Underground engineering rockburst dynamic prediction method and device based on machine learning

Abstract

The invention provides an underground engineering rockburst dynamic prediction method and device based on machine learning, and relates to the technical field of tunnel engineering safety monitoring. According to the method, three-dimensional classification labels of rock materials, burial depth intervals and hardness intervals are constructed, four types of actual measurement data including underground water pressure, rock temperature, sound wave speed and excavation footage are collected, the actual measurement data intervals are classified under the three-dimensional classification labels after being divided, and under each three-dimensional classification label, the three-dimensional classification labels of the underground water pressure, the rock temperature, the sound wave speed and the excavation footage are obtained. The method comprises the following steps of: counting rockburst and non-rockburst probabilities of each interval based on historical data, calculating a posterior probability by utilizing a Bayesian formula, fusing a quality function of multi-source evidence through a D-S evidence theory, constructing a comprehensive confidence coefficient, matching a three-dimensional classification label in real time according to new measured data in a new underground engineering implementation process, inputting the new data into a model, and constructing a new underground engineering model. And outputting the minimum confidence coefficient and the maximum confidence coefficient, constructing a probability formula based on the minimum confidence coefficient and the maximum confidence coefficient, and comparing a result with a preset threshold value to judge a rockburst occurrence probability level.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel engineering safety monitoring, and in particular to a method and device for dynamic prediction of underground engineering rockburst based on machine learning. Background Art

[0002] Rockburst is a sudden dynamic disaster caused by high ground stress during underground engineering construction, which seriously threatens construction safety and efficiency. As underground engineering extends to deep and complex geological environments, rockburst prediction has become a core problem in the field of geotechnical engineering. Traditional prediction methods mainly rely on empirical formulas, single physical indicators or numerical simulations, but under the coupling of multiple factors, such as geological structure, groundwater, and construction disturbances, their prediction accuracy and real-time performance are difficult to meet engineering needs. In recent years, machine learning technology has been introduced into the field of rockburst prediction. Through data-driven methods, the correlation between multi-source monitoring data and rockburst events is mined, providing new ideas for dynamic risk assessment. However, existing technologies still have significant bottlenecks in multi-source data fusion, uncertainty modeling and dynamic prediction mechanisms.

[0003] Prior art, publication number CN117332240B discloses a rockburst prediction model construction method, storage medium, rockburst prediction method, and system. This technology analyzes the types and content ratios of elements such as silicon, calcium, and iron in the rocks within the construction area, combines historical rockburst level data to construct a multi-class dataset, and utilizes machine learning algorithms such as random forests and neural networks to train the rockburst prediction model, enabling dynamic assessment of rockburst risk in the early stages of construction. However, the underground engineering environment is not always fixed. Relying solely on microscopic element information without integrating macroscopic engineering parameters limits the model's prediction accuracy under the influence of multiple coupled factors. Static models based on historical data cannot update weights in real time, making it difficult to adapt to dynamic changes in geological conditions during construction. Furthermore, the model lacks adaptability to different engineering scenarios.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and device for dynamic prediction of underground engineering rockburst based on machine learning to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The dynamic prediction method of underground engineering rockburst based on machine learning includes the following steps:

[0008] Step 1: Collect measured data from historical underground engineering data and divide it into intervals. The rock material at the underground engineering location is collected, and the depth of the underground engineering location is divided into intervals. The rock hardness is classified by uniaxial compressive strength, and a three-dimensional classification label is constructed for material, depth, and hardness. The measured data includes groundwater pressure, rock temperature, acoustic wave velocity, and excavation footage.

[0009] Step 2: Under each three-dimensional classification label, count the number of times each measured data falls within each measured data interval, construct a measured data set, and calculate the probability of the rockburst data interval and the probability of the non-rockburst data interval. Calculate the overall probability of the measured data interval based on the probability of the rockburst data interval and the non-rockburst data interval, and then use the Bayesian formula to calculate the posterior probability of rockburst.

[0010] Step 3: Construct the DS evidence theory model, map the rockburst posterior probability into a quality function in proportion, fuse the quality functions through the Dempster rule, and construct a comprehensive quality function;

[0011] Step 4: Dynamically match the three-dimensional classification labels for the underground engineering construction material, burial depth interval and hardness interval, collect new measured data and divide the measured data interval, input the new measured data into the measured data interval and output the minimum confidence and maximum confidence after entering the DS evidence theory model, and predict the rock burst probability based on the minimum confidence and maximum confidence.

[0012] Furthermore, the method of collecting measured data and dividing the measured data intervals is as follows:

[0013] Set the sampling time interval to T cj , in minutes, and T cj >0, according to the collection time interval T cj Collect four types of measured data: groundwater pressure, rock temperature, sound wave velocity and excavation footage, denoted as x i , where i = 1, 2, 3, 4, represents the index, representing groundwater pressure, rock temperature, acoustic wave velocity and excavation footage respectively;

[0014] For each measured data x i , divided into several continuous intervals:

[0015]

[0016] Where a i，j represents the lower bound of the jth interval of the i-th type of measured data, a i，j+1 represents the upper bound of the jth interval of the i-th type of measured data, represents the jth interval of the i-th type of measured data, where j = 1, 2, ..., z i , z iIndicates the total number of intervals divided by the i-th category data.

[0017] Furthermore, the depth of the underground project is divided into depth intervals, and the rock hardness is divided by uniaxial compressive strength as follows:

[0018] The burial depth range is divided according to the actual depth of the underground project:

[0019]

[0020] Where D ms represents the classification label of the buried depth interval, and D represents the actual depth of the underground project;

[0021] The rock is tested through uniaxial compressive strength and the hardness range is divided into:

[0022]

[0023] Where U yd represents the hardness classification label, and UCS represents the uniaxial compressive strength.

[0024] Furthermore, the number of times each measured data falls within each measured data interval is counted, and the method for constructing the measured data set and calculating the probability of the rockburst data interval and the probability of the non-rockburst data interval is as follows:

[0025] Preset the recording period to T zq , that is, the length of a single time window, in minutes, and T zq ≥30, set the sliding step length to 30 minutes;

[0026] Calculate the number of data points collected during the logging period:

[0027]

[0028] Where N nub Indicates the number of data points collected during the recording period;

[0029] Calculate the number of data point slides:

[0030]

[0031] Where S represents the number of data point sliding;

[0032] Record the measured data of each time window. At this time, the number of overlapping data points in adjacent time windows is:

[0033] O=N nub -S

[0034] Where O represents the number of overlapping data points in adjacent time windows;

[0035] According to the divided time windows, four types of measured data, namely groundwater pressure, rock temperature, acoustic wave velocity, and excavation footage, are recorded window by window. The number of time windows where rockburst occurs and does not occur is counted in a ratio of 1:4. The total number of accumulated data points is more than 1000 to generate a measured data set. Based on the generated measured data set, the number of times the groundwater pressure, rock temperature, acoustic wave velocity, and excavation footage are in each measured data interval is counted. The number of times each measured data is in each interval when rockburst occurs, the total number of rockburst occurrences, the number of times each measured data is in each interval when rockburst does not occur, and the total number of times rockburst does not occur are counted. The probability of the rockburst data interval is calculated:

[0036]

[0037] Where, N represents the number of times the i-th measured data is in interval j when the rock burst occurs. E represents the total number of rock bursts, represents the probability of rock burst data interval, that is, the measured data x under the condition of rock burst i The probability in interval j, E represents the label of rockburst;

[0038] Calculate the probability of non-rockburst data interval:

[0039]

[0040] Where, represents the number of times the i-th measured data is in interval j when rock burst does not occur, Indicates the total number of times rock burst did not occur, represents the probability of the non-rockburst data interval, that is, the measured data x under the condition that rockburst does not occur i The probability in interval j is, Label indicating non-rockburst.

[0041] Furthermore, the overall probability of the measured data interval is calculated based on the probability of the rockburst data interval and the probability of the non-rockburst data interval, and then the method of calculating the posterior probability of rockburst using the Bayesian formula is as follows:

[0042] First, extract the total number of rock bursts N E Total number of rock bursts that did not occur Calculate the prior probability of rockburst:

[0043]

[0044] Where P(E) represents the prior probability of rockburst;

[0045] Using the total probability formula, calculate the measured data x i The overall probability in the jth interval is:

[0046]

[0047] Where, Represents data x i The overall probability in the jth interval, where

[0048] The posterior probability of rockburst is calculated using the Bayesian formula:

[0049]

[0050] Where, represents the posterior probability of rockburst.

[0051] Furthermore, the DS evidence theory model is constructed, and the method of mapping the rockburst posterior probability into a mass function in proportion is as follows:

[0052] Extracting rockburst posterior probability Mapped to DS quality function according to the following formula:

[0053] Construct the rockburst mass function:

[0054]

[0055] Where, Represents the rockburst mass function, that is, the measured data x i In interval j, the mass assignment of rock burst is supported, α represents the scaling factor, and the value is α = 0.9. When i = 1, the rock burst mass function is marked as m1(E), when i = 2, the rock burst mass function is marked as m2(E), when i = 3, the rock burst mass function is marked as m3(E), and when i = 4, the rock burst mass function is marked as m4(E);

[0056] Construct the non-rockburst mass function:

[0057]

[0058] Where, represents the non-rockburst mass function, that is, the mass assignment that supports non-rockburst. When i = 1, the non-rockburst mass function is marked as m1(E); when i = 2, the non-rockburst mass function is marked as m2(E); when i = 3, the non-rockburst mass function is marked as m3(E); when i = 4, the non-rockburst mass function is marked as m4(E);

[0059] Construct the uncertainty quality function:

[0060]

[0061] Where, represents the uncertainty mass function, i.e., the mass assigned to the measured data due to uncertainty, where When i=1, the uncertainty mass function is labeled as m1(Θ), when i=2, the uncertainty mass function is labeled as m2(Θ), when i=3, the uncertainty mass function is labeled as m3(Θ), and when i=4, the uncertainty mass function is labeled as m4(Θ).

[0062] Furthermore, the method of constructing a comprehensive quality function by fusing the quality functions with the Dempster rule is as follows:

[0063] The mass functions are combined using the Dempster combination rule. First, the mass functions of groundwater pressure and rock temperature are combined to calculate the intersection mass of groundwater pressure and rock temperature:

[0064]

[0065] Where k 12 represents the conflict coefficient between groundwater pressure and rock temperature,

[0066] Then, the mass function of the acoustic wave velocity is integrated to calculate the intersection mass of groundwater pressure, rock temperature, and acoustic wave velocity:

[0067]

[0068] Where k 123 represents the conflict coefficient of groundwater pressure, rock temperature and sound wave velocity,

[0069] Then, the mass function of the excavation footage is integrated to calculate the intersection mass of groundwater pressure, rock temperature, acoustic wave velocity, and excavation footage:

[0070]

[0071] Where m total represents the comprehensive quality function, k total It represents the conflict coefficient of groundwater pressure, rock temperature, sound wave velocity and excavation footage.

[0072] Furthermore, the measured data is divided into intervals and then input into the DS evidence theory model to output the minimum confidence and maximum confidence as follows:

[0073] According to the real-time rock material, burial depth and uniaxial compressive strength, the new measured data is matched with the three-dimensional classification label. Divide the intervals and input them into the DS evidence theory model to extract the rockburst data interval probability of the new measured data Non-rockburst data interval probability and the prior probability of rockburst P(E), where Represents the i-th category of real-time data The interval index to which it belongs, Substitute the Bayesian formula to calculate the posterior probability of rockburst:

[0074]

[0075] Where, represents the posterior probability of rockburst for new measured data;

[0076] Quality function construction:

[0077]

[0078] Where, Indicates measured data In interval j, the mass assignment supporting rock burst is: Indicates that non-rockburst mass assignment is supported. It represents the quality assigned to the measured data due to uncertainty;

[0079] Calculate m1(E), m2(E), m3(E), m4(E), m1(Θ), m2(Θ), m3(Θ);

[0080] Use Dempster's combination rule to combine the quality functions:

[0081]

[0082] Furthermore, the method for predicting the rockburst probability based on the minimum confidence level and the maximum confidence level is as follows:

[0083] Preset the threshold T of rock burst probability high and T low , and T high >T low >0, the new measured data m total (E) Output is the minimum confidence Bel(E), extracting the m of the new measured data total (E) and m total (Θ) Calculate the maximum confidence:

[0084] Pl(E)=m total (E)+m total (Θ)

[0085] Where Pl(E) represents the maximum confidence level;

[0086] Construct the formula for the probability of rock burst occurrence:

[0087]

[0088] Where, P pred (E) represents the probability of rock burst;

[0089] When P pred (E)≥T high , then the probability of rock burst is judged to be high; when T low ≤P pred (E) <T high , then the probability of rock burst is judged to be medium; when P pred (E) <T low , then the probability of rock burst is judged to be low.

[0090] See also Figure 11 The present invention further provides a device for predicting underground engineering rockburst dynamics based on machine learning, wherein the device is used to execute the above-mentioned method for predicting underground engineering rockburst dynamics based on machine learning, comprising:

[0091] A label construction module is used to collect measured data from historical data of underground projects and divide the measured data into intervals. The module collects the rock material of the underground project location, divides the burial depth of the underground project location into intervals, divides the rock hardness by uniaxial compressive strength, and constructs three-dimensional classification labels for material, burial depth, and hardness. The measured data includes groundwater pressure, rock temperature, sound wave velocity, and excavation footage.

[0092] The posterior probability module is used to count the number of times each measured data falls within each measured data interval under each three-dimensional classification label, construct a measured data set, and calculate the probability of rockburst data interval and the probability of non-rockburst data interval. The overall probability of the measured data interval is calculated based on the probability of rockburst data interval and non-rockburst data interval, and then the posterior probability of rockburst is calculated using the Bayesian formula.

[0093] Theoretical model construction module is used to construct the DS evidence theoretical model, map the rockburst posterior probability into a quality function in proportion, fuse the quality functions through the Dempster rule, and construct a comprehensive quality function;

[0094] The rockburst probability prediction module is used to dynamically match three-dimensional classification labels to underground engineering construction materials, burial depth intervals, and hardness intervals. After collecting new measured data, the measured data intervals are divided. After the new measured data is divided into measured data intervals, it is input into the DS evidence theory model and the minimum and maximum confidence levels are output. The rockburst probability is predicted based on the minimum and maximum confidence levels.

[0095] Compared with the prior art, the present invention has the following beneficial effects:

[0096] The present invention constructs three-dimensional classification labels based on material, burial depth, and hardness, dynamically matches groundwater pressure, rock temperature, acoustic wave velocity, and excavation footage measured data to the three-dimensional classification labels, and adapts to changes in construction conditions in real time, so that the probability of rockburst can be predicted and misjudgment caused by dynamic environmental changes can be avoided. The present invention also uses the DS evidence theory fusion method to jointly quantify the predicted probability of rockburst occurrence by combining groundwater pressure, rock temperature, acoustic wave velocity, and excavation footage measured data, avoiding the limitations of traditional single-data judgment and the misjudgment of rockburst occurrence caused by excessive fluctuations in single measured data. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1 Schematic diagram of the overall method flow of the present invention;

[0098] Figure 2 This is a correlation diagram between groundwater pressure and rock burst in the present invention;

[0099] Figure 3 This is a correlation diagram between rock temperature and rock burst in the present invention;

[0100] Figure 4 This is a correlation diagram between the acoustic wave velocity and rock burst of the present invention;

[0101] Figure 5 This is the correlation diagram between excavation footage and rock burst of the present invention

[0102] Figure 6 The groundwater pressure interval probability and total probability distribution diagram of the present invention;

[0103] Figure 7 The rock temperature interval probability and total probability distribution diagram of the present invention;

[0104] Figure 8 The probability distribution diagram of the sound wave velocity interval and the total probability distribution diagram of the present invention;

[0105] Figure 9 The excavation footage interval probability and total probability distribution diagram of the present invention;

[0106] Figure 10 This is a schematic diagram of the risk levels of the present invention;

[0107] Figure 11 It is a schematic diagram of the overall device flow of the present invention. DETAILED DESCRIPTION

[0108] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0109] It should be noted that, unless otherwise defined, the technical or scientific terms used in the present invention should have the usual meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative position relationships. When the absolute position of the object being described changes, the relative position relationship may also change accordingly.

[0110] Example:

[0111] See also Figures 1 to 10 , the present invention provides a technical solution:

[0112] The dynamic prediction method of underground engineering rockburst based on machine learning includes the following steps:

[0113] Step 1: Collect measured data from historical underground engineering data and divide it into intervals. The rock material at the underground engineering location is collected, and the depth of the underground engineering location is divided into intervals. The rock hardness is classified by uniaxial compressive strength, and a three-dimensional classification label is constructed for material, depth, and hardness. The measured data includes groundwater pressure, rock temperature, acoustic wave velocity, and excavation footage.

[0114] Set the sampling time interval to T cj , in minutes, and T cj >0. As rock bursts are mostly sudden disasters, too long an acquisition time interval will lead to data lag, while too short an acquisition time interval will increase the computing cost. Four types of measured data, namely groundwater pressure, rock temperature, acoustic wave velocity and excavation footage, are collected within a fixed time interval. According to the recommended key parameters described in the Technical Specifications for Railway Tunnel Monitoring and Measurement, such as stress and displacement monitoring intervals that must be less than or equal to 30 minutes, the acquisition time interval T can be set to 0. cj It is set to 10 minutes to meet the real-time requirements of rock burst prediction and can capture the changes of measured data in real time. The four types of measured data are recorded as x i, where i represents the index, i = 1, 2, 3, 4, representing groundwater pressure, rock temperature, acoustic wave velocity, and excavation footage, respectively. High-pressure groundwater will weaken the rock mass strength, increase the fracture water pressure, and induce rock burst. Too low or too high water pressure may indicate internal fracture of the rock mass, which is consistent with the precursor of rock burst; rock friction or energy release will cause the temperature to rise, which is a thermodynamic signal before rock burst; the decrease in acoustic wave velocity reflects the destruction of rock mass integrity and indicates potential crack expansion; too fast footage will accelerate stress concentration and reduce the self-stabilization ability of the surrounding rock. Therefore, these four types of measured data are selected as the data for predicting rock burst.

[0115] For each measured data x i , divided into several continuous intervals:

[0116]

[0117] Where a i，j represents the lower bound of the jth interval of the i-th type of measured data, a i，j+1 represents the upper bound of the jth interval of the i-th type of measured data, represents the jth interval of the i-th type of measured data, where j = 1, 2, ..., z i , z i It represents the total number of intervals divided by the i-th category data. Different intervals may correspond to different rockburst probabilities.

[0118] For example, the continuous interval of groundwater pressure is set to [0,1), [1,2), [2,3), [3,5), with the unit being MPa. High-pressure groundwater will significantly increase the fracture water pressure and weaken the rock mass strength, while the low-pressure area corresponds to dry rock or closed fractures, and the rock burst risk is low. The groundwater pressure in the interval [0,1) is low and may not affect the stability of the rock mass. The interval [1,2) reflects normal water pressure and good rock stability. The interval [2,3) reflects slightly higher water pressure, which may begin to affect the rock mass and increase fractures. The interval [3,5) is in a high water pressure state, which may significantly increase the rock burst risk.

[0119] The continuous interval of rock temperature is set to [20,25), [25,30), [30,35), [35,40), in °C. Friction or stress release in the rock mass before a rockburst can cause the temperature to rise. Therefore, the high temperature area corresponds to a high rockburst probability, the medium temperature area is medium, and the low temperature area is low. The rock temperature is normal in the interval [20,25), and the rock mass generally remains stable. The temperature rises in the interval [25,30), which may be due to friction or energy release. The high temperature in the interval [30,35), which may be a thermal signal before a rockburst. The temperature in the interval [35,40), which is even higher, may indicate serious damage to the rock mass.

[0120] The continuous interval of the acoustic wave velocity is set to [1000, 2500), [2500, 4000), [4000, 5000], with the unit of m / s. The internal cracks of the broken rock mass are connected, the stress concentration is easy to release, and the probability of rock burst is high. The acoustic wave velocity is positively correlated with the integrity of the rock mass. According to the engineering rock mass classification standard, it is set to [1000, 2500), [2500, 4000), [4000, 5000].

[0121] The continuous intervals of excavation footage are set to [0,1.5), [1.5,2.5), and [2.5,4.0), in meters. Increased stress concentration will reduce the self-stabilizing capacity of the surrounding rock and increase the risk of rockburst. According to the railway tunnel construction specifications, it is recommended that the excavation footage for soft rock be less than 1.5. Excavation footage greater than 2.5 will result in insufficient release of surrounding rock stress. Slow advancement in the interval [0,1.5) meets the specifications and results in less rock stress concentration. Fast advancement in the interval [1.5,2.5) may increase stress concentration. Rapid advancement in the interval [2.5,4.0) may reduce the self-stabilizing capacity of the surrounding rock. The intervals [3,5) for groundwater pressure, [35,40) for rock temperature, [1000,2500) for acoustic wave velocity, and [2.5,4.0) for excavation footage are considered high-risk intervals.

[0122] Determine the rock type, such as granite, sandstone, shale, etc., through geological survey reports or on-site sampling, and mark it as R m , m represents the index of rock type. For example, when m=1, R1 represents granite, and when m=2, R2 represents sandstone, etc., so as to clearly distinguish different rock types. The mineral composition, structural structure and physical and mechanical properties of different rock types are significantly different, which has a profound impact on the rock burst mechanism. For example, crystalline igneous rocks such as granite usually have high hardness and elastic modulus due to the dense embedding of mineral particles. They have strong energy accumulation ability under high ground stress environment and high rock burst tendency. Sedimentary rocks such as shale often have thin lamellar bedding structure, weak mechanical properties and prominent anisotropy. Energy release is mainly plastic deformation, and rock burst tendency is low.

[0123] The burial depth range is divided according to the actual depth of the underground project:

[0124]

[0125] Where D ms Indicates the classification label of the buried depth interval. D represents the actual depth of the underground project. The depth of the current excavation position can be obtained based on the engineering design drawings.

[0126] The rock is tested through uniaxial compressive strength and the hardness range is divided into:

[0127]

[0128] Where U yd represents the hardness classification label, UCS represents the uniaxial compressive strength, and the constructed three-dimensional classification label is:

[0129] Ω=R m ×D ms ×U yd

[0130] Where Ω represents the three-dimensional classification label. Rocks of different materials, burial depths, and hardness have significantly different rockburst mechanisms. For example, deeply buried hard rock is more likely to accumulate elastic energy, causing rockbursts, while soft rock may undergo plastic deformation under high stress, reducing the risk of rockburst. Using three-dimensional labels to classify historical data by engineering scenario ensures that real-time data is matched to the most similar historical subset, avoiding data confusion across scenarios.

[0131] Step 2: Under each three-dimensional classification label, count the number of times each measured data falls within each measured data interval, construct a measured data set, and calculate the probability of the rockburst data interval and the probability of the non-rockburst data interval. Calculate the overall probability of the measured data interval based on the probability of the rockburst data interval and the non-rockburst data interval, and then use the Bayesian formula to calculate the posterior probability of rockburst.

[0132] Preset the recording period to T zq , that is, the length of a single time window, in minutes, and T zq ≥30, which is an integer multiple of the collection time interval. Set the sliding step to 30 minutes, and ensure that the sliding step is an integer multiple of the collection time interval to avoid the problem of non-integer overlap caused by the decimal number of data points in the time window;

[0133] Calculate the number of data points collected during a single logging cycle:

[0134]

[0135] Where N nub Indicates the number of data points collected during the recording period;

[0136] Calculate the number of data point slides:

[0137]

[0138] Where S represents the number of data point sliding;

[0139] Record the measured data of each time window. At this time, the number of overlapping data points in adjacent time windows is:

[0140] O=N nub -S

[0141] Where O represents the number of overlapping data points in adjacent time windows;

[0142] Among them, the number of data points collected in a single recording cycle is divided into time windows, and four types of measured data, namely groundwater pressure, rock temperature, sound wave velocity, and excavation footage, are recorded window by window. For example, T zq It is set to 60. According to a large number of engineering practices, a 50% overlapping window is used, which is stable in both sudden and gradual rock bursts, with a low missed reporting rate. Therefore, the sliding step is set to 30 minutes, that is, the first 3 data points are removed each time, and the four types of measured data each record 6 data points. At the same time, it is recorded whether the rock burst occurs when each data point is collected. The measured data are divided into intervals according to the method described above, and three-dimensional classification labels of material, burial depth interval and hardness interval are constructed. Under each three-dimensional classification label, the groundwater pressure, rock temperature, acoustic wave velocity and excavation footage in each measured data interval are statistically analyzed. The number of times, according to the divided time windows, four types of measured data, namely groundwater pressure, rock temperature, acoustic wave velocity, and excavation footage, are recorded window by window. Table 1 shows the rockburst data of underground engineering with a certain three-dimensional classification label, covering four key parameters of groundwater pressure, rock temperature, acoustic wave velocity, and excavation footage, as well as the annotation of whether rockburst occurred within the monitoring period. Each monitoring time window corresponds to an independent engineering scenario, recording the four types of measured data parameters and rockburst status within a certain recording period, providing a structured data foundation for subsequent Bayesian probability calculation, DS evidence theory fusion, and dynamic prediction of rockburst risk based on three-dimensional classification labels.

[0143]

[0144]

[0145]

[0146] Table 1 Measured data statistics

[0147] like Figure 2-Figure 5 As shown in Figure 2, the changing trends of groundwater pressure, rock temperature, acoustic wave velocity and excavation footage under different rock burst conditions with the sequence number of data points are shown. Figure 2 The horizontal axis is the data point number, and the vertical axis is the groundwater pressure. The black line represents the groundwater pressure change before the rock burst occurs, and the red line represents the pressure change when the rock burst occurs. It can be observed that the fluctuation range and trend of the groundwater pressure are different when the rock burst occurs and when it does not occur. Figure 3 The horizontal axis is the data point number, and the vertical axis is the rock temperature. The black line represents the temperature change before the rock burst occurs, and the red line represents the temperature change after the rock burst occurs. By comparing them, we can analyze the impact of whether the rock burst occurs or not on the rock temperature change characteristics. Figure 4The horizontal axis is the data point number, and the vertical axis is the acoustic velocity. The black line represents the change in acoustic velocity before rock burst occurs, and the red line represents the change in velocity when rock burst occurs. This allows us to explore the potential connection between rock burst occurrence and acoustic velocity fluctuations. Figure 5 The horizontal axis is the data point number, and the vertical axis is the excavation footage. The black line represents the change in excavation footage before rockburst occurs, and the red line represents the change in excavation footage when rockburst occurs. By statistically analyzing the changing trends of groundwater pressure, rock temperature, sound wave velocity, and excavation footage before and after rockburst occurs, the correlation between the occurrence of rockburst and the measured data can be analyzed in the subsequent analysis.

[0148] For each 3D classification label, the number of time windows where rockbursts occurred and did not occur was counted in a 1:4 ratio. The total number of accumulated data points was more than 1000, and a measured data set was generated. Based on the generated measured data set, the data was divided into two groups according to whether rockbursts occurred or not: a rockburst occurrence group and a rockburst non-occurrence group. The rockburst data interval probability was then calculated:

[0149]

[0150] Where, N represents the number of times the i-th measured data is in interval j when the rock burst occurs. E represents the total number of rock bursts, represents the probability of rock burst data interval, that is, the measured data x under the condition of rock burst i The probability in interval j, E represents the label of rockburst. The historical data contains the distribution patterns of measured data when rockburst occurs and when it does not occur. By counting the number of occurrences of each data interval and calculating the probability, these patterns can be extracted from the historical data, providing a basis for subsequent rockburst prediction.

[0151] Calculate the probability of non-rockburst data interval:

[0152]

[0153] Where, represents the number of times the i-th measured data is in interval j when rock burst does not occur, Indicates that no rock burst occurred

[0154] Total number of times, represents the probability of the non-rockburst data interval, that is, the measured data x under the condition that rockburst does not occur i The probability in interval j is, Label indicating non-rockburst.

[0155] When the Bayesian formula is used to calculate the posterior probability of rockburst, the probability of rockburst data interval and the probability of non-rockburst data interval are needed. First, the total number of rockburst occurrences N is extracted. ETotal number of rock bursts that did not occur Calculate the prior probability of rockburst:

[0156]

[0157] Where P(E) represents the prior probability of rockburst. In the subsequent dynamic prediction of rockburst in new underground engineering projects, this prior probability of rockburst is dynamically updated to enhance the accuracy of the model.

[0158] Using the total probability formula, calculate the measured data x i The overall probability in the jth interval is:

[0159]

[0160] Where, Represents data x i The overall probability in the jth interval, where

[0161] The posterior probability of rockburst is calculated using the Bayesian formula:

[0162]

[0163] Where, It represents the posterior probability of rockburst, which can directly correspond to the possibility of rockburst occurrence. The prior probability, that is, the rockburst conditional probability in the current data interval, is combined with the Bayesian formula to dynamically update the posterior probability of rockburst occurrence, so that the prediction results can be continuously optimized as new measured data are input.

[0164] Step 3: Construct the DS evidence theory model, map the rockburst posterior probability into a quality function in proportion, fuse the quality functions through the Dempster rule, and construct a comprehensive quality function;

[0165] Extracting rockburst posterior probability Mapped to DS quality function according to the following formula:

[0166] The rockburst mass function is constructed by multiplying the posterior probability by the scaling factor:

[0167]

[0168] Where, Represents the rockburst mass function, that is, the measured data x iIn interval j, the quality of supporting rockburst is assigned. α represents the scaling factor, which is set to α = 0.9. It is a commonly used ratio for quantifying uncertainty in DS evidence theory, indicating that 90% of the quality is assigned to the proposition supporting / opposing rockburst, and 10% is reserved for uncertainty, balancing certainty and unknown factors. DS evidence theory focuses on the support of the main evidence through basic probability allocation. The main evidence is groundwater pressure, temperature, sound wave velocity, and excavation footage to ensure that they play a dominant role. 10% is reserved for uncertainty, leaving room for error for unknown factors and avoiding absolute judgments caused by incomplete data or model assumption deviations. It is to prevent incorrect support allocation caused by fluctuations in a single parameter. 10% uncertainty can reduce its excessive impact on the final result. When i = 1, the rockburst quality function is labeled m1(E); when i = 2, the rockburst quality function is labeled m2(E); when i = 3, the rockburst quality function is labeled m3(E); and when i = 4, the rockburst quality function is labeled m4(E).

[0169] Construct the non-rockburst mass function:

[0170]

[0171] Where, represents the non-rockburst mass function, that is, the mass assignment that supports non-rockburst. When i = 1, the non-rockburst mass function is marked as m1(E); when i = 2, the non-rockburst mass function is marked as m2(E); when i = 3, the non-rockburst mass function is marked as m3(E); when i = 4, the non-rockburst mass function is marked as m4(E);

[0172] Construct the uncertainty quality function:

[0173]

[0174] Where, represents the uncertainty mass function, i.e., the mass assigned to the measured data due to uncertainty, where When i=1, the uncertainty mass function is labeled as m1(Θ), when i=2, the uncertainty mass function is labeled as m2(Θ), when i=3, the uncertainty mass function is labeled as m3(Θ), and when i=4, the uncertainty mass function is labeled as m4(Θ).

[0175] The Dempster combination rule is used to fuse the quality functions. By gradually fusing the four pieces of evidence instead of processing them simultaneously, the problem of misjudgment is reduced. First, the quality functions of groundwater pressure and rock temperature are fused to calculate the intersection quality of groundwater pressure and rock temperature:

[0176]

[0177] Where k 12 represents the conflict coefficient between groundwater pressure and rock temperature,

[0178]

[0179] Then, the mass function of the acoustic wave velocity is integrated to calculate the intersection mass of groundwater pressure, rock temperature, and acoustic wave velocity:

[0180]

[0181] Where k 123 represents the conflict coefficient of groundwater pressure, rock temperature and sound wave velocity,

[0182] Then, the mass function of the excavation footage is integrated to calculate the intersection mass of groundwater pressure, rock temperature, acoustic wave velocity, and excavation footage:

[0183]

[0184] Where m total It represents the comprehensive quality function, and its significance is reflected in the three dimensions of multi-source data fusion, uncertainty quantification and engineering decision support, especially m total (E) The independent judgments of groundwater pressure, rock temperature, acoustic wave velocity and excavation footage are converted into a unified support value. A high value indicates a high probability of rock burst. total It represents the conflict coefficient of groundwater pressure, rock temperature, sound wave velocity and excavation footage.

[0185] DS evidence theory allows different evidence to support the same proposition to different degrees. That is, in the measured data, for example, groundwater pressure fluctuations and temperature measurement errors may cause different sensors to provide contradictory information. The conflict coefficient is used to quantify the severity of this contradiction. Generally, if it exceeds 0.7 or 0.8, the conflict is considered to be large. For example, groundwater pressure supports rock burst but acoustic wave velocity does not support it. In this case, the conflict coefficient k total The value of is too high, so a threshold k of the conflict coefficient can be set. ct , and k ct >0.5, when k total ≥k ct When the data reliability is checked, for example, rock burst prediction requires extremely high reliability. To avoid misjudgment and serious consequences, the threshold of the conflict coefficient can be set to 0.65. At the same time, the conflict coefficient is used for the denominator 1-k total, which can avoid the unreasonable amplification of the results by contradictory evidence. Table 2 shows that under a certain three-dimensional classification label, the number of time windows where rockburst occurred and did not occur follows a ratio of 1:4. 200 rockburst occurrence groups and 800 rockburst non-occurrence groups were collected, totaling 1000 time window data. The number of times the four types of measured data were in different intervals, and the rockburst data interval probability, non-rockburst data interval probability and total probability were calculated according to the above formula, and a table was generated by summarizing them.

[0186]

[0187]

[0188] Table 2 Model parameter statistics

[0189] like Figure 6-Figure 9 As shown in the figure, the rockburst data interval probability, non-rockburst data interval probability and total probability of the four types of measured data in different intervals are intuitively displayed. For example, when the groundwater pressure is in the range of [0,1) and [1,2), the probability of rockburst is only 10%, and the probability of no rockburst is 0.8, which is consistent with the characteristic that low pressure corresponds to stable rock mass. When the rock temperature is in the range of [30,35), the probability of rockburst reaches 0.45, which indicates that the increase in rock temperature may be related to the increase in rockburst risk. After calculating the rockburst data interval probability and non-rockburst data interval probability, it provides a basis for subsequent Bayesian analysis and provides data support for the comprehensive assessment of rockburst risk, thereby making a more accurate multi-dimensional comprehensive prediction of the risk of rockburst.

[0190] Step 4: Dynamically match the three-dimensional classification labels for the underground engineering construction material, burial depth interval, and hardness interval. After collecting new measured data, divide the measured data interval. After the new measured data is divided into measured data intervals, it is input into the DS evidence theory model and the minimum and maximum confidence levels are output. The rockburst probability is predicted based on the minimum and maximum confidence levels.

[0191] According to the real-time collected rock material, burial depth and uniaxial compressive strength, the three-dimensional classification labels are matched and the new measured data are Divide the intervals, assign them to three-dimensional classification labels, and input them into the DS evidence theory model. The entire process from data input to warning output is automatically executed without manual intervention, and the probability of rock burst data intervals from new measured data is extracted. Non-rockburst data interval probability and the prior probability of rockburst P(E), where Represents the i-th category of real-time data The interval index to which it belongs, Substitute the Bayesian formula to calculate the posterior probability of rockburst:

[0192]

[0193] Where, represents the posterior probability of rockburst for new measured data;

[0194] Quality function construction:

[0195]

[0196] Where, Indicates measured data In interval j, the mass assignment supporting rock burst is: Indicates that non-rockburst mass assignment is supported. Indicates the quality assigned due to uncertainty in measured data, reducing the risk of misjudgment.

[0197] Calculate m1(E), m2(E), m3(E), m4(E), m1(Θ), m2(Θ), m3(Θ);

[0198] Use Dempster's combination rule to combine the quality functions:

[0199]

[0200] Preset the threshold T of rock burst probability high and T low , and T high >T low >0, continuous probability values ​​are difficult to directly guide engineering decisions, and threshold division facilitates the formulation of targeted measures. It can also integrate the uncertainty of groundwater pressure, rock temperature, sound wave velocity, and excavation footage measured data, quantify the degree of conflict and support through the quality function, and provide a reference framework for the risk assessment of other complex geological hazards, such as landslides and gas outbursts.

[0201] The new measured data m total (E) The output is the minimum confidence Bel(E), which reflects the lowest probability of rock burst. total (E) and m total (Θ) Calculate the maximum confidence:

[0202] Pl(E)=m total (E)+m total (Θ)

[0203] Where Pl(E) represents the maximum confidence level, reflecting the highest probability of rock burst occurrence;

[0204] Construct the formula for the probability of rock burst occurrence:

[0205]

[0206] Where p pred (E) represents the probability of rock burst, which is the average of the lowest probability of rock burst and the highest probability of rock burst, comprehensively considering the certainty and uncertainty of evidence to avoid the bias of a single indicator;

[0207] When P pred (E)≥T high , then the estimated probability of rock burst is high; when T low ≤P pted (E) <T high , then the estimated probability of rock burst is judged to be medium; when P pred (E) <T low , then the estimated probability of rock burst is judged to be low. According to the risk control and data verification method, T is set. high =0.7, T low =0.5, when T high =0.7, if at least P pred (E) ≥ 0.7, at least 2 measured data must fall into the high-risk range;

[0208] like Figure 10 As shown in the figure, the probability of rock burst occurrence P is calculated for 100 rock burst time windows. pred (E), P calculated for 76% of the time window pred (E) is higher than 0.7, and the probability of rock burst P calculated in the remaining time window pred (E) is generally between 0.5 and 0.7, which is consistent with actual rockburst prediction scenarios. If the misjudgment rate is high in moderate areas, the threshold can be slightly adjusted upward. Setting a higher threshold means that the risk is only determined to be high when the model has extremely high confidence in the occurrence of a rockburst. This can avoid frequent unnecessary shutdowns or protective measures due to oversensitivity, while ensuring timely response in truly high-risk situations. When the probability is between 50% and 70%, it indicates a certain degree of uncertainty, but close monitoring is still required to take preventive measures before the risk escalates. This setting balances the risks of misjudgment and missed detection, neither being overly conservative nor ignoring potential threats. The threshold can be adjusted according to actual conditions. By dividing the confidence interval and the threshold, this method transforms complex uncertainty reasoning into an intuitive risk level, providing a scientific basis for engineering decision-making.

[0209] See also Figure 11 The present invention further provides a device for predicting underground engineering rockburst dynamics based on machine learning, wherein the device is used to execute the above-mentioned method for predicting underground engineering rockburst dynamics based on machine learning, comprising:

[0210] A label construction module is used to collect measured data from historical data of underground projects and divide the measured data into intervals. The module collects the rock material of the underground project location, divides the burial depth of the underground project location into intervals, divides the rock hardness by uniaxial compressive strength, and constructs three-dimensional classification labels for material, burial depth, and hardness. The measured data includes groundwater pressure, rock temperature, sound wave velocity, and excavation footage.

[0211] The posterior probability module is used to count the number of times each measured data falls within each measured data interval under each three-dimensional classification label, construct a measured data set, and calculate the probability of rockburst data interval and the probability of non-rockburst data interval. The overall probability of the measured data interval is calculated based on the probability of rockburst data interval and non-rockburst data interval, and then the posterior probability of rockburst is calculated using the Bayesian formula.

[0212] Theoretical model construction module is used to construct the DS evidence theoretical model, map the rockburst posterior probability into a quality function in proportion, fuse the quality functions through the Dempster rule, and construct a comprehensive quality function;

[0213] The rockburst probability prediction module is used to dynamically match three-dimensional classification labels to underground engineering construction materials, burial depth intervals, and hardness intervals. After collecting new measured data, the measured data intervals are divided. After the new measured data is divided into measured data intervals, it is input into the DS evidence theory model and the minimum and maximum confidence levels are output. The rockburst probability is predicted based on the minimum and maximum confidence levels.

[0214] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0215] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software depends on the specific application and design constraints of the technical solution.

[0216] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0217] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A dynamic prediction method for underground engineering rockburst based on machine learning, characterized by: The specific steps include: Step 1: Collect measured data from historical underground engineering data and divide it into intervals. The rock material at the underground engineering location is collected, and the depth of the underground engineering location is divided into intervals. The rock hardness is classified by uniaxial compressive strength, and a three-dimensional classification label is constructed for material, depth, and hardness. The measured data includes groundwater pressure, rock temperature, acoustic wave velocity, and excavation footage. Step 2: Under each three-dimensional classification label, count the number of times each measured data falls within each measured data interval, construct a measured data set, and calculate the probability of the rockburst data interval and the probability of the non-rockburst data interval. Calculate the overall probability of the measured data interval based on the probability of the rockburst data interval and the non-rockburst data interval, and then use the Bayesian formula to calculate the posterior probability of rockburst. Step 3: Construct the DS evidence theory model, map the rockburst posterior probability into a quality function in proportion, fuse the quality functions through the Dempster rule, and construct a comprehensive quality function; Step 4: Dynamically match the three-dimensional classification labels for the underground engineering construction material, burial depth interval and hardness interval, collect new measured data and divide the measured data interval, input the new measured data into the measured data interval and output the minimum confidence and maximum confidence after entering the DS evidence theory model, and predict the rock burst probability based on the minimum confidence and maximum confidence.

2. The method for dynamic prediction of underground engineering rockburst based on machine learning according to claim 1, characterized in that: The method for collecting measured data and dividing the measured data interval is: Set the sampling time interval to T cj , in minutes, and T cj >0, according to the collection time interval T cj Collect four types of measured data: groundwater pressure, rock temperature, sound wave velocity and excavation footage, denoted as x i , where i = 1, 2, 3, 4, represents the index, representing groundwater pressure, rock temperature, acoustic wave velocity and excavation footage respectively; For each measured data x i , divided into several continuous intervals: Where a i，j represents the lower bound of the jth interval of the i-th type of measured data, a i，j+1 represents the upper bound of the jth interval of the i-th type of measured data, represents the jth interval of the i-th type of measured data, where j = 1, 2, ..., z i , z i Indicates the total number of intervals divided by the i-th category data.

3. The method for dynamic prediction of underground engineering rockburst based on machine learning according to claim 1, characterized in that: The method of dividing the depth interval of the underground project location and the rock hardness by uniaxial compressive strength is as follows: The burial depth range is divided according to the actual depth of the underground project: Where D ms represents the classification label of the buried depth interval, and D represents the actual depth of the underground project; The rock is tested through uniaxial compressive strength and the hardness range is divided into: Where U ud represents the hardness classification label, and UCS represents the uniaxial compressive strength.

4. The method for dynamic prediction of underground engineering rockburst based on machine learning according to claim 1, characterized in that: The method of counting the number of times each measured data falls within each measured data interval, constructing a measured data set, and calculating the probability of rockburst data interval and non-rockburst data interval is as follows: Preset the recording period to T zq , that is, the length of a single time window, in minutes, and T zq ≥30, set the sliding step length to 30 minutes; Calculate the number of data points collected during the logging period: Where N nub Indicates the number of data points collected during the recording period; Calculate the number of data point slides: Where S represents the number of data point sliding; Record the measured data of each time window. At this time, the number of overlapping data points in adjacent time windows is: O=N nub -S Where O represents the number of overlapping data points in adjacent time windows; According to the divided time windows, four types of measured data, namely groundwater pressure, rock temperature, acoustic wave velocity, and excavation footage, are recorded window by window. The number of time windows where rockburst occurs and does not occur follows a 1:4 ratio for statistics. The total number of accumulated data points is more than 1000, and a measured data set is generated. Based on the generated measured data set, the number of times the groundwater pressure, rock temperature, acoustic wave velocity, and excavation footage are in each measured data interval is counted. The number of times each measured data is in each interval when rockburst occurs, the total number of rockburst occurrences, the number of times each measured data is in each interval when rockburst does not occur, and the total number of times rockburst does not occur are counted. The probability of the rockburst data interval is calculated: Where, N represents the number of times the i-th measured data is in interval j when the rock burst occurs. E represents the total number of rock bursts, represents the probability of rock burst data interval, that is, the measured data x under the condition of rock burst i The probability in interval j, E represents the label of rockburst; Calculate the probability of non-rockburst data interval: Where, represents the number of times the i-th measured data is in interval j when rock burst does not occur, Indicates the total number of times rock burst did not occur, represents the probability of the non-rockburst data interval, that is, the measured data x under the condition that rockburst does not occur i The probability in interval j is, Label indicating non-rockburst.

5. The method for dynamic prediction of underground engineering rockburst based on machine learning according to claim 4, characterized in that: The method of calculating the overall probability of the measured data interval based on the probability of the rockburst data interval and the probability of the non-rockburst data interval, and then using the Bayesian formula to calculate the posterior probability of rockburst is as follows: First, extract the total number of rock bursts N E Total number of rock bursts that did not occur Calculate the prior probability of rockburst: Where P(E) represents the prior probability of rockburst; Using the total probability formula, calculate the measured data x i The overall probability in the jth interval is: Where, Represents data x i The overall probability in the jth interval, where The posterior probability of rockburst is calculated using the Bayesian formula: Where, represents the posterior probability of rockburst.

6. The method for dynamic prediction of underground engineering rockburst based on machine learning according to claim 5, characterized in that: The method of constructing the DS evidence theory model and mapping the rockburst posterior probability into a mass function in proportion is as follows: Extracting rockburst posterior probability Mapped to DS quality function according to the following formula: Construct the rockburst mass function: Where, Represents the rockburst mass function, that is, the measured data x i In interval j, the mass assignment of rock burst is supported, α represents the scaling factor, and the value is α = 0.

9. When i = 1, the rock burst mass function is marked as m1(E), when i = 2, the rock burst mass function is marked as m2(E), when i = 3, the rock burst mass function is marked as m3(E), and when i = 4, the rock burst mass function is marked as m4(E); Construct the non-rockburst mass function: Where, represents the non-rockburst mass function, that is, the mass assignment that supports non-rockburst. When i = 1, the non-rockburst mass function is marked as m1(E); when i = 2, the non-rockburst mass function is marked as m2(E); when i = 3, the non-rockburst mass function is marked as m3(E); when i = 4, the non-rockburst mass function is marked as m4(E); Construct the uncertainty quality function: Where, represents the uncertainty mass function, i.e., the mass assigned to the measured data due to uncertainty, where When i=1, the uncertainty mass function is labeled as m1(Θ), when i=2, the uncertainty mass function is labeled as m2(Θ), when i=3, the uncertainty mass function is labeled as m3(Θ), and when i=4, the uncertainty mass function is labeled as m4(Θ).

7. The method for dynamic prediction of underground engineering rockburst based on machine learning according to claim 6, characterized in that: The method of constructing a comprehensive quality function by fusing quality functions through Dempster rule is as follows: The mass functions are combined using the Dempster combination rule. First, the mass functions of groundwater pressure and rock temperature are combined to calculate the intersection mass of groundwater pressure and rock temperature: Where k 12 represents the conflict coefficient between groundwater pressure and rock temperature, Then, the mass function of the acoustic wave velocity is integrated to calculate the intersection mass of groundwater pressure, rock temperature, and acoustic wave velocity: Where k 123 represents the conflict coefficient of groundwater pressure, rock temperature and sound wave velocity, Then, the mass function of the excavation footage is integrated to calculate the intersection mass of groundwater pressure, rock temperature, acoustic wave velocity, and excavation footage: Where m total represents the comprehensive quality function, k total It represents the conflict coefficient of groundwater pressure, rock temperature, sound wave velocity and excavation footage.

8. The method for dynamic prediction of underground engineering rockburst based on machine learning according to claim 1, characterized in that: The method of dividing the measured data into intervals and inputting them into the DS evidence theory model to output the minimum confidence level and the maximum confidence level is as follows: According to the real-time rock material, burial depth and uniaxial compressive strength, the new measured data is matched with the three-dimensional classification label. Divide the intervals and input them into the DS evidence theory model to extract the rockburst data interval probability of the new measured data Non-rockburst data interval probability and the prior probability of rockburst P(E), where Represents the i-th category of real-time data The interval index to which it belongs, Substitute the Bayesian formula to calculate the posterior probability of rockburst: Where, represents the posterior probability of rockburst for new measured data; Quality function construction: Where, Indicates measured data In interval j, the mass assignment supporting rock burst is: Indicates that non-rockburst mass assignment is supported. It represents the quality assigned to the measured data due to uncertainty; Calculate m1(E), m2(E), m3(E), m4(E), m1(Θ), m2(Θ), m3(Θ); Use Dempster's combination rule to combine the quality functions:

9. A dynamic prediction device for underground engineering rockburst based on machine learning, characterized by: The method for predicting the rockburst probability based on the minimum confidence level and the maximum confidence level is: Preset the threshold T of rock burst probability high and T low , and T high >T low >0, the new measured data m total (E) Output is the minimum confidence Bel(E), extracting the m of the new measured data total (E) and m total (Θ) Calculate the maximum confidence: Pl(E)=m total (E)+m total (Θ) Where Pl(E) represents the maximum confidence level; Construct the formula for the probability of rock burst occurrence: Where, P pred (E) represents the probability of rock burst; When P pred (E)≥T high , then the probability of rock burst is judged to be high; when T low ≤P pred (E) <T high , then the probability of rock burst is judged to be medium; when P pred (E) <T low , then the probability of rock burst is judged to be low.

10. A dynamic prediction device for underground engineering rockburst based on machine learning, characterized by: The device is used in the underground engineering rockburst dynamic prediction method based on machine learning according to any one of claims 1 to 9: A label construction module is used to collect measured data from historical data of underground projects and divide the measured data into intervals. The module collects the rock material of the underground project location, divides the burial depth of the underground project location into intervals, divides the rock hardness by uniaxial compressive strength, and constructs three-dimensional classification labels for material, burial depth, and hardness. The measured data includes groundwater pressure, rock temperature, sound wave velocity, and excavation footage. The posterior probability module is used to count the number of times each measured data falls within each measured data interval under each three-dimensional classification label, construct a measured data set, and calculate the probability of rockburst data interval and the probability of non-rockburst data interval. The overall probability of the measured data interval is calculated based on the probability of rockburst data interval and non-rockburst data interval, and then the posterior probability of rockburst is calculated using the Bayesian formula. Theoretical model construction module is used to construct the DS evidence theoretical model, map the rockburst posterior probability into a quality function in proportion, fuse the quality functions through the Dempster rule, and construct a comprehensive quality function; The rockburst probability prediction module is used to dynamically match three-dimensional classification labels to underground engineering construction materials, burial depth intervals, and hardness intervals. After collecting new measured data, the measured data intervals are divided. After the new measured data is divided into measured data intervals, it is input into the DS evidence theory model and the minimum and maximum confidence levels are output. The rockburst probability is predicted based on the minimum and maximum confidence levels.

Images