Geological risk-torque prediction driven jamming risk evaluation method and system

Abstract

The present application belongs to the technical field of tunnel, in order to solve the problem that the existing monitoring method is difficult to realize the collaborative perception of geological inducement and equipment response in the process of tunneling, resulting in inaccurate risk prediction and lagging early warning, etc., a geological risk-torque prediction driven jam risk evaluation method and system are proposed, for geological risk evaluation, a standard cloud model is constructed to realize, a dynamic adjustment mechanism based on real-time data fluctuation and risk sensitivity is introduced, so that the cloud model can not only more accurately respond to complex and dynamic changing geological conditions, but also maintain efficient and accurate risk assessment ability in geological mutation; The prediction method driven by geological risk prediction and torque prediction realizes the collaborative perception of geological inducement and equipment response, comprehensively and dynamically masters the cause and evolution process of jam risk, and ensures the prediction accuracy.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field related to tunnel excavation, and particularly relates to a geological risk-torque prediction driven jam risk evaluation method and system. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute prior art.

[0003] TBM, i.e., tunnel boring machine, is widely used in the construction of long tunnels and large tunnels in highway, railway and water conservancy engineering, etc., and has the advantages of high efficiency, safety and environmental protection. However, with the expansion of construction scale and the increase of complexity of geological conditions, the TBM often faces the risk of jamming during the tunneling process, especially in complex strata and fault areas.

[0004] In the process of evaluating the jamming risk of the tunneling machine, the existing technology generally has the problems of single model and delayed response. On the one hand, the risk evaluation model based on geological parameters usually relies on static geological exploration data, and it is difficult to reflect the dynamic risk factors such as stratum mutation or evolution of soft surrounding rock in the tunneling process in a timely manner. Moreover, there is a lack of quantitative modeling of uncertainty and volatility between geological data, resulting in rough risk level division and insufficient reliability of prediction results. On the other hand, although the monitoring method based on the main control parameters such as cutter torque can realize real-time perception of the state of the equipment, its essence is still "result driven", and usually only after the abnormal fluctuation of the equipment parameters, the warning is carried out, and it is difficult to identify the jamming risk caused by the latent geological causes in advance, and the warning is obviously delayed. In addition, the current prevention and control strategy relies on the experience of on-site personnel for parameter adjustment and emergency response, and lacks a systematic and standardized intelligent decision support platform, resulting in delay in risk identification and response measures, and easy to induce serious equipment damage and engineering delay problems.

[0005] In summary, the existing geological parameter risk evaluation usually relies on static geological exploration data and empirical models, and has the problems of insufficient prediction accuracy, poor real-time performance and poor response to unexpected situations. The existing monitoring method in the tunneling process is difficult to realize the coordinated perception of geological causes and equipment response, and cannot comprehensively and dynamically grasp the causes and evolution process of the jamming risk, resulting in the problems of being difficult to identify the jamming risk caused by the latent geological causes in advance and causing the warning to be delayed. SUMMARY

[0006] In order to overcome the shortcomings of the prior art, the present application provides a geological risk-torque prediction driven jam risk evaluation method and system, which adopts a prediction method driven by geological risk prediction and torque prediction, realizes the coordinated perception of geological causes and equipment response, comprehensively and dynamically grasps the causes and evolution process of the jamming risk, and ensures the prediction accuracy.

[0007] In order to achieve the above object, the present application adopts the following technical solutions:

[0008] In a first aspect, the present application provides a geological risk-torque prediction driven jamming risk evaluation method, comprising:

[0009] obtaining geological parameters, geophysical parameters and main control parameters of a tunneling machine in a tunneling section;

[0010] According to the adverse geology and its corresponding geological parameters and geophysical parameters, the initial weight of the risk indicators of the geological parameters and the geophysical parameters is given, and the geology is divided into multiple risk levels to construct a standard cloud model; based on the standard cloud model, the weight of the geological parameters and / or the geophysical parameters is dynamically adjusted according to the fluctuation and risk sensitivity of the geological parameters and / or the geophysical parameters in the tunneling process, and the prediction results of all risk indicators are weighted and fused to obtain the geological risk level of the tunneling section;

[0011] Based on the geological parameters, the geophysical parameters and the main control parameters of the tunneling machine, a trained torque prediction model is used to obtain the running risk level of the tunneling machine;

[0012] Based on the jamming risk prediction model constructed by the gradient boosting tree, the geological risk level and the running risk level of the tunneling machine are fused to obtain the final jamming risk evaluation result.

[0013] In a second aspect, the present application provides a tunneling machine jamming risk evaluation system, comprising:

[0014] The data collection module is configured to obtain geological parameters, geophysical parameters and main control parameters of a tunneling machine in a tunneling section;

[0015] The geological risk evaluation module is configured to: according to the adverse geology and its corresponding geological parameters and geophysical parameters, the initial weight of the risk indicators of the geological parameters and the geophysical parameters is given, and the geology is divided into multiple risk levels to construct a standard cloud model; based on the standard cloud model, the weight of the geological parameters and / or the geophysical parameters is dynamically adjusted according to the fluctuation and risk sensitivity of the geological parameters and / or the geophysical parameters in the tunneling process, and the prediction results of all risk indicators are weighted and fused to obtain the geological risk level of the tunneling section;

[0016] The torque prediction module is configured to: based on the geological parameters, the geophysical parameters and the main control parameters of the tunneling machine, a trained torque prediction model is used to obtain the running risk level of the tunneling machine;

[0017] The double-driving evaluation module is configured to fuse the geological risk grade and the tunneling machine operation risk grade based on a machine jamming risk prediction model constructed by the gradient boosting tree to obtain a machine jamming risk evaluation result.

[0018] The above one or more technical solutions have the following beneficial effects:

[0019] In the present application, the geological risk evaluation is realized by constructing a comprehensive cloud model. The cloud model has strong advantages in describing the uncertainty, fuzziness and evolution trend among multi-source data, has the ability to fuse qualitative cognition and quantitative analysis, can accurately depict the randomness and fuzzy boundary of risk distribution in a complex geological environment with strong parameter volatility and insufficient information integrity, and can improve the stability and adaptability of risk grade judgment. The double-driving prediction method of geological risk prediction and torque prediction is used, gradient boosting tree algorithm is used for fusion analysis, and the final machine jamming risk evaluation result is obtained. The gradient boosting tree algorithm automatically learns the nonlinear relationship between the two and the actual risk grade, realizes the collaborative perception of geological causes and equipment response, comprehensively and dynamically masters the causes and evolution process of machine jamming risk, ensures the prediction accuracy, and realizes the quantization of the model result.

[0020] The present application introduces a dynamic adjustment mechanism based on real-time data fluctuation and risk sensitivity, so that the comprehensive cloud model can not only respond more accurately to complex and dynamic geological conditions, but also maintain efficient and accurate risk assessment ability in the case of geological mutation.

[0021] The advantages of the additional aspects of the present application will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0022] The drawings accompanying the specification of the present application form a part thereof and serve to provide further understanding of the present application, the illustrative embodiments of the present application and its description serve to explain the present application, and do not constitute improper limitations on the present application.

[0023] Figure 1 A flowchart for the process of index selection, weight adjustment, cloud droplet generation and comprehensive cloud model construction in the geological risk evaluation in the present application embodiment one;

[0024] Figure 2 A deep learning network structure based on LSTM and attention mechanism and its specific implementation process in torque dynamic prediction in the present application embodiment one;

[0025] Figure 3 A process of gradient boosting tree fusion analysis of geological risk evaluation and torque prediction results and an output process of comprehensive risk grade in the present application embodiment one;

[0026] Figure 4 A flowchart of a process for generating a tunneling parameter adjustment scheme or a breakout measure under different risk levels in the first embodiment of the present application is shown in Figure 1.

[0027] Figure 5 A schematic diagram of a geological risk-torque prediction driven jamming risk assessment system in the second embodiment of the present application is shown in Figure 2.

[0028] Figure 6 A schematic diagram of a jamming risk classification in the first embodiment of the present application is shown in Figure 3. DETAILED DESCRIPTION

[0029] It should be noted that the following detailed description is exemplary in nature and is intended to provide further description of the present application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0030] It is to be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit exemplary embodiments according to the present application.

[0031] In the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0032] Embodiment One

[0033] The present embodiment discloses a geological risk-torque prediction driven jamming risk assessment method, comprising:

[0034] Obtaining geological parameters of a tunneling section and main control parameters of a tunneling machine;

[0035] According to adverse geology in the tunneling process and its corresponding geological parameters and geophysical parameters, assigning initial weights to the geological parameters and the geophysical parameters and dividing the geology into multiple risk levels to construct a comprehensive cloud model; dynamically adjusting the weights of the geological parameters and / or the geophysical parameters according to the fluctuation and risk sensitivity of the geological parameters and / or the geophysical parameters in the tunneling process, and obtaining the geological risk level of the tunneling section according to the updated comprehensive cloud model;

[0036] Based on the geological parameters, the geophysical parameters and the main control parameters of the tunneling machine, a trained torque prediction model is used to obtain the running risk level of the tunneling machine;

[0037] Based on the jamming risk prediction model constructed by the gradient boosting tree, the geological risk level and the running risk level of the tunneling machine are fused to obtain the final jamming risk assessment result.

[0038] The embodiment is realized by constructing a comprehensive cloud model. The cloud model has strong advantages in describing the uncertainty, fuzziness and evolution trend among multi-source data, has the ability to fuse qualitative cognition and quantitative analysis, can accurately depict the randomness and fuzzy boundary of risk distribution in a complex geological environment with strong parameter volatility and insufficient information integrity, and can improve the stability and adaptability of risk level judgment. In addition, the invention introduces a dynamic adjustment mechanism based on real-time data volatility and risk sensitivity, so that the comprehensive cloud model can not only respond more accurately to complex and dynamic geological conditions, but also maintain efficient and accurate risk assessment ability in the event of geological mutations. In the embodiment, according to the predicted geological risk level and the tunneling machine operation risk level, gradient boosting tree algorithm is used for fusion analysis to obtain the final jamming risk evaluation result. The gradient boosting tree algorithm automatically learns the nonlinear relationship between the two and the actual risk level, ensuring the accuracy of the prediction while realizing the quantification and interpretation of the model results.

[0039] The following will be combined Figures 1-4 The geological risk-torque prediction driven jamming risk evaluation method proposed in the embodiment will be described in detail:

[0040] Step 1: Obtain the geological parameters, geophysical parameters and main control parameters of the tunneling machine of the tunneling section.

[0041] Specifically, the geological parameters, geophysical parameters and main control parameters of the excavator related to the tunneling process are obtained. Among them, according to the division of typical geological unit types, mainly including three types of unfavorable geological conditions such as fault fracture zone, karst and alteration zone. For the fault fracture zone, the collected geological parameters include fault width, rock integrity coefficient (RQD), joint spacing, water seepage and rock compressive strength; for karst geology, the collected geological parameters include the spatial distribution and volume of karst cave, fracture water pressure, permeability coefficient, distribution ratio of carbonate rock and water inflow; and in the alteration zone geology, the collected geological parameters cover the changes of rock chemical composition, rock mechanical properties, swelling index, fracture development degree and water seepage and water inflow and other key indicators.

[0042] The geophysical parameters include resistivity, relaxation time, longitudinal wave velocity, polarization rate and other parameters. The main control parameters of the excavator include TBM cutterhead diameter, cutterhead thickness, cutterhead opening rate and other key indicators.

[0043] Step 2: According to the adverse geological types in the tunneling process and their corresponding geological parameters and geophysical parameters, initial weights of risk indicators are given to the geological parameters and geophysical parameters, and the geological types are divided into multiple risk levels to construct a standard cloud model; according to the fluctuation and risk sensitivity of the geological parameters and / or geophysical parameters in the tunneling process, the weights of the geological parameters and / or geophysical parameters are dynamically adjusted, and the real-time values are substituted into the cloud generator to generate a to-be-identified cloud, and the membership between the to-be-identified cloud and the standard cloud is calculated to obtain the geological risk level of the tunneling section.

[0044] In this embodiment, the collected geological parameters and geophysical parameters are comprehensively evaluated, and a cloud model is used as a core evaluation tool. The cloud model has strong advantages in describing the uncertainty, fuzziness and evolution trend between multi-source data. Compared with the traditional weighted scoring method or analytic hierarchy process, the cloud model has the ability to integrate qualitative cognition and quantitative analysis, can accurately depict the randomness and fuzzy boundary of risk distribution in a complex geological environment with strong parameter volatility and insufficient information integrity, and can improve the stability and adaptability of risk level judgment. To further enhance the sensitivity of the cloud model, a dynamic weight adjustment mechanism is introduced based on the cloud model in this embodiment to realize feedback correction of real-time monitoring data, thereby enhancing the response capability to geological evolution.

[0045] In practical application, first, according to different risk characteristics of typical adverse geological types such as fracture and crush zone, karst, alteration zone, etc., a geological risk evaluation index system is constructed. The geological risk evaluation index system includes several geological key parameter indexes, such as rock integrity index , underground water permeability index , development degree of soft interlayer , etc., which are uniformly denoted as , where represents the total number of geological parameters and geophysical parameters. In order to reasonably express the influence degree of each parameter index on the risk evaluation result, the initial weight of each parameter index needs to be initially given according to expert experience or historical case data, and the sum of the weights satisfies:

[0046] To characterize geological risk levels, this embodiment constructs a standard cloud model based on cloud model theory. Each parameter is divided into multiple risk level ranges according to practical engineering experience, typically into six levels: "Stable," "Weak Risk," "Medium Risk," "Relatively High Risk," "High Risk," and "Extremely High Risk." For each risk level's numerical range, its cloud model triplet parameters are calculated. The triplet parameters include the expected value Ex, entropy En, and hyperentropy He. The expected value Ex represents the typical characteristic value of that risk level; entropy En reflects the uncertainty within that risk level; and hyperentropy He further characterizes the fluctuation of entropy itself, reflecting the random perturbation characteristics of the data.

[0047] The calculation formulas for the three are as follows:

[0048]

[0049] in, and These are the maximum and minimum values ​​corresponding to this risk level, respectively; It is an empirical constant, and its value generally ranges from 0.1 to 0.5.

[0050] Subsequently, a standard cloud model was constructed using the aforementioned parameters, and a large number of cloud droplets were generated using a forward cloud generator. The specific process includes: first using... As expected, To determine the variance, generate random entropy values. , and then As expected, Generate random variables for variance Finally, the membership degree of the cloud droplets is calculated. This is used to describe the degree of matching between the data point and the risk level.

[0051] Large-scale cloud droplet sampling can generate standard-level cloud maps, which can be used for subsequent cloud similarity assessment. The calculation formula is as follows:

[0052]

[0053]

[0054]

[0055] During the actual tunneling process, the monitoring equipment continuously collects real-time values ​​of various parameters. , Indicates the first The parameters at time... The value of the cloud model. In order to make the cloud model have dynamic response ability, the embodiment introduces "fluctuation analysis" and "sensitivity analysis" mechanism in the risk identification stage, which constitutes the real-time self-adaptive adjustment system of weight. Fluctuation analysis is used to capture the change amplitude of the index in time series, which is defined as:

[0056]

[0057] If the fluctuation amplitude exceeds the fluctuation value threshold , it is considered that the parameter has mutation, which is a significant disturbance, and its weight should be increased.

[0058] At the same time, the sensitivity analysis is further used to evaluate the influence degree of each parameter index on the overall risk result. The risk assessment function is defined as In the embodiment, the risk assessment function can be constructed in linear or nonlinear combination form, such as linear weighted model, fuzzy reasoning function or neural network output, and the specific form can be flexibly set according to the actual engineering scene.

[0059] The sensitivity of the first index is defined as:

[0060]

[0061] When , it is considered that the parameter is a high-sensitive index, and its influence should be enhanced; wherein, is the sensitivity threshold.

[0062] Based on the above result analysis, the dynamic weight can be updated according to the following formula:

[0063]

[0064] Wherein, represents the weight of the first index at time ; represents the weight adjustment factor, which reflects the influence strength of fluctuation and sensitivity; the denominator is used for normalization, to ensure that the sum of all weights is still 1.

[0065] Finally, after completing the identification cloud construction of each parameter index, the comprehensive identification cloud diagram is formed by weighting and fusing all parameter index clouds. The comprehensive identification cloud diagram is compared with the cloud diagram of each standard risk level, and the maximum membership degree principle is used to judge the closest level.

[0066] The specific judgment criteria are as follows:

[0067]

[0068]

[0069] wherein, represents the degree of matching of the cloud with the th risk level; represents the membership degree of the th cloud drop in the th risk level criterion cloud; represents the total number of cloud drops; is the risk level of the current tunneling section determined by the maximum membership degree principle of the final output.

[0070] Step 3: Based on the geophysical parameters, stratum parameters and main control parameters of the tunneling machine, the trained torque prediction model is used to obtain the running risk level of the tunneling machine.

[0071] In this embodiment, the torque prediction model is constructed based on a deep learning framework, combining a long short-term memory network (LSTM), an attention mechanism and a geological disturbance perception gating mechanism, aiming to accurately model and dynamically predict the change trend of the cutter torque in the tunneling process, and improve the forward-looking identification ability of the running risks such as machine jamming. The torque prediction model of this embodiment has significant advantages in processing TBM multi-source input parameters, complex geological coupling effects and sudden disturbance responses, and can be widely used in intelligent decision-making auxiliary systems for shield construction in complex geological conditions.

[0072] The torque prediction model input is the multi-dimensional data collected in real time, including the device running state and the surrounding geological environment, covering main control parameters such as cutter speed, propulsion thrust, penetration, shield friction, hydraulic system pressure, while integrating typical geological indicators such as surrounding rock strength (RQD, UCS), karst cave characteristics, fracture width, and combining geophysical parameters such as resistivity, relaxation time, polarization rate, and longitudinal wave velocity.

[0073] After each item of data is collected, it is preprocessed by a data normalization method, and the standardization calculation formula is:

[0074] wherein, represents the value of the th sample in the original feature vector, for example, the original value of the propulsion speed, current, etc. at the th time point; and represent the minimum and maximum values of the feature taken in all samples, respectively. represents the normalized value, i.e. the value after linearly mapping to the interval .

[0075] All input samples are organized as multi-dimensional tensors in a sliding time window manner, with tensor dimensions of where, represents the tensor of input data, which is a three-dimensional array; represents that each element in the tensor is a real number; represents the batch size, represents the time step, represents the number of feature dimensions. The preprocessed data is input into the LSTM network to extract the memory and dependency structure of the input sequence in the time dimension.

[0076] The internal structure of the LSTM network includes an input gate, a forget gate, an output gate, and a state update mechanism. The function of the input gate is to update the unit state and control the input vector entering the storage unit; the function of the forget gate is to decide which information should be discarded or retained, and its opening and closing means whether the information can be passed to the next moment; the function of the output gate is to determine the value of the next hidden state, which contains the information of the previous input. Its update formula is as follows:

[0077] Input gate :

[0078] Forget gate :

[0079] Output gate :

[0080] Candidate state :

[0081] Cell state update :

[0082] Hidden state update :

[0083] where, represents the input feature of the th time step; represents the hidden state corresponding to the th time step is the hidden state of the previous time step; is the memory unit of the th time step, i.e., the cell state, is the memory unit of the previous step; are the weight matrices connected to the input and the current state, respectively, which act on ; are the weight matrices for input and past hidden state, respectively, acting on ; are the bias terms; represents the Sigmoid function; represents the hyperbolic tangent function. The output of the LSTM network is a sequence of hidden states , which are used to characterize the influence of features at each time step on the target variable.

[0084] To further enhance the model's ability to identify and express key time segments, an attention mechanism is introduced to weight the aggregation of hidden state sequences. The calculation of attention weights first constructs a scoring function to evaluate the relevance between each hidden state and the global state . The global state may be set as or a trainable vector, and the scoring function takes the following form:

[0085]

[0086] where the result of the calculation represents the "importance" of the th time step to the target; is the weight matrix that maps hidden states to the attention space; is the weight matrix that maps the global state to the attention space; represents the hyperbolic tangent function; is a trainable weight vector that acts to compress the nonlinear result into a scalar score. The superscript T represents the transpose.

[0087] The attention weights are obtained by Softmax normalization:

[0088]

[0089] where is the total number of time steps.

[0090] The weighted context vector is then obtained as:

[0091]

[0092] Based on the traditional LSTM and attention structure, this embodiment further introduces a geological disturbance perception gating mechanism to achieve dynamic response adjustment to sudden geological anomalies. The disturbance feature vector According to the lithology change rate, fault density, and water seepage quantity slope, a disturbance event label is generated when the variation coefficient of these disturbance indicators in the sample section exceeds a threshold value, such as 3σ or the water seepage rate change slope is greater than a critical value. The context vector is spliced, and a disturbance adjustment factor is generated via a gating function , and the expression is as follows:

[0093]

[0094] wherein, is the adjustment coefficient calculated by the disturbance gating mechanism, ranging from , reflecting whether the geology is smooth. represents a vector splicing operation, and are the gating layer weights and biases, respectively. The final prediction output after disturbance adjustment is a weighted fusion form of the current prediction value and the historical trend value:

[0095]

[0096] wherein, is the torque value predicted based on the context vector , i.e., the torque value predicted by the LSTM and the attention mechanism; represents the historical torque sliding mean or weighted trend value, which is used to provide a robust reference when the disturbance signal is significant. By introducing the geological disturbance perception gating mechanism, the torque prediction model can realize dynamic correction of the prediction output when facing geological mutations, effectively improving the sensitivity and response robustness to abnormal fluctuations.

[0097] During the training process of the torque prediction model, the mean square error is used as the loss function, and the optimization target is:

[0098]

[0099] wherein, is the sample number, and are the true value and the predicted value, respectively. The optimizer uses the Adam algorithm combined with the learning rate adjustment strategy and the EarlyStopping mechanism to ensure the efficiency and generalization ability of the training process. After training is completed, the model is evaluated on the test set, and common indicators include the mean square error (MSE), the mean absolute error (MAE), and the determination coefficient (R ), which are used to quantify the prediction accuracy and stability.

[0100] Finally, the torque prediction model can be deployed in the TBM tunneling real-time control system to continuously predict the trend of cutterhead torque changes based on current input parameters, and combined with the set threshold to divide the risk level. When the predicted value is in the stable interval, it is judged as low risk; when the predicted value approaches the abnormal upper limit or is accompanied by significant increase in disturbance signal, it is judged as medium-high risk, and the early warning mechanism is triggered or the adjustment of tunneling parameters is suggested. The overall structure of the model has good generalization ability and field adaptability, and is especially suitable for intelligent TBM tunneling control system under complex geological conditions and multiple disturbance conditions.

[0101] Step 4: Based on the gradient boosting tree constructed by the jam risk prediction model, the geological risk level and the tunneling machine operation risk level are fused to obtain the final jam risk evaluation result.

[0102] In this embodiment, the jam risk evaluation model based on double-drive information fusion adopts gradient boosting tree algorithm (XGBoost) to build, aiming to realize the coupling modeling between geological risk and equipment torque response, and output a unified TBM jam risk level. As the comprehensive discrimination core of the system, the jam risk evaluation model receives the key feature information output by the cloud model and the torque prediction model, automatically learns the nonlinear relationship between the two and the actual jam risk level through machine learning, while ensuring the prediction accuracy, realizing the quantization and interpretation of the model results.

[0103] In the process of constructing the jam risk evaluation model, first, the various input features are structured and preprocessed, including the comprehensive description information of the geological evaluation results and the equipment operation state, among which the geological part covers the risk level and its corresponding representative indicators output by the comprehensive cloud model, such as fault width, water seepage, swelling index, etc.; the torque part introduces the torque level, mean, fluctuation range, penetration, etc. Feature parameters obtained by the torque prediction model.

[0104] To enhance the discrimination ability of the jam risk evaluation model, this embodiment processes the classification features through one-hot encoding, and the continuous data is normalized, thereby constructing a unified data input format. Each set of samples corresponds to an actual construction scene, and the risk level is supervised and trained through historical labeled data, and the output label is a three-level classification result, representing low risk, medium risk and high risk respectively.

[0105] In the double-drive information fusion jam risk evaluation model based on XGBoost constructed in this embodiment, the input features are composed of geological risk features and equipment operation features, which are unified to form a structured risk feature vector, including:

[0106] The geological feature subset from the cloud model includes geological risk levels and their corresponding key geological parameters, such as fault width, RQD value, water permeability, swelling index, joint spacing, and fracture water pressure; the equipment feature subset includes the equipment operation risk levels output by the torque prediction model and their corresponding main control parameters of the tunneling machine, such as the cutter torque prediction value, fluctuation amplitude, penetration, thrust, and shield frictional force.

[0107] Among them, the geological risk level and the equipment operation risk level are used as classification features and are represented by one-hot encoding; the remaining continuous variables are normalized and input into the fusion model, and finally a high-dimensional risk feature vector in a unified format is constructed as the input of the XGBoost model.

[0108] The XGBoost algorithm is an enhanced gradient decision tree method that has strong generalization ability for high-dimensional heterogeneous data and sensitive capture ability for changes in feature importance. During the training process of the XGBoost model, the coupling law between geological features and torque features is automatically fitted through an iterative structure, further optimizing the tree structure and node division.

[0109] According to the division of the training set, the validation set, and the test set, by adjusting the learning rate, tree depth, and the number of weak classifiers, the XGBoost model can avoid overfitting while ensuring the fitting performance, and improve its discriminant ability for unknown samples. Its loss function includes two parts: model prediction error and structural complexity.

[0110]

[0111] Among them, represents the entire parameter set of the XGBoost model; represents the th sample ( ), represents the total number of samples; represents the th sample's true label, i.e., the actual risk level; represents the th sample's XGBoost model prediction value; is the loss function, such as squared error or log loss, which measures the difference between the prediction value and the true value; represents the th sample, , represents the total number of constructed decision trees; represents the th regression tree; represents the complexity penalty term of the th tree, such as the number of leaf nodes and the square of the weight.

[0112] During the training process of the XGBoost model, the GridSearch and K-fold cross-validation methods are used to adjust hyperparameters such as learning rate, tree depth, and number of sub-trees, and the model performance is evaluated based on the training set and the validation set.

[0113] To enhance the interpretability of the XGBoost model, this embodiment introduces a feature importance analysis mechanism to quantify the contribution of each input variable. In the XGBoost-based dual-drive information fusion card risk assessment model, the quantification of variable contribution is not only used for post-hoc explanation, but also directly participates in the optimization design of the XGBoost model structure and input space.

[0114] The feature importance analysis mechanism introduces three types of feature importance indicators as follows: The number of times the feature is used as a split node in all trees; The loss function reduction amount caused by using the feature for each split; The weight sum of the samples covered by the feature in the split node.

[0115] Due to the physical property differences between geological and equipment features, a differentiated index priority strategy is adopted. Geological features prefer to use Gain to measure their fine partitioning ability, torque features prefer to use Cover to measure their global response ability, and for features that frequently participate in structure but have large fluctuations, Weight and Gain are used for comprehensive evaluation.

[0116] The importance variation coefficient (Coefficient of Variation, CV) is introduced to monitor the stability of the feature:

[0117]

[0118] wherein, represents the set of Gain values obtained by the th feature in multiple rounds of training; represents the standard deviation of the Gain value of the th feature, used to measure the fluctuation; the mean of the Gain value of the th feature, used to measure the contribution level; if a feature has too large fluctuations in consecutive rounds, it is determined to be a "weak stable factor" and is removed from the current training round; if a feature has stable contribution in multiple rounds, it is preferentially selected in node partitioning.

[0119] To capture the cross-coupling effect between geological variables and torque variables, a joint contribution evaluation mechanism is introduced. If a combination of features significantly improves the target function reduction, it will be automatically added as a cross-term to participate in tree building. The joint contribution evaluation mechanism defines the joint importance function of the interactive feature combination:

[0120]

[0121] wherein, represents the characteristic contribution degree of the joint contribution of ; represents the first geological class feature, represents the first torque class feature. If the value exceeds the set threshold value, an intersection feature item is automatically generated and is included in the candidate split feature set of the current round tree structure to participate in tree building. This mechanism can effectively capture the nonlinear interaction effect between cross-domain variables and improve the recognition ability of the model to complex induction mechanisms.

[0122] After training, XGBoost outputs the final prediction result by weighted voting. For each set of input features, the XGBoost model will calculate the probability of each category and select the category corresponding to the maximum probability as the final risk level. This process is based on the Softmax function, which is used to convert the prediction value output by XGBoost into a probability value:

[0123]

[0124] wherein, represents the prediction score of the XGBoost model for the first risk level; represents all category indexes for normalization; represents the probability of being classified as category . By selecting the class with the maximum probability, XGBoost gives the risk level corresponding to the current construction state.

[0125] To enhance the interpretability of the XGBoost model output, the SHAP (Shapley Additive Explanations) mechanism is further introduced, and its calculation formula is as follows:

[0126]

[0127] wherein, each SHAP value represents the marginal contribution value of feature i to the final prediction of the XGBoost model; F represents the set of all input features in the XGboost model, including variables from geological prediction and torque prediction, such as water seepage, pushing speed, torque fluctuation, etc. represents all feature subsets that do not contain feature i .​

[0128] The embodiment visualizes the contribution ranking of variables such as "water infiltration amount", "propulsion fluctuation" and "fault width" through SHAP analysis results, thereby clarifying the influence weight of each feature on high-risk prediction and assisting construction personnel in developing risk source traceability and directional adjustment of control strategies.

[0129] The machine jam risk assessment model constructed in the embodiment not only breaks through the bottleneck that traditional experience rules are difficult to depict the interaction of complex working conditions, but also establishes the mapping relationship between risk level and input parameters through machine learning technology, realizes the logical closed loop between risk identification and proactive prevention and control strategies. The model can be supported by local deployment in the field monitoring system, and can also continuously absorb new construction data through a cloud training platform to realize model self-learning and continuous optimization, thereby meeting the requirements of dynamic adaptability under complex geological conditions and universality under large-scale construction environment.

[0130] Step 5: Based on the machine jam risk evaluation results, realize the dynamic output of risk response strategies.

[0131] In the embodiment, based on the aforementioned comprehensive risk level judgment results, the excavation parameters are dynamically adjusted, the escape suggestions are generated, and the automatic control and early warning response are realized under necessary conditions, so as to realize the closed-loop management of machine jam risk and the real-time guarantee of construction safety.

[0132] Integrating risk response identification, parameter optimization decision, emergency disposal execution and multi-level information support, its core operation mechanism revolves around the main line of "risk level driving parameter behavior adjustment". When it is determined as a low risk level, the existing excavation parameters remain unchanged, and the continuous monitoring mechanism of the equipment state is started. When it is determined as a medium risk level, the preset optimization model is called, and based on the torque change trend and the geological environment evolution characteristics, the parameter adjustment suggestions such as reducing the propulsion thrust, reducing the cutterhead speed or optimizing the penetration depth are intelligently generated to slow down the equipment load and delay the risk evolution. And through the control link, the parameter adjustment instructions are issued to the equipment control system in real time to realize the immediate adjustment and feedback of the excavation working condition.

[0133] When identified as a high-risk state, the emergency linkage mode is entered. At this time, according to the current risk type such as jamming, water inrush, surrounding rock instability, etc. and the combination of inducing characteristics, the pre-defined response strategy library is called to automatically generate a targeted escape plan. For example, in the typical jamming risk situation, combined with the torque fluctuation rate, the rate of penetration depth decline and the geological grade information, it can be judged as "bit blocked type jamming", and then it is suggested to switch the tunneling mode such as switching from double shield to single shield, synchronously enabling the front face grouting device to reduce the surrounding rock resistance, and realizing step-by-step relief through shield tail thrust feedback. In the water inrush risk scenario, the speed reduction mechanism and the face grouting module are linked, and the face seepage pressure is reduced through the shield drainage system to achieve rapid drainage and risk transfer. All schemes can be executed by the control system through a programmable logic controller (PLC) to ensure that the system has automatic processing capability in high-risk scenarios.

[0134] To enhance operation transparency and assist decision-making ability, the embodiment synchronously outputs a graphical interface and a textual suggestion, including the current risk judgment basis, recommended adjustment strategy and expected effect evaluation. Through this auxiliary interface, the operator can master the device running state, risk source and control response path in real time, effectively reducing human decision-making delay and improving reaction efficiency.

[0135] The parameter adjustment and strategy execution during the operation of the embodiment have a complete closed-loop mechanism. When the device receives a parameter adjustment instruction, the execution effect is automatically verified through feedback signals, and the risk state is re-evaluated in combination with real-time monitoring data, and the control strategy is updated again if necessary, forming a continuous cycle mechanism of "evaluation-response-feedback-re-evaluation", ensuring that the tunneling conditions are stable and controlled.

[0136] The active prevention and control of the embodiment can realize risk perception based on data driving, parameter optimization based on intelligent algorithm, strategy execution based on control system and continuous adjustment based on feedback mechanism in complex geological and high-risk scenarios, forming a whole-process dynamic closed-loop control system from "identification" to "response", significantly improving the adaptability and safe operation level of the tunneling machine in complex geological conditions, and providing key control support for TBM construction intelligence.

[0137] For example, the tunneling construction in the fracture and fracture zone is described as an example:

[0138] First, a fracture and fracture zone risk database is established. The database contains multi-dimensional information related to the fracture and fracture zone, such as fault width, rock mass integrity coefficient (RQD), joint spacing, water seepage, surrounding rock strength (UCS), historical construction records and corresponding jamming risk level. The database also covers the classification of fracture and fracture zone such as high, medium and low risk and the corresponding characteristic description, providing data support for subsequent risk assessment.

[0139] Next, the data for training is screened and prepared. Geological information and equipment operation data are collected through field construction records and geophysical, drilling, and other means. The geological information includes parameters such as fault width of fracture zone, rock mass integrity coefficient (RQD), water seepage, surrounding rock strength (UCS), etc.; the equipment operation information includes cutterhead speed, propulsion thrust, penetration, shield friction, hydraulic system pressure, etc. After cleaning and normalization processing of the collected data, features with high correlation to the fracture zone jamming risk are selected as model inputs.

[0140] Subsequently, model training is performed. The screened data is input into the dual-drive risk comprehensive analysis model. The model is composed of a geological risk evaluation model and a torque prediction model. The geological risk evaluation model uses a cloud model to evaluate the geological risk based on geological information, and generates and fuses cloud droplets to calculate the geological risk level, such as A, B, C. The torque prediction model uses a deep learning model (LSTM + attention mechanism) to perform time series analysis on equipment operation parameters (such as cutterhead speed, penetration), predict torque values and fluctuation trends, and evaluate equipment operation risk levels. Training is completed by minimizing the loss function, and historical data of geological risk and torque risk are used to optimize model weights, and the contribution of both to comprehensive risk is learned through gradient boosting trees (XGBoost).

[0141] Then, risk assessment and level output are performed. The geological information and equipment operation parameters of the fracture zone section that needs to be assessed are input into the trained dual-drive risk comprehensive analysis model. The geological information is calculated by the cloud model module, and the fracture zone geological risk level is output, such as A, B, C; the equipment operation parameters are analyzed by the deep learning module, and the equipment operation risk level is output, such as low, medium, and high. The geological risk and torque risk are analyzed by gradient boosting trees (XGBoost), and the comprehensive risk level is output, such as low risk, medium risk, and high risk. The comprehensive risk level is compared with the results in the fracture zone risk database, the results are checked, and the reliability of the assessment results is verified.

[0142] Next, linkage active prevention and control. According to the comprehensive risk assessment results, linkage active prevention and control is performed, and excavation parameter adjustment or escape plan suggestions are provided. When the risk is low, the current excavation parameters are maintained, and real-time monitoring is performed; when the risk is medium, the excavation parameters are adjusted, such as reducing the cutterhead speed, propulsion thrust, and optimizing the penetration; when the risk is high, an escape plan is generated, such as switching the excavation mode, such as switching from double shield to single shield, implementing chemical grouting or advanced support.

[0143] Finally, the results are checked and optimized. The model output results are compared with the actual risk conditions recorded during construction to verify the effectiveness of the model in predicting the risk of machine jamming in fault and fracture zones. By continuously accumulating new construction data to update the fault and fracture zone risk database and model parameters, the applicability and accuracy of the risk assessment model are improved.

[0144] The geological risk assessment-torque prediction dual-drive machine jamming risk evaluation method proposed in this embodiment has the following significant beneficial effects:

[0145] 1. Dynamic response and scientific decision-making. By linking with risk assessment, this embodiment can monitor the geological and equipment operating conditions in real time, trigger corresponding control strategies based on risk levels, and dynamically adjust the tunneling parameters and generate escape plans. It provides scientific decision-making support based on data-driven, avoids misjudgments that may be caused by relying solely on human experience, and improves the accuracy and reliability of decision-making.

[0146] 2. Active prevention and control and efficient execution. The active prevention and control of this embodiment has the functions of tunneling parameter optimization and automatic execution, which can actively adjust the tunneling parameters such as propulsion thrust, cutterhead speed or directly start supporting equipment such as advanced grouting and small catheter arrangement under medium and high risk levels. In high-risk state, the system can trigger emergency shutdown and alarm, realizing full automation of rapid response and risk prevention and control, significantly reducing the mistakes and delays caused by human intervention.

[0147] 3. Risk prevention and enhanced escape ability. Under medium risk level, this embodiment provides optimized construction parameter suggestions to actively reduce equipment operating load and prevent risk from further escalation. Under high risk level, it can quickly generate and implement targeted escape plans such as switching tunneling mode, chemical grouting support, drainage hole dredging, etc. Through precise and efficient risk handling, the system's escape ability is significantly enhanced, reducing the impact of common construction problems such as machine jamming, water inrush, and surrounding rock instability on construction progress.

[0148] 4. Dynamic adaptation and closed-loop control. This embodiment uses real-time monitoring data to optimize tunneling parameters to form a closed-loop control mechanism, which can dynamically adjust control measures according to risk changes to ensure that equipment operates within a safe range. Through real-time dynamic linkage of risk levels and parameter adjustments, the system can adapt to changes in the construction environment, ensuring the safety and continuity of the construction process under complex geological conditions.

[0149] 5、Construction safety and efficiency are improved. The embodiment effectively reduces the equipment downtime and safety accident rate caused by machine jamming, water inrush or surrounding rock instability during construction through scientific decision-making and proactive prevention and control, and optimizes the tunneling parameters to improve the construction efficiency. In complex geological conditions, especially in high-risk scenes such as fault fracture zone, karst, alteration zone, etc., the embodiment can achieve the best balance between construction efficiency and safety.

[0150] 6、Technical scalability and applicability. The proactive prevention and control of the embodiment has good technical scalability and can adapt to various risk types and construction conditions. For example, the model accuracy and application range can be further optimized by combining more rich equipment operation parameters such as vibration, temperature and geological monitoring data such as pressure and water inflow. In addition, the embodiment can be directly deployed to the field control system, supporting dual-end operation of local and cloud, realizing real-time risk prevention and control and construction optimization in multiple scenes.

[0151] In summary, the embodiment realizes intelligent monitoring, dynamic adjustment and efficient prevention and control of the whole construction process through the proactive prevention and control module, effectively solves the safety and efficiency problems of tunneling in complex geological conditions, provides strong technical support for the intelligentization and automation development of TBM construction, and has significant engineering application value.

[0152] Embodiment two

[0153] As shown in Figure 5 , the purpose of the embodiment is to provide a tunneling machine jamming risk evaluation system, comprising:

[0154] A data acquisition module is configured to acquire geological parameters, geophysical parameters and tunneling machine main control parameters of a tunneling section;

[0155] A geological risk evaluation module is configured to assign initial weights to the geological parameters and the geophysical parameters according to adverse geology in the tunneling process and the corresponding geological parameters and geophysical parameters, divide the geology into multiple risk levels, and construct a comprehensive cloud model; dynamically adjust the weights of the geological parameters and / or the geophysical parameters according to the fluctuation and risk sensitivity of the geological parameters and / or the geophysical parameters in the tunneling process, and obtain the geological risk level of the tunneling section according to the updated comprehensive cloud model;

[0156] A torque prediction module is configured to obtain a tunneling machine operation risk level by using a trained torque prediction model based on the geological parameters, the geophysical parameters and the tunneling machine main control parameters;

[0157] A double-drive evaluation module is configured to fuse the geological risk level and the tunneling machine operation risk level based on a machine jamming risk prediction model constructed by gradient boosting tree, and obtain a final machine jamming risk evaluation result.

[0158] In the embodiment, an active prevention and control module is further included, which is configured to enter an emergency linkage mode when the system identifies a high-risk state, at which time the active prevention and control module automatically generates a targeted escape plan according to a predefined response strategy library based on the current risk type such as a jamming machine, water inrush, surrounding rock instability, etc. and a triggering feature combination. For example, in a typical jamming machine risk situation, the system can combine the torque fluctuation rate, the rate of penetration depth decline, and the geological grade information to determine that it is a "cutterhead blocked type jamming machine", and then suggest switching the tunneling mode such as switching from a double shield to a single shield, synchronously starting a front face grouting device to reduce the surrounding rock resistance, and achieving step-by-step relief through shield tail thrust feedback. In a water inrush risk scenario, the system links the speed reduction mechanism and the face grouting module, and reduces the face seepage pressure through the shield drainage system to achieve rapid drainage and risk transfer.

[0159] In more embodiments, the following are further provided:

[0160] An electronic device includes a memory and a processor, and computer instructions stored on the memory and run on the processor, when the computer instructions are run by the processor, the method described in embodiment one is completed. For brevity, this will not be described here.

[0161] It should be understood that in the embodiments, the processor can be a central processing unit CPU, and the processor can also be other general-purpose processors, digital signal processors DSPs, application-specific integrated circuits ASICs, ready-to-program gate arrays FPGA or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0162] The memory can include read-only memory and random access memory, and provide instructions and data to the processor, and a part of the memory can also include non-volatile random access memory. For example, the memory can also store device type information.

[0163] A computer readable storage medium for storing computer instructions, when the computer instructions are executed by a processor, the method described in embodiment one is completed.

[0164] The method in embodiment one can be directly embodied as a hardware processor to execute and complete, or executed by a combination of hardware and software modules in the processor. The software module can be located in a storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, register, etc. The storage medium is located in the memory, and the processor reads the information in the memory, and combines the hardware to complete the steps of the above method. To avoid repetition, it will not be described in detail here.

[0165] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the present embodiment can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software manner depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0166] Although the specific embodiments of the present application are described above in combination with the drawings, it is not a limitation on the scope of protection of the present application. Those skilled in the art should understand that various modifications or variations made by those skilled in the art on the basis of the technical solutions of the present application without creative labor are still within the scope of protection of the present application.

Claims

1. A method for evaluating the risk of a stuck pipe, driven by the prediction of the torque, characterized in that, The method comprises the following steps: obtaining geological parameters, geophysical parameters and main control parameters of a tunneling machine of a tunneling section; according to adverse geology and corresponding geological parameters and geophysical parameters, assigning initial weights of risk indicators to the geological parameters and the geophysical parameters, and dividing geology into multiple risk levels to construct a standard cloud model; based on the standard cloud model, dynamically adjusting weights of the geological parameters and / or the geophysical parameters according to fluctuation conditions and risk sensitivity of the geological parameters and / or the geophysical parameters in a tunneling process, and obtaining a geological risk level of the tunneling section by weighted fusion of prediction results of all risk indicators; based on the geological parameters, the geophysical parameters and the main control parameters of the tunneling machine, obtaining a tunneling machine operation risk level by using a trained torque prediction model; the processing process of the torque prediction model on the geological parameters, the geophysical parameters and the main control parameters of the tunneling machine is as follows: using a long short-term memory neural network to extract long-term dependence of input data of a time series; using an attention mechanism to weight and aggregate hidden states of the long short-term memory neural network to obtain a weighted context vector; introducing a geological disturbance perception gating mechanism to splice a disturbance feature and the weighted context vector to obtain a disturbance adjustment factor; using the disturbance adjustment factor to weight and fuse torque values predicted by the trained long short-term memory network based on the attention mechanism, and combining a historical torque sliding mean or a weighted trend value to obtain a prediction result of the torque prediction model; based on a gradient boosting tree, a machine jam risk prediction model is constructed to fuse the geological risk level and the tunneling machine operation risk level to obtain a machine jam risk evaluation result.

2. The geological risk-torque prediction driven jam risk assessment method of claim 1, wherein, According to adverse geology and corresponding geological parameters and geophysical parameters, assigning initial weights of risk indicators to the geological parameters and the geophysical parameters, and dividing geology into multiple risk levels to construct a standard cloud model, specifically: according to three kinds of adverse geology, i.e., fracture and fracture zone, karst and alteration zone, establishing a corresponding risk indicator system; each risk indicator is divided into multiple risk level intervals, and the three-element parameter of the cloud model is calculated according to the numerical range of each risk level; according to the calculated three-element parameter, using a forward cloud generator to generate cloud droplets to establish a standard cloud model.

3. The geological risk-torque prediction driven jam risk assessment method of claim 1, wherein, Based on the standard cloud model, dynamically adjusting weights of the geological parameters and / or the geophysical parameters according to fluctuation conditions and risk sensitivity of the geological parameters and / or the geophysical parameters in a tunneling process, specifically: determining fluctuation values of the geological parameters and the geophysical parameters according to variation amplitudes of the geological parameters and the geophysical parameters in a time dimension; determining sensitivities of the geological parameters and the geophysical parameters according to correlations between the geological parameters and the geophysical parameters and risk evaluation results; dynamically adjusting initial weights of the geological parameters and / or the geophysical parameters according to comparison results of the fluctuation values and the sensitivities of the geological parameters and the geophysical parameters with corresponding fluctuation value thresholds and sensitivity thresholds.

4. The geological risk-torque prediction driven jam risk assessment method of claim 1 or 3, wherein, The prediction results of all risk indicators are weighted and fused to obtain the geological risk grade of the tunneling section, and the specific process is as follows: Based on the standard cloud model corresponding to each risk indicator, the cloud drops of each risk indicator are obtained according to the geological parameters and the geophysical parameters of the tunneling section, the cloud drops of all risk indicators are weighted and fused to obtain a comprehensive identification cloud chart, and the geological risk grade of the tunneling section is determined by comparing the comprehensive identification cloud chart with the cloud charts of the standard risk grades and using the maximum membership degree principle. In the process of constructing the machine jam risk prediction model, a joint contribution degree evaluation mechanism is introduced, the joint contribution degrees of the geological features and the torque features are calculated based on the gain values of the geological features and the coverage values of the torque features, and it is determined whether the joint contribution degrees of the geological features and the torque features exceed a set threshold value.

5. The geological risk-torque prediction driven jam risk assessment method of claim 1, wherein, The cross features generated by the geological features and the torque features are included in the candidate split feature set of the current tree structure to participate in tree construction if the joint contribution degrees of the geological features and the torque features exceed the set threshold value.

6. The geological risk-torque prediction driven jam risk assessment method of claim 1, wherein, The machine jam risk prediction model is constructed by using the geological risk grade and the corresponding risk indicators and the tunneling machine operation risk grade and the corresponding tunneling machine main control parameters.

7. The geological risk-torque prediction driven jam risk assessment method of claim 2, wherein, The geological parameters of the fault fracture zone include fault width, rock mass integrity coefficient, joint spacing, water seepage amount and rock compressive strength; the geological parameters of the karst include spatial distribution and volume of karst cave, fracture water pressure, permeability coefficient, distribution proportion of carbonate rock and water inflow; the geological parameters of the alteration zone include rock chemical composition change, rock mechanical property, swelling index, fracture development degree and water seepage and water inflow; The geophysical parameters include resistivity, relaxation time, P-wave velocity and polarizability; The tunneling machine main control parameters include TBM cutterhead diameter, cutterhead thickness and cutterhead opening rate.

8. A geological risk-torque prediction driven jam risk assessment system, characterized in that, The method comprises the following steps: The data collection module is configured to obtain the geological parameters, the geophysical parameters and the tunneling machine main control parameters of the tunneling section; The geological risk evaluation module is configured to assign initial weights of risk indicators to the geological parameters and the geophysical parameters according to adverse geology and corresponding geological parameters and geophysical parameters, divide geology into multiple risk grades, and construct a standard cloud model; Based on the standard cloud model, the weights of the geological parameters and / or the geophysical parameters are dynamically adjusted according to the fluctuation and risk sensitivity of the geological parameters and / or the geophysical parameters in the tunneling process, and the geological risk grade of the tunneling section is obtained by weighting and fusing the prediction results of all risk indicators; The torque prediction module is configured to obtain a tunneling machine operation risk grade by using a trained torque prediction model based on the geological parameters, the geophysical parameters and the tunneling machine main control parameters; and the processing process of the torque prediction model on the geological parameters, the geophysical parameters and the tunneling machine main control parameters is as follows: The long-term dependence relationship of the input data of the time series is extracted by using a long short-term memory neural network; The weighted context vector is obtained by weighting and aggregating the hidden states of the long short-term memory neural network by using an attention mechanism. A geological disturbance perception gating mechanism is introduced, and a disturbance feature is spliced with the weighted context vector to obtain a disturbance adjustment factor; The disturbance adjustment factor is used to weight and fuse the torque value predicted by the trained attention mechanism-based long short-term memory network, and the historical torque sliding mean or weighted trend value is combined to obtain the prediction result of the torque prediction model; The double-drive evaluation module is configured to fuse the geological risk grade and the tunneling machine operation risk grade based on the machine jamming risk prediction model constructed by the gradient boosting tree to obtain a machine jamming risk evaluation result.

Images