal. Second, when relying solely on the b value, the downward trend may fluctuate several times, and not every time corresponds directly to macroscopic failure, lacking cross-validation with parameters such as fractal dimension that reflect the spatial aggregation degree of rupture, making the early warning index one-sided and unable to accurately capture the critical instability point. Finally, existing methods often separate the time, space and energy information of the failure process, making it difficult to form a comprehensive criterion that can simultaneously predict the exact time, spatial location and instability energy intensity of failure occurrence.

[0005] Specifically, the shortcomings of the prior art can be summarized as follows: First, the early warning accuracy is poor: Because the b value and fractal dimension parameters are easily affected by local micro-crack adjustment, noise interference or sample heterogeneity under complex loading conditions, their fluctuations may not correspond uniquely to the occurrence of main rupture, resulting in a high false alarm rate and an inability to accurately indicate the critical instability state.

[0006] Second, the early warning index is one-sided: Existing methods fail to effectively cross-fuse and co-analyze key parameters that reflect the overall state of internal damage (such as b value) and the concentration degree of damage spatial evolution (such as fractal dimension). The abnormality of a single index may only be a stage phenomenon in the damage evolution process, with weak anti-interference ability and low reliability in identifying true pre-rupture signs.

[0007] Third, the prediction is not comprehensive: Traditional techniques are difficult to provide accurate time, exact spatial location and instability energy information simultaneously and synchronously. For example, the method may warn of impending failure but fail to locate the dominant rupture zone, or identify the damage zone but fail to determine whether it will evolve into a through-going main rupture and the corresponding energy level. This limitation results in poor early warning accuracy and low timeliness, failing to meet the high-precision determination needs of the time, space and energy elements of the entire failure process, restricting the in-depth understanding of coal rock failure mechanism and the improvement of precise prediction ability of disasters.

[0008] In summary, the existing technology has low early warning accuracy, one-sided index and incomplete prediction in monitoring and early warning of coal rock failure, and there is an urgent need for a new method that can integrate multi-parameter and multi-dimensional information to achieve accurate short-term prediction of rock burst. SUMMARY

[0009] The technical problems to be solved by the present application are to provide a coal rock damage short-term prediction method based on acoustic emission b value and fractal dimension cross warning, to solve the problems of accurately predicting the instability damage time, the dominant position of the rupture and the instability release energy of the coal rock sample during the loading process. First, the accurate prediction of the coal rock sample damage time is realized, which breaks through the dependence on the macroscopic trend of the stress-strain curve of the sample as a whole, realizes the minute or second level advanced prediction of when the macroscopic main rupture occurs, and captures the critical instability point. Second, the spatial positioning problem is solved, which accurately locates the core area of the main rupture caused by stress accumulation in the sample, which is not just observing the microcracks throughout, thereby revealing the localization mechanism of the damage. Third, the energy value estimation problem is solved, which predicts the size of the elastic strain energy (corresponding to the acoustic emission energy) that may be released by the sample when it is unstable, so as to evaluate the severity of the damage and provide key basis for quantitative impact tendency and countermeasures research.

[0010] To solve the above technical problems, the technical scheme adopted by the present application is: a coal rock damage short-term prediction method based on acoustic emission b value and fractal dimension cross warning, comprising: Step 1: Collect acoustic emission monitoring data of coal rocks with different lithology, and construct an acoustic emission monitoring database; Step 2: Based on the coal rock acoustic emission monitoring data of the acoustic emission monitoring database, a time series prediction model, a spatial prediction model and an energy prediction model are established respectively, and the three models are combined to obtain a coal rock damage short-term prediction system; Step 3: Collect the coal rock acoustic emission monitoring data to the coal rock damage short-term prediction system, and output the damage time, damage position and damage energy; Wherein, the input of the coal rock damage short-term prediction system is the acoustic emission monitoring data of coal rocks with different lithology, and the output is the damage time, damage position and damage energy.

[0011] Further, the coal rock acoustic emission monitoring data includes: time, ring count, energy and event coordinates.

[0012] Further, the time series prediction model takes time, ring count and energy as input, and damage time as output; The spatial prediction model takes time, ring count and energy as input, and damage position and damage energy as output; The energy prediction model takes time, ring count, energy and event coordinates as input, and damage time and damage energy as output.

[0013] Further, the data processing flow of the time series prediction model comprises: Step A1: continuously calculate the b value and fractal dimension of acoustic emission according to continuous time, ring count and energy, and normalize the continuous b value and fractal dimension of acoustic emission; Step A2: determine whether the normalized continuous b value and fractal dimension of acoustic emission continuously decrease, if yes, proceed to step A3, if not, return to step A1; Step A3: determine whether the stress value of the current coal rock exceeds 60% of the estimated coal rock, if yes, proceed to step A4, if not, return to step A1; Step A4: start the prediction process, calculate the intersection point of the normalized continuous b value and fractal dimension of acoustic emission, and continuously calculate the descending slope value of the b value and fractal dimension of acoustic emission, if the slope deviation is greater than 50%, proceed to step A5, if the slope deviation is less than or equal to 50%, return to step A1; Step A5: calculate the intersection point of the b value and fractal dimension of acoustic emission with the time axis respectively to obtain the damage time interval; Step A6: perform steps A1-A6 on both sides to obtain the damage time interval of the continuous two times, and determine whether the damage time interval is gradually reduced, if yes, output the latest damage time interval, if not, return to step A1; Step A7: determine whether the coal rock is damaged, if yes, output the latest damage time interval, if not, return to step A1.

[0014] Further, in step 4, calculating the intersection point of the normalized continuous b value and fractal dimension of acoustic emission comprises: Preliminary calculation of intersection point, and determine whether the normalized continuous b value and fractal dimension of acoustic emission intersect; If they intersect, output the complementary calculated intersection point, if they do not intersect, return to step A1.

[0015] Further, the data processing flow of the spatial prediction model comprises: Step B1: receive time, ring count, energy and event coordinates; Step B2: divide the spatial cubic grid according to the size of the coal rock, and calculate the multi-element acoustic emission parameters in the cubic grid; Step B3: take the multi-element acoustic emission parameters in the cubic grid and the maximum acoustic emission energy in the cubic grid as the first data set, input the data set into the first machine learning model trained in advance, and output the damage location and damage energy; The first machine learning model takes the first data set as input and outputs the damage location and damage energy.

[0016] Further, the first machine learning model is an XGBoost model, and the XGBoost model optimizes model parameters through a grid search algorithm and a cross-validation algorithm to determine optimal hyperparameters of the model.

[0017] Further, the data processing procedure of the energy prediction model comprises: Step C1: receiving time, ring count, energy, and event coordinates, and calculating multivariate acoustic emission parameters in time sequence; Step C2: taking the multivariate acoustic emission parameters in time sequence and the maximum acoustic emission energy per second as a second data set, inputting the data set into a second machine learning model that has been pre-trained, and outputting a failure time and a failure energy; The second machine learning model takes the second data set as input and outputs the failure time and the failure energy.

[0018] Further, the second machine learning model is a CNN+LSTM time sequence model, and the CNN+LSTM time sequence model optimizes model parameters through a grid search algorithm and a cross-validation algorithm to determine optimal hyperparameters of the model.

[0019] Compared with the prior art, the present application has the following advantages: (1) The prediction reliability is improved: The present application effectively filters false alarms of a single parameter caused by experimental noise or local micro-crack adjustment through deep fusion of the b value (reflecting the overall damage degree of the medium) and the fractal dimension (reflecting the spatial concentration of damage) from "single criterion" to "cross-validation", greatly improving the credibility of the early warning signal.

[0020] (2) The prediction accuracy is improved: The present application provides qualitative or trend judgment from "qualitative trend" to "quantitative positioning" by predicting energy through the b value and locking space through the fractal dimension, which can more accurately indicate the accurate time of main fracture occurrence, the accurate spatial position of large-scale fracture, and the potential instability energy level, realizing quantitative prediction of time, space, and energy.

[0021] (3) The prediction comprehensiveness is improved: The present application breaks the situation of mutual fragmentation of time, space, and energy information in traditional methods from "fragmentation of three-dimensional information of time, space, and energy" to "cooperative early warning", providing a unified and cooperative analysis framework, which enables researchers to obtain a more comprehensive and in-depth understanding of the whole process of coal rock failure, and provides more sufficient decision basis for precise prevention and control.

[0022] The technical solutions of the present application will be further described in detail below with reference to the accompanying drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1A schematic diagram of the overall process of the coal rock damage short-term prediction method provided by the application based on acoustic emission b value and fractal dimension cross early warning is shown.

[0024] Figure 2 A schematic diagram of the flow of the time series prediction model provided by the application is shown.

[0025] Figure 3 A schematic diagram of the flow of the spatial prediction model provided by the application is shown.

[0026] Figure 4 A schematic diagram of the flow of the energy prediction model provided by the application is shown.

[0027] Figure 5 A case of the time series prediction model in the application. DETAILED DESCRIPTION

[0028] As shown in Figures 1-4 The coal rock damage short-term prediction method provided by the application based on acoustic emission b value and fractal dimension cross early warning comprises the following steps: Step 1: Collect coal rock acoustic emission monitoring data of different lithology, and construct an acoustic emission monitoring database; Step 2: Based on the coal rock acoustic emission monitoring data of the acoustic emission monitoring database, respectively establish a time series prediction model, a spatial prediction model and an energy prediction model, and combine the three models to obtain a coal rock damage short-term prediction system; Step 3: Collect coal rock acoustic emission monitoring data to the coal rock damage short-term prediction system, and output to obtain the damage time, damage location and damage energy; The input of the coal rock damage short-term prediction system is different lithology coal rock acoustic emission monitoring data, and the output is damage time, damage location and damage energy.

[0029] The method provided by the application comprises collecting coal rock acoustic emission (AE) monitoring data of different lithology (sandstone sample, siltstone sample, coal sample, etc.), constructing an acoustic emission monitoring database of different lithology, further constructing a time series prediction model, a spatial prediction model and an energy prediction model, and constructing a coal rock damage short-term prediction system based on the time and space strong prediction model. When performing test prediction, the basic parameters (lithology, geometric parameters) of the coal rock sample to be predicted are input into the short-term prediction system, and the system finally outputs the time, spatial position and energy size of the coal rock damage.

[0030] In step 1, a large amount of coal rock acoustic emission (AE) monitoring data of different lithology is collected and obtained, so as to construct a basic database. The basic database samples mainly include lithology, stress-strain curve, acoustic emission ringing, counting, elastic modulus, Poisson's ratio and other related physical and mechanical parameters.

[0031] On the basis of the basic database, the time parameters, space parameters, energy parameters, geometric parameters, and index parameters (acoustic emission b value and acoustic emission fractal dimension) of the compression failure of different samples were calculated. The calculation methods of acoustic emission b value and fractal dimension are as follows: For the calculation of acoustic emission b value of rock mass failure in rock mechanics test, the magnitude in the "G-R" relationship is replaced by amplitude to obtain an equivalent acoustic emission magnitude similar to the magnitude of earthquake, and the actual event number of magnitude is obtained, which has the following formula:

[0032] In the formula, N is the sum of the number of acoustic emission in the magnitude grade, that is, the actual event number of magnitude; a is a parameter representing seismic activity, that is, the frequency of microseismic events; b is the distribution ratio of acoustic emission events in a certain time period or spatial segment (that is, b value), which represents the activity level of the coal rock acoustic emission; A dB is the maximum amplitude of acoustic emission events in decibels. The magnitude interval is set to 5 dB, and the b value is obtained by the least square method.

[0033] In the formula, (A dB / 20) is the magnitude, which is represented by M. In practice, the energy frequency relationship of microseismic data can be converted into the magnitude frequency relationship of G-R relationship, and the specific calculation method of M is as follows: lgE=4.8+15M In the formula, E is the energy released by the source.

[0034] Based on the above M and N, the b value is calculated, and the main method is the least square method, and the absolute value of the slope of the fitting straight line is taken as the calculation result, and the calculation method is as follows:

[0035] In the formula, m is the total number of magnitude grades; M i is the magnitude of the first i grade; N i is the actual event number of the first i magnitude grade.

[0036] The acoustic emission fractal dimension value of coal rock failure is the same as the b value, which is time series data, and the commonly used algorithm is Gassberger-Procaccia (G-P) algorithm. G-P algorithm takes the time series signal of acoustic emission monitoring in the process of coal rock failure as the research object, which corresponds to each acoustic emission time series Xi, i=1, 2, …, N:

[0037] In the formula, xi is the amplitude of acoustic emission, and N is the sequence length.

[0038] The phase space is reconstructed with delay time and embedding dimension m. The time series is taken with sliding points, and the phase points of the phase space are:

[0039] In the formula, M is the number of phase points of the phase space.

[0040] The correlation integral is a cumulative distribution function, which represents the probability that the distance between any two phase points in the phase space is less than the neighborhood radius, and is used to characterize the degree of phase point aggregation. Given a neighborhood radius r, the correlation integral is defined as follows:

[0041] In the formula, is the Heaviside function, if , if , is the infinity norm.

[0042] When r tends to 0, if the strange attractor is a one-dimensional structure, C(m, r) is proportional to r, if it is a two-dimensional structure, C(m, r) is proportional to r2, and so on. The strange attractor of a D-dimensional structure has C(m, r) proportional to rD. That is, there is always a corresponding correlation function C(m, r) corresponding to the value of the neighborhood radius r. Take multiple sets of coordinate values, draw the relationship between C(m, r) and r in the double logarithmic coordinate system, and perform a linear fitting in the coordinate system to obtain the slope of the regression function, which is the fractal dimension, that is:

[0043] In the formula, D is the fractal dimension of the time series to be solved.

[0044] When calculating the acoustic emission b value and fractal dimension, a certain number of acoustic emission events in the time window can ensure the stability of the b value fitting result, but a too large window is easily affected by early data, and a too small window is more volatile. Different calculation windows and sliding steps will affect the calculation results, but the overall trend is basically the same. Referring to the characteristics of the experiment, the calculation window is selected as 200 event numbers, and the sliding step is 20 in the time window. The b value and fractal dimension are repeatedly calculated.

[0045] Based on the above calculation, a model is established to determine the coal rock acoustic emission monitoring data, including time, ring count, energy, and event coordinates.

[0046] The time series prediction model takes time, ring count, and energy as input, and takes failure time as output. The space prediction model takes time, ringing count and energy as input, and takes damage position and damage energy as output; The energy prediction model takes time, ringing count, energy and event coordinates as input, and takes damage time and damage energy as output.

[0047] The data processing flow of the time sequence prediction model comprises: Step A1: According to the continuous time, ringing count and energy, the b value and the fractal dimension of the acoustic emission are continuously calculated, and the b value and the fractal dimension of the continuous acoustic emission are normalized; Step A2: Determine whether the normalized b value and fractal dimension of the continuous acoustic emission are continuously decreasing, if yes, proceed to step A3, if no, return to step A1; Step A3: Determine whether the stress value of the current coal rock exceeds 60% of the estimated coal rock, if yes, proceed to step A4, if no, return to step A1; Step A4: Start the prediction process, calculate the intersection point of the normalized b value and fractal dimension of the continuous acoustic emission, and continuously calculate the descending slope value of the b value and fractal dimension of the acoustic emission, if the slope deviation is greater than 50%, proceed to step A5, if the slope deviation is less than or equal to 50%, return to step A1; Step A5: Calculate the intersection point of the b value and fractal dimension of the continuous acoustic emission with the time axis respectively to obtain the damage time interval; Step A6: Perform steps A1-A6 on both sides to obtain the damage time interval of the continuous two times, and determine whether the damage time interval is gradually reduced, if yes, output the latest damage time interval, if no, return to step A1; Step A7: Determine whether the coal rock is damaged, if yes, output the latest damage time interval, if no, return to step A1.

[0048] Further, in step 4, calculating the intersection point of the normalized b value and fractal dimension of the continuous acoustic emission comprises: Preliminary calculation of intersection point, and determine whether the normalized b value and fractal dimension of the continuous acoustic emission intersect; If intersected, output the complementary calculated intersection point, if not intersected, return to step A1.

[0049] In the application, the b value and the fractal dimension can better represent the precursor information of the coal rock mass failure instability. In the mechanical test, the b value and the fractal dimension change trend is close at the initial loading stage of the test piece, with the continuous increase of the damage degree of the coal rock sample and the stress level, the b value and the fractal dimension rapidly decrease as a whole, until the coal rock mass completely loses the bearing capacity, and the values reach the minimum or maximum. The extreme value and the change trend of the two values can be used as the precursor information of the early warning, and there is a common rule and mutual unity in the prediction and prediction application of the rock burst. When the acoustic emission or microseismic activity is active and then suddenly decreases or is in a calm stage, it should be analyzed in time, so that the reasonable prediction can be made. In addition, it can be observed that within the first 80% of the critical failure of the coal rock, the two values will appear a significant downward trend, and after the scale normalization, when the two normalized values intersect on the time axis, the coal rock sample will fail in a short period of time. Based on this theory, a time prediction model is constructed.

[0050] In the specific implementation, the time prediction model works as follows: 1) First, the monitoring data sequence is continuously obtained from the acoustic emission monitoring system, including time data, ring count and ring energy data 2) The b value and the fractal dimension of the acoustic emission are continuously calculated according to the calculation window and the sliding step, and the obtained values are continuously normalized, and the original values and the normalized values calculated are stored.

[0051] 3) Determine whether the two values have a continuous downward trend, if yes, go to step 4, otherwise go to step 2.

[0052] 4) Determine whether the current stress value exceeds 60% of the estimated strength, if yes, go to step 5, otherwise go to step 2.

[0053] 5) Start the prediction process and calculate the intersection point. Specifically, the algorithm automatically determines whether the two values intersect, and if the two normalized values are continuous 0, go to step 6, otherwise recalculate.

[0054] 6) After determining that the two values intersect, the slope of the continuous calculation of the original values of the two parameters is calculated, and it is judged whether the slope value of the two parameters is greater than 50%, if yes, continue to calculate, otherwise go to step 7.

[0055] 7) Calculate the intersection point of the straight line of the slope of the two parameters and the time axis, so as to obtain the damage time interval. Determine whether the time interval obtained by the continuous two times is gradually reduced, if yes, go to step 8, otherwise continue to operate.

[0056] 8) Update and output the damage time interval, and determine whether the coal sample is damaged, if yes, go to step 9, otherwise continue the calculation process to gradually approach the damage time interval.

[0057] 9) output the last time interval before failure as the final output result of the time prediction model.

[0058] The data processing flow of the spatial prediction model comprises: Step B1: receiving time, ring count, energy, and event coordinates; Step B2: dividing a spatial cubic grid according to the size of the coal rock, and calculating the multi-element acoustic emission parameters in the cubic grid; Step B3: taking the multi-element acoustic emission parameters in the cubic grid and the maximum acoustic emission energy in the cubic grid as a first data set, inputting the data set into a first machine learning model that has been pre-trained, and outputting the failure location and failure energy; The first machine learning model takes the first data set as input and outputs the failure location and failure energy.

[0059] Further, the first machine learning model is an XGBoost model, and the XGBoost model optimizes the model parameters through a grid search algorithm and a cross-validation algorithm to determine the optimal hyperparameters of the model.

[0060] In specific implementation, the working steps of the spatial prediction model are: 1) input a large number of acoustic emission monitoring data sequences (time, ring count, energy, and event coordinates) of coal rock samples 2) divide a spatial cubic grid according to the size of the sample, and distribute acoustic emission events according to their coordinates and time into the corresponding grid to form acoustic emission data sets of different spatial regions. Calculate the multi-element acoustic emission monitoring parameters (ring count, energy size, b value, fractal dimension, seismic quiescence, and seismic activity S value) in the cubic grid. These monitoring indicators are used to describe the precursor information of the coal rock sample failure, reflect the abnormal characteristics before the failure of the coal rock sample, and have high sensitivity. Therefore, by analyzing the acoustic emission parameters in the grid window, the dynamic evolution process of the crack in the spatial region of the sample can be tracked, and the cumulative damage of the coal body in the spatial region can be evaluated for accurate prediction of the spatial position.

[0061] 3) construct a spatial data set with the multi-element acoustic emission parameters as data features and the maximum acoustic emission energy in the cubic grid as a label. Divide the data set into training samples (80% of the total number of samples) and test samples (20% of the total number of samples) to prepare for model training.

[0062] 4) establish an XGBoost (extreme gradient boosting tree) machine learning model, train it using the training samples, and optimize the model parameters through the GridSearchCV (grid search algorithm and cross-validation) method to determine the optimal hyperparameter combination of the model. After that, the coal rock failure short-term prediction spatial prediction model is obtained.

[0063] 5) test sample is input into the spatial prediction model, and the model predicts and outputs the damage risk area and the damage risk level size.

[0064] The data processing procedure of the energy prediction model comprises: Step C1: receiving time, ringing count, energy and event coordinates, and calculating multivariate acoustic emission parameters in time sequence; Step C2: taking the multivariate acoustic emission parameters in time sequence and the maximum acoustic emission energy per second as a second data set, inputting the data set into a second machine learning model which has been pre-trained, and outputting damage time and damage energy; The second machine learning model takes the second data set as input and outputs damage time and damage energy.

[0065] Further, the second machine learning model is a CNN+LSTM time sequence model, and the CNN+LSTM time sequence model optimizes model parameters through a grid search algorithm and a cross-validation algorithm to determine optimal hyperparameters of the model.

[0066] In a specific implementation, the working steps of the energy prediction model are: 1) input acoustic emission monitoring data sequences (time, ringing count, energy, event coordinates) of a large number of coal rock samples 2) form different time sequence acoustic emission data sets. Calculate multivariate acoustic emission monitoring parameters (ringing count, energy size, b value, fractal dimension, missing earthquake, mine earthquake activity S value) in time sequence 3) take multivariate acoustic emission parameters as data features, take the selected maximum acoustic emission energy in a prediction time length of 1s as a label to construct a time sequence data set, and also divide the data set into training samples and test samples according to 8:2.

[0067] 4) establish a CNN (convolutional neural network) + LSTM (long short-term memory neural network) time sequence model, train the training samples, and optimize model parameters through a grid search algorithm and a cross-validation method to determine an optimal hyperparameter combination of the model, and obtain a coal rock damage short-term prediction energy prediction model.

[0068] 5) input test samples into the spatial prediction model, and the model predicts and outputs the energy size and damage time point of the sample damage.

[0069] The key point of the present application is to creatively cross and fuse analysis of the acoustic emission b value and the acoustic emission event space-time fractal dimension which are two parameters with different physical meanings, so as to take them as core criteria for coal rock sample short-term damage prediction.

[0070] (1) It pioneered the idea and specific early warning algorithm for short-term rock prediction based on the cross-decreasing of b-value and fractal dimension, and realized a comprehensive method for predicting the time, spatial location and energy of coal and rock failure based on the dynamic synergistic evolution law of b-value and fractal dimension.

[0071] (2) A data processing algorithm for real-time acquisition of acoustic emission signals, a data processing unit for calculating b-value and fractal dimension and executing cross-early warning algorithm, a forecasting algorithm module for short-term forecasting of coal and rock time, space and energy, and an early warning module for issuing alarms were constructed.

[0072] (3) Based on the laboratory study of coal and rock mechanical behavior, this method can be extended to the intelligent evaluation of rockburst, mine tremors, rock bursts and other fields, and can be reliably applied in disaster monitoring and early warning at engineering sites. It provides feasible ideas and algorithm design guidance for intelligent prediction and forecasting of mines.

[0073] The present invention describes one example of a time series prediction model as follows, such as... Figure 5 As shown, the prediction model iterated through two prediction interval outputs. In this case, two crossover events occurred, both of which met the necessary conditions for judgment in the process. The first crossover resulted in a prediction interval of [281s~284s] based on the slope intersection. The coal sample remained intact after this, and the model continued real-time monitoring. After the second crossover, the prediction interval was updated and re-output based on the slope intersection, becoming [320s~339s]. The coal sample was destroyed at 332s, demonstrating the accuracy of the predicted time interval. Furthermore, the model combined spatial and energy prediction models to provide the specific time, spatial location, and energy magnitude of the destruction.

[0074] This invention provides a standardized, quantitative, and intelligent forecasting and evaluation system for short-term prediction of coal and rock failure. This technical system has broad application prospects and significant engineering value in the fields of short-term rock mechanics prediction, coal mine tremors, and rockburst prevention, mainly reflected in the following three dimensions: Basic scientific research: It can serve as a powerful tool for in-depth research on the fracture mechanism of brittle materials such as coal and rock. By revealing the intrinsic co-evolution law of acoustic emission parameters before instability, it can intuitively demonstrate the precursor information and early warning principle of material failure, and promote the development of solid failure theory.

[0075] Precise evaluation of mine tremors and rockbursts: At the laboratory level, this method can provide a more accurate and reliable quantitative indicator for evaluating the rockburst tendency of coal and rock under different mining areas and mechanical conditions. Simultaneously, the acoustic emission short-term prediction method can be analogous to the microseismic system for mine monitoring. Based on this prediction method, it can be extended to engineering practice to design intelligent prediction and forecasting systems for rockbursts and mine tremors, serving the design of safe mining operations.

[0076] Frontier technology development: the core algorithm and early warning strategy established by the method can be further transplanted and developed into a dedicated laboratory acoustic emission real-time analysis and early warning software system, forming a core technology product with independent intellectual property rights, which has significant market value.

[0077] The above is only a preferred embodiment of the present application, not any limitation on the present application, any simple modification, change and equivalent structure change of the above embodiment according to the technical essence of the present application are still within the protection scope of the technical solution of the present application.

Claims

1. A short-term prediction method for coal and rock failure based on the cross-prediction of acoustic emission b-value and fractal dimension, characterized in that, include: Step 1: Collect acoustic emission monitoring data of coal and rock with different lithologies, and construct an acoustic emission monitoring database; Step 2: Based on the acoustic emission monitoring data of coal and rock from the acoustic emission monitoring database, establish a time-series prediction model, a spatial prediction model, and an energy prediction model respectively, and merge the three models to obtain a short-term prediction system for coal and rock failure; Step 3: Collect acoustic emission monitoring data of coal and rock into the short-term prediction system for coal and rock failure, and output the failure time, failure location and failure energy; The input of the short-term prediction system for coal and rock failure is acoustic emission monitoring data of coal and rock with different lithologies, and the output is failure time, failure location and failure energy.

2. The short-term prediction method for coal and rock failure based on the cross-prediction of acoustic emission b-value and fractal dimension according to claim 1, characterized in that, The coal and rock acoustic emission monitoring data includes: time, ring count, energy, and event coordinates.

3. The short-term prediction method for coal and rock failure based on the cross-prediction of acoustic emission b-value and fractal dimension according to claim 2, characterized in that, The time-series prediction model takes time, ring count, and energy as inputs and destruction time as output. The spatial prediction model takes time, ring count, and energy as inputs and destruction location and destruction energy as outputs. The energy prediction model takes time, ring count, energy, and event coordinates as inputs, and outputs destruction time and destruction energy.

4. The short-term prediction method for coal and rock failure based on the cross-prediction of acoustic emission b-value and fractal dimension according to claim 3, characterized in that, The data processing flow of the time series prediction model includes: Step A1: Calculate the b-value and fractal dimension of the acoustic emission continuously based on the continuous time, ring count, and energy, and normalize the b-value and fractal dimension of the continuous acoustic emission. Step A2: Determine whether the b-value and fractal dimension of the continuous acoustic emission after normalization decrease continuously. If yes, proceed to step A3; otherwise, return to step A1. Step A3: Determine whether the current stress value of the coal exceeds 60% of the estimated stress value. If yes, proceed to step A4; otherwise, return to step A1. Step A4: Start the forecast process, calculate the intersection of the normalized continuous acoustic emission b-value and fractal dimension, and continuously calculate the decreasing slope of the acoustic emission b-value and fractal dimension. If the slope deviation is greater than 50%, proceed to step A5; if the slope deviation is less than or equal to 50%, return to step A1. Step A5: The intersection points of the continuously calculated acoustic emission b-value and fractal dimension with the time axis are used to obtain the destruction time interval; Step A6: Execute steps A1-A6 consecutively on both sides to obtain two consecutive destruction time intervals, and determine whether the destruction time interval gradually shrinks. If so, output the latest destruction time interval; otherwise, return to step A1. Step A7: Determine whether the coal and rock have been damaged. If so, output the latest damage time interval; otherwise, return to step A1.

5. The short-term prediction method for coal and rock failure based on the cross-prediction of acoustic emission b-value and fractal dimension according to claim 4, characterized in that, In step 4, the intersection of the normalized continuous acoustic emission b-value and the fractal dimension is calculated, including: Preliminary calculation of the intersection point, and determination of whether the b-value and fractal dimension of the normalized continuous acoustic emission intersect; If they intersect, output the intersection point calculated by complementary operations; if they do not intersect, return to step A1.

6. The short-term prediction method for coal and rock failure based on the cross-prediction of acoustic emission b-value and fractal dimension according to claim 3, characterized in that, The data processing flow of the spatial prediction model includes: Step B1: Receive time, ring count, energy, and event coordinates; Step B2: Divide the space into three-dimensional grids according to the size of the coal and rock, and calculate the multi-element acoustic emission parameters within the three-dimensional grid; Step B3: Use the multivariate acoustic emission parameters and the maximum acoustic emission energy within the 3D mesh as the first dataset, input the dataset into the pre-trained first machine learning model, and output the destruction location and destruction energy; The first machine learning model takes the first dataset as input and the destruction location and destruction energy as output.

7. The short-term prediction method for coal and rock failure based on the cross-prediction of acoustic emission b-value and fractal dimension according to claim 6, characterized in that, The first machine learning model is an XGBoost model, which uses a grid search algorithm and a cross-validation algorithm to optimize the model parameters and determine the optimal hyperparameters of the model.

8. The short-term prediction method for coal and rock failure based on the cross-prediction of acoustic emission b-value and fractal dimension according to claim 3, characterized in that, The data processing flow of the energy prediction model includes: Step C1: Receive time, ring count, energy, and event coordinates, and calculate the multi-element acoustic transmission parameters in the timing sequence; Step C2: Use the time-series multivariate acoustic emission parameters and the maximum acoustic emission energy per second as the second dataset, input the dataset into the pre-trained second machine learning model, and output the destruction time and destruction energy; The second machine learning model takes the second dataset as input and the destruction time and destruction energy as output.

9. The short-term prediction method for coal and rock failure based on the cross-prediction of acoustic emission b-value and fractal dimension according to claim 8, characterized in that, The second machine learning model is a CNN+LSTM time series model. The CNN+LSTM time series model optimizes the model parameters through grid search algorithm and cross-validation algorithm to determine the optimal hyperparameters of the model.

Images