Rock burst risk early warning method based on TBM multi-source data fusion and hybrid algorithm

Abstract

The invention discloses a rockburst risk early warning method based on TBM multi-source data fusion and a hybrid algorithm, and the method comprises the steps: collecting TBM tunneling parameters, geological exploration data and micro-seismic monitoring data of different rockburst levels, achieving feature complementation through multi-source data fusion, improving the completeness and reliability of early warning, constructing a hybrid neural network model based on CNN, LSTM and an attention mechanism, and carrying out the early warning of the rockburst risk. Feature weight distribution is optimized by introducing an attention mechanism, so that the model can adaptively focus key risk signals, and the applicability of model early warning is improved; and in combination with a fuzzy comprehensive evaluation algorithm and a Bayesian probability model, outputting four rockburst grades of no rockburst, slight rockburst, medium rockburst and strong rockburst and occurrence probabilities thereof, and realizing real-time dynamic early warning and probabilistic early warning of the rockburst grades. Compared with an existing method, the method has the advantages that the accuracy, timeliness and engineering applicability of rockburst early warning are remarkably improved, and real-time and accurate early warning of potential rockburst and the grade of the potential rockburst can be achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel engineering safety monitoring and geological disaster early warning, and in particular to a rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm. Background Art

[0002] Rockburst is a dynamic phenomenon in which the elastic deformation energy accumulated in deep or highly tectonically stressed rock masses is suddenly released under excavation or other external disturbances, resulting in the explosion and ejection of surrounding rock. Rockbursts can cause casualties, economic losses, and construction delays, and can even trigger artificial earthquakes, bringing immeasurable disasters to human society. As full-face tunnel boring machines (TBMs) become the dominant construction method, rockburst hazards are unavoidable during construction. Therefore, how to quickly and effectively implement rockburst early warning systems for TBM tunnels has long been a pressing concern.

[0003] Currently, rockburst disasters encountered during TBM tunneling are mainly evaluated based on laboratory testing, numerical calculations, and field monitoring. In laboratory testing, traditional rockburst assessment methods, such as those based on the Russenes criterion and the Hoek criterion, rely primarily on static parameters such as the uniaxial compressive strength and ground stress of rock, and are unable to reflect the dynamic changes during the tunneling process. Rockburst prediction and early warning methods based on numerical calculations are unable to monitor and warn of rockburst risks in real time and have significant lags. Field monitoring often uses technologies such as microseismic monitoring and acoustic emission, but these technologies suffer from issues such as event recognition lags and high false alarm rates. In addition, machine learning methods for warning of rockburst risks, such as the BP neural network method, are limited by single data source input and fixed time window analysis, resulting in low accuracy in rockburst risk assessment.

[0004] Therefore, how to achieve real-time and accurate early warning assessment of rockburst risks is an urgent problem that technical personnel in this field need to solve. Summary of the Invention

[0005] In view of this, the present invention provides a rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm, which is used to achieve real-time and accurate early warning assessment of rockburst risk.

[0006] Therefore, the present invention provides a rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm, comprising the following steps:

[0007] S1: Collect TBM excavation parameters, geological exploration data, and microseismic monitoring data for multiple rockburst cases. Rockburst cases are classified into four levels: no rockburst, slight rockburst, moderate rockburst, and severe rockburst.

[0008] S2: The TBM excavation parameters, geological exploration data, and microseismic monitoring data corresponding to each rockburst case are constructed into a spatiotemporal correlation feature matrix. The rockburst levels and spatiotemporal correlation feature matrices corresponding to all rockburst cases constitute a sample library.

[0009] S3: Divide the sample library into a training set and a test set;

[0010] S4: constructing a hybrid neural network model including a convolutional neural network, a long short-term memory network, and an attention mechanism, training the hybrid neural network model using the training set, and verifying the trained hybrid neural network model using the test set;

[0011] S5: TBM excavation parameters, geological exploration data and microseismic monitoring data are collected in real time for the tunnels to be detected for early warning. The real-time collected data are input into the trained and verified hybrid neural network model. The rockburst grade membership is calculated using the fuzzy comprehensive evaluation algorithm. The probability of occurrence of each rockburst grade is calculated using the Bayesian probability model. Real-time rockburst risk warning is carried out for the tunnels to be detected for early warning.

[0012] In one possible implementation, in the rockburst risk warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention, in step S1, the collected TBM excavation parameters specifically include cutterhead torque, thrust pressure, penetration and rotation speed, with a sampling frequency of not less than 1 Hz, and the data accuracy requirements are: the measurement error of the cutterhead torque ≤ ±2, the resolution of the thrust pressure ≥ 0.1 MPa, and the calculation period of the penetration ≤ 0.5 s.

[0013] In one possible implementation, in the rockburst risk warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention, in step S1, the geological exploration data collected specifically include rock uniaxial compressive strength, rock integrity coefficient, ground stress component, rock wave velocity and basic quality indicators of surrounding rock.

[0014] In a possible implementation, in the rockburst risk warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention, in step S1, the collected microseismic monitoring data specifically includes the number of microseismic events, microseismic release energy, microseismic apparent volume and source coordinates.

[0015] In one possible implementation, in the rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention, step S1 collects TBM excavation parameters, geological exploration data, and microseismic monitoring data for multiple rockburst cases that have occurred, specifically including:

[0016] If the tunnel to be detected for early warning belongs to a newly built underground project, an existing project with the same excavation method and similar geological conditions as the tunnel to be detected for early warning is selected, and the TBM excavation parameters, geological exploration data and microseismic monitoring data corresponding to multiple rock burst cases that occurred during the excavation process of the existing project are collected; if the tunnel to be detected for early warning belongs to a project under construction, the TBM excavation parameters, geological exploration data and microseismic monitoring data corresponding to multiple rock burst cases that occurred in the excavated area of the tunnel to be detected for early warning are collected.

[0017] In one possible implementation, in the rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention, in step S2, the TBM excavation parameters, geological exploration data, and microseismic monitoring data corresponding to each rockburst case are constructed into a spatiotemporal correlation feature matrix. The method for constructing the spatiotemporal correlation feature matrix specifically includes:

[0018]

[0019] n=ceil(1.5D / v) (2)

[0020] m=floor(L / Δ) (3)

[0021] Among them, S t represents the spatiotemporal correlation feature matrix, F TBM represents the TBM excavation parameter matrix, G geo Represents the geological exploration data matrix, M micro represents the microseismic monitoring data matrix, represents the spatiotemporal convolution operator, T n×m represents the time window and spatial dimension matrix, n represents the time window size, m represents the spatial correlation dimension; D represents the TBM cutterhead diameter, in meters; v represents the TBM advance speed, in meters per minute; L represents the monitoring range in the tunnel axis direction, in meters; Δ represents the spatial resolution; the ceil function represents the rounding up of the value, and the floor function represents the rounding down of the value.

[0022] In one possible implementation, in the rockburst risk warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention, in step S4, a hybrid neural network model including a convolutional neural network, a long short-term memory network and an attention mechanism is constructed, and the network structure of the hybrid neural network model includes, in sequence, 1 input layer, 1 convolutional layer, 1 LSTM parameter layer, 1 attention weight layer, 2 fully connected layers and 1 Softmax layer.

[0023] In one possible implementation, in the rockburst risk warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention, after executing step S5, TBM excavation parameters, geological exploration data, and microseismic monitoring data are collected in real time for the tunnel to be detected for warning. The real-time collected data is input into a trained and verified hybrid neural network model, and the rockburst grade membership is calculated using a fuzzy comprehensive evaluation algorithm. The occurrence probability of each rockburst grade is calculated using a Bayesian probability model. After providing a real-time rockburst risk warning for the tunnel to be detected for warning, the following steps are further included:

[0024] S6: Check on-site whether the warning results are consistent with the actual situation; if so, construct the TBM excavation parameters, geological exploration data, and microseismic monitoring data corresponding to the rockburst case into a spatiotemporal correlation feature matrix, and add the rockburst level and spatiotemporal correlation feature matrix corresponding to the rockburst case to the original sample library; if not, discard the rockburst case and do not include it in the sample library.

[0025] The rockburst risk warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention collects TBM excavation parameters, geological exploration data and microseismic monitoring data of different rockburst levels, and realizes cross-dimensional and cross-temporal and spatial feature complementarity through multi-source data fusion, which can significantly improve the completeness and reliability of rockburst warning. A hybrid neural network model based on convolutional neural network, long short-term memory network and attention mechanism is constructed. By introducing the attention mechanism, the feature weight distribution can be optimized, so that the model can adaptively focus on key risk signals and improve the applicability of model warning. Further combined with the fuzzy comprehensive evaluation algorithm and the Bayesian probability model, a four-dimensional probabilistic warning system for rockburst risk is established, which outputs four rockburst levels and their occurrence probabilities of no rockburst, slight rockburst, moderate rockburst and severe rockburst, thereby realizing real-time dynamic warning and probabilistic warning of rockburst risk levels. Compared with existing rockburst warning methods, the present invention has significantly improved the accuracy, timeliness and engineering applicability of rockburst warning. It is particularly suitable for underground projects such as deep tunnels and mine tunnels. It can achieve real-time and accurate warning of potential rockbursts and their levels, thereby providing a scientific basis for the subsequent formulation of targeted rockburst prevention and control measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A schematic flow chart of a rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention;

[0027] Figure 2 This is a structural schematic diagram of the hybrid neural network model in the rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention. DETAILED DESCRIPTION

[0028] The specific implementation of the rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention is described in detail below with reference to the accompanying drawings.

[0029] The rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention is as follows: Figure 1 As shown, the following steps may be included:

[0030] Step 1: Collect TBM excavation parameters, geological exploration data, and microseismic monitoring data for multiple rockburst cases. Rockburst cases can be categorized into four levels: no rockburst, minor rockburst, moderate rockburst, and severe rockburst.

[0031] In specific implementation, the tunnel to be detected by early warning can be a newly built underground project, that is, a project that has not yet been excavated or has been excavated for a short time (no rock burst cases have occurred or the number of rock burst cases has been insufficient); or, the tunnel to be detected by early warning can also be a project under construction, that is, a project that has been excavated and has accumulated a sufficient number of rock burst cases, which is not limited here. When the tunnel to be detected by early warning belongs to a newly built underground project, that is, the tunnel to be detected by early warning has not yet had a rock burst case or the number of rock burst cases has been insufficient, due to the lack of rock burst cases and corresponding microseismic monitoring information, it is possible to select an existing project with the same excavation method (such as TBM) and similar geological conditions (for example, consistent lithology, similar ground stress, etc.) as the tunnel to be detected by early warning, and use multiple rock burst cases of known levels that occurred during the excavation process and their corresponding TBM excavation parameters, geological exploration data and microseismic monitoring data to establish a sample library. When the tunnel to be detected for early warning belongs to a project under construction, that is, the tunnel to be detected for early warning has been excavated and a sufficient number of rockburst cases have accumulated. Since a sufficient number of rockburst cases have been accumulated in the early stage of the tunnel to be detected for early warning, multiple rockburst cases with known rockburst levels that occurred in the excavated area of the tunnel to be detected for early warning and their corresponding TBM excavation parameters, geological exploration data, and microseismic monitoring data can be directly collected to establish a sample library.

[0032] In specific implementations, the collected TBM excavation parameters may include cutterhead torque (unit: kN·m), thrust pressure (unit: MPa), penetration (unit: mm / r), and rotational speed (i.e., the time required for the cutterhead to make one revolution, unit: rpm). Preferably, the sampling frequency of TBM excavation parameters may be no less than 1Hz, and the data accuracy requirements may be: cutterhead torque measurement error ≤ ±2, thrust pressure resolution ≥ 0.1MPa, and penetration calculation period ≤ 0.5s. Of course, the data types, sampling frequency, and accuracy requirements for TBM excavation parameters are not limited to these and can be optimized and adjusted based on actual conditions, and are not limited here.

[0033] During the specific implementation, the collected geological exploration data may include the uniaxial compressive strength of rock (unit: MPa), rock mass integrity coefficient, ground stress component (σ x ,σ y ,σ z ), rock mass wave velocity (unit: m / s) and basic quality indicators of surrounding rock. Of course, the types of geological exploration data are not limited to these and can be optimized and adjusted according to actual conditions, so they are not limited here.

[0034] In the specific implementation, the collected microseismic monitoring data may include the number of microseismic events (unit: number), microseismic release energy (unit: J), microseismic apparent volume (unit: m 3 ) and earthquake source coordinates (x, y, z). Of course, the types of microseismic monitoring data are not limited to these and can be optimized and adjusted according to actual conditions, so they are not limited here.

[0035] Step 2: The TBM excavation parameters, geological exploration data, and microseismic monitoring data corresponding to each rockburst case are constructed into a spatiotemporal correlation feature matrix. The rockburst levels and spatiotemporal correlation feature matrices corresponding to all rockburst cases constitute a sample library.

[0036] The method for constructing the spatiotemporal correlation feature matrix may specifically include:

[0037]

[0038] n=ceil(1.5D / v) (2)

[0039] m=floor(L / Δ) (3)

[0040] Among them, S t represents the spatiotemporal correlation feature matrix, F TBM represents the TBM excavation parameter matrix, G geo Represents the geological exploration data matrix, M micro represents the microseismic monitoring data matrix, represents the spatiotemporal convolution operator, T n×m represents the time window and spatial dimension matrix, n represents the time window size, m represents the spatial correlation dimension; D represents the TBM cutterhead diameter, in meters; v represents the TBM advance speed, in meters per minute; L represents the monitoring range in the tunnel axis direction, in meters; Δ represents the spatial resolution, for example, Δ can be between 2 meters and 5 meters; the ceil function represents the rounding up of the value, and the floor function represents the rounding down of the value.

[0041] Step 3: Divide the sample library into training set and test set.

[0042] In specific implementations, most samples are generally selected as the training set for training the hybrid neural network model constructed later, and the remaining samples are used as the test set to verify the trained hybrid neural network model. For example, 80% of the samples can be randomly selected from the sample library as training samples, and the remaining 20% of the samples can be used as test samples.

[0043] Step 4: Build a hybrid neural network model that includes convolutional neural network (CNN), long short-term memory network (LSTM) and attention mechanism (Attention), train the hybrid neural network model using the training set, and verify the trained hybrid neural network model using the test set.

[0044] In specific implementation, the network structure of the hybrid neural network model, such as Figure 2 As shown in the figure, it can include: 1 input layer, 1 convolutional layer, 1 LSTM parameter layer, 1 attention weight layer, 2 fully connected layers, and 1 Softmax layer. The input layer inputs the training set data, and the Softmax layer serves as the output layer to output the rockburst grade and occurrence probability.

[0045] In the rockburst risk warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention, the constructed hybrid neural network model is trained and optimized using a training set, and the model parameters with the highest rockburst grade classification accuracy are obtained based on the classification results of the test set.

[0046] Step 5: Real-time collection of TBM excavation parameters, geological exploration data, and microseismic monitoring data is performed for the tunnels undergoing early warning detection. The real-time collected data is input into a trained and verified hybrid neural network model. The rockburst grade membership is calculated using a fuzzy comprehensive evaluation algorithm, and the probability of occurrence of each rockburst grade is calculated using a Bayesian probability model. Real-time rockburst risk warnings are then provided for the tunnels undergoing early warning detection.

[0047] Preferably, in order to further optimize the training of the hybrid neural network model, in the rockburst risk warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention, after executing the fifth step, that is, using the trained hybrid neural network model to perform real-time rockburst risk warning on the tunnel to be detected, if the on-site inspection warning result is consistent with the actual situation, the TBM excavation parameters, geological exploration data and microseismic monitoring data corresponding to the rockburst case can be added to the sample library. In this way, the training set can be dynamically supplemented and updated, so that the hybrid neural network model can be continuously optimized.

[0048] Step 6: On-site verification to determine whether the warning results are consistent with the actual situation; if so, construct a spatiotemporal correlation feature matrix using the TBM excavation parameters, geological exploration data, and microseismic monitoring data corresponding to the rockburst case, and add the rockburst level and spatiotemporal correlation feature matrix corresponding to the rockburst case to the original sample library; if not, discard the rockburst case and exclude it from the sample library.

[0049] The following describes in detail the implementation of the rockburst risk warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention through a specific embodiment.

[0050] Example 1:

[0051] Taking a deep TBM tunnel as an example, during the initial excavation process, 30 cases of no rockburst, 25 cases of minor rockburst, 20 cases of moderate rockburst, and 3 cases of severe rockburst (a total of 78 rockburst cases) occurred. A sample library was established using these 78 rockburst cases and their corresponding TBM excavation parameters, geological exploration data, and microseismic monitoring data. A sample library was randomly selected from these 78 rockburst cases (24 cases of no rockburst, 20 cases of minor rockburst, 16 cases of moderate rockburst, and 2 cases of severe rockburst, a total of 62 rockburst cases). A spatiotemporal correlation feature matrix was constructed from these 62 rockburst cases and their corresponding TBM excavation parameters, geological exploration data, and microseismic monitoring data to form the training set. A spatiotemporal correlation feature matrix was constructed from the remaining 16 rockburst cases and their corresponding TBM excavation parameters, geological exploration data, and microseismic monitoring data to form the test set.

[0052] The tunnel is 18.6 km long and has a burial depth of approximately 680–720 m. The lithology is mainly granite with a uniaxial compressive strength of 175 MPa. The tunnel diameter is 5 m. The average TBM advancement speed is 0.83 m / min. The tunnel axis monitoring range is 15 m, and the spatial resolution is 3 m. From formulas (2) and (3), it can be calculated that the time window n of the tunnel is 6 and the spatial correlation dimension m is 5.

[0053] The training set was fed into a hybrid neural network model, which then performed deep learning on TBM excavation parameters, geological exploration data, and microseismic monitoring data during the development of different rockburst levels. After training, the rockburst class membership was calculated using a fuzzy comprehensive evaluation algorithm, and the probability of occurrence of each rockburst class was calculated using a Bayesian probability model. The rockburst class was then output based on a comprehensive assessment of the rockburst class membership and probability. This model was then used to test the test set. Model parameters were adjusted based on the test results until the model with the highest warning accuracy was obtained. The network structure of the hybrid neural network model finally obtained can include: 1) input layer, which inputs TBM excavation parameters, geological exploration data and microseismic monitoring data; 2) a convolution layer, which inputs a 6×5×3 tensor, a 3×3 convolution kernel, and a convolution kernel movement step size automatically determined by the same-padding strategy, using ReLU as the activation function; 3) an LSTM parameter layer with 128 hidden units, the forget gate bias initialized to 1.0, and iterative calculation along 6 time steps; 4) an attention weight layer; 5) the first fully connected layer, including 12 8 neuron nodes, weight initialization uses He normal distribution; 6) The second fully connected layer contains 64 neuron nodes, and adopts the Dropout strategy with a probability of 0.2, and uses ReLU as the activation function; 7) The Softmax layer, the category weights are distributed in inverse proportion to the number of samples, contains 4 nodes, corresponding to the number of rockburst grades; 8) The rockburst grade membership is calculated through the fuzzy comprehensive evaluation algorithm, and the occurrence probability of each rockburst grade is calculated through the Bayesian probability model. The rockburst grade membership and occurrence probability are used to comprehensively judge the rockburst risk warning and output it.

[0054] The test set was tested using the hybrid neural network model, and the results are shown in Table 1. As can be seen from Table 1, the test accuracy of the hybrid neural network model reached 87.5%. During the subsequent excavation of the tunnel, the model was used for rock burst warning, and a total of 133 rock burst warnings were carried out. The comparison between the warning results and the actual situation on site is shown in Table 2. As can be seen from Table 2, the accuracy of the potential rock burst warning of Example 1 of the present invention reached 84.9%. This application shows that the rock burst risk warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention has high practicality and accuracy, and can achieve real-time and accurate warning of potential rock bursts and their levels, thereby providing a scientific basis for the subsequent formulation of targeted rock burst prevention and control measures.

[0055] Table 1 Test results of hybrid neural network model

[0056]

[0057] Table 2 Comparison of potential rockburst warning results of a TBM tunnel project based on the present invention and actual situation on site

[0058]

[0059] The rockburst risk warning method based on TBM multi-source data fusion and hybrid algorithm provided by the present invention collects TBM excavation parameters, geological exploration data and microseismic monitoring data of different rockburst levels, and realizes cross-dimensional and cross-temporal and spatial feature complementarity through multi-source data fusion, which can significantly improve the completeness and reliability of rockburst warning. A hybrid neural network model based on convolutional neural network, long short-term memory network and attention mechanism is constructed. By introducing the attention mechanism, the feature weight distribution can be optimized, so that the model can adaptively focus on key risk signals and improve the applicability of model warning. Further combined with the fuzzy comprehensive evaluation algorithm and the Bayesian probability model, a four-dimensional probabilistic warning system for rockburst risk is established, which outputs four rockburst levels and their occurrence probabilities of no rockburst, slight rockburst, moderate rockburst and severe rockburst, thereby realizing real-time dynamic warning and probabilistic warning of rockburst risk levels. Compared with existing rockburst warning methods, the present invention has significantly improved the accuracy, timeliness and engineering applicability of rockburst warning. It is particularly suitable for underground projects such as deep tunnels and mine tunnels. It can achieve real-time and accurate warning of potential rockbursts and their levels, thereby providing a scientific basis for the subsequent formulation of targeted rockburst prevention and control measures.

[0060] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm is characterized by: The steps include: S1: Collect TBM excavation parameters, geological exploration data, and microseismic monitoring data for multiple rockburst cases. Rockburst cases are classified into four levels: no rockburst, slight rockburst, moderate rockburst, and severe rockburst. S2: The TBM excavation parameters, geological exploration data, and microseismic monitoring data corresponding to each rockburst case are constructed into a spatiotemporal correlation feature matrix. The rockburst levels and spatiotemporal correlation feature matrices corresponding to all rockburst cases constitute a sample library. S3: Divide the sample library into a training set and a test set; S4: constructing a hybrid neural network model including a convolutional neural network, a long short-term memory network, and an attention mechanism, training the hybrid neural network model using the training set, and verifying the trained hybrid neural network model using the test set; S5: TBM excavation parameters, geological exploration data and microseismic monitoring data are collected in real time for the tunnels to be detected for early warning. The real-time collected data are input into the trained and verified hybrid neural network model. The rockburst grade membership is calculated using the fuzzy comprehensive evaluation algorithm. The probability of occurrence of each rockburst grade is calculated using the Bayesian probability model. Real-time rockburst risk warning is carried out for the tunnels to be detected for early warning.

2. The rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm according to claim 1 is characterized in that: In step S1, the collected TBM excavation parameters include cutterhead torque, thrust pressure, penetration, and rotational speed. The sampling frequency is not less than 1 Hz, and the data accuracy requirements are: the measurement error of the cutterhead torque is ≤ ±2, the resolution of the thrust pressure is ≥ 0.1 MPa, and the calculation period of the penetration is ≤ 0.5 s.

3. The rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm according to claim 1 is characterized in that: In step S1, the geological exploration data collected specifically include rock uniaxial compressive strength, rock mass integrity coefficient, ground stress component, rock mass wave velocity and basic quality indicators of surrounding rock.

4. The rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm according to claim 1 is characterized in that: In step S1, the collected microseismic monitoring data specifically include the number of microseismic events, microseismic release energy, microseismic apparent volume and source coordinates.

5. The rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm according to any one of claims 1 to 4, characterized in that: Step S1: Collect TBM excavation parameters, geological exploration data, and microseismic monitoring data for multiple rockburst cases, including: If the tunnel to be detected for early warning belongs to a newly built underground project, an existing project with the same excavation method and similar geological conditions as the tunnel to be detected for early warning is selected, and the TBM excavation parameters, geological exploration data and microseismic monitoring data corresponding to multiple rock burst cases that occurred during the excavation process of the existing project are collected; if the tunnel to be detected for early warning belongs to a project under construction, the TBM excavation parameters, geological exploration data and microseismic monitoring data corresponding to multiple rock burst cases that occurred in the excavated area of the tunnel to be detected for early warning are collected.

6. The rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm according to claim 1, characterized in that: In step S2, the TBM excavation parameters, geological exploration data, and microseismic monitoring data corresponding to each rockburst case are constructed into a spatiotemporal correlation feature matrix. The method for constructing the spatiotemporal correlation feature matrix specifically includes: n=ceil(1.5D / v) (2) m=floor(L / Δ) (3) Among them, S t represents the spatiotemporal correlation feature matrix, F TBM represents the TBM excavation parameter matrix, G geo Represents the geological exploration data matrix, M micro represents the microseismic monitoring data matrix, represents the spatiotemporal convolution operator, T n×m represents the time window and spatial dimension matrix, n represents the time window size, m represents the spatial correlation dimension; D represents the TBM cutterhead diameter, in meters; v represents the TBM advance speed, in meters per minute; L represents the monitoring range in the tunnel axis direction, in meters; Δ represents the spatial resolution; the ceil function represents the rounding up of the value, and the floor function represents the rounding down of the value.

7. The rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm according to claim 1, characterized in that: In step S4, a hybrid neural network model including a convolutional neural network, a long short-term memory network and an attention mechanism is constructed. The network structure of the hybrid neural network model includes, in sequence, 1 input layer, 1 convolutional layer, 1 LSTM parameter layer, 1 attention weight layer, 2 fully connected layers and 1 Softmax layer.

8. The rockburst risk early warning method based on TBM multi-source data fusion and hybrid algorithm according to claim 1, characterized in that: In step S5, TBM excavation parameters, geological exploration data, and microseismic monitoring data are collected in real time for the tunnel to be detected for early warning. The real-time collected data are input into a trained and verified hybrid neural network model. The rockburst grade membership is calculated using a fuzzy comprehensive evaluation algorithm, and the occurrence probability of each rockburst grade is calculated using a Bayesian probability model. After providing a real-time rockburst risk warning for the tunnel to be detected for early warning, the following steps are also included: S6: Check on-site whether the warning results are consistent with the actual situation; if so, construct the TBM excavation parameters, geological exploration data, and microseismic monitoring data corresponding to the rockburst case into a spatiotemporal correlation feature matrix, and add the rockburst level and spatiotemporal correlation feature matrix corresponding to the rockburst case to the original sample library; if not, discard the rockburst case and do not include it in the sample library.

Images