Deep underground engineering large deformation real-time monitoring and control method based on deep learning

Abstract

The invention discloses a deep underground engineering large deformation real-time monitoring and control method based on deep learning. The method comprises the steps of deployment of a monitoring system, preprocessing of monitoring data, construction of a neural network model, training of the neural network model, output of a prediction result, generation of a control instruction and quantitative evaluation of a control effect. According to synchronously acquired multi-source information such as surrounding rock strain, temperature, seepage pressure and micro-seismic disturbance energy, a deep learning model based on physical mechanism constraint is introduced, so that the reliability and generalization ability of a prediction result are remarkably improved; intelligent identification of surrounding rock stability assessment and failure modes is realized, a control instruction is automatically generated, and an implementation effect is quantitatively assessed.

Description

Technical Field

[0001] The present invention belongs to the field of underground engineering technology and relates to a real-time monitoring and control method for large deformation of deep underground engineering based on deep learning. Background Art

[0002] Deep underground projects such as deep-buried tunnels and mine tunnels are often located in complex geological environments with high ground stress, high temperature, and high seepage pressure. Large deformations of the surrounding rock are prone to occur during construction. This type of deformation may not only damage the support structure and cause the cavern to converge beyond the allowable range, but may also cause the expansion of loose areas in the surrounding rock, further causing stress redistribution and increased deformation, and even triggering major engineering accidents such as landslides. In addition, the development and penetration of internal cracks in the rock mass during large deformation will significantly weaken the surrounding rock's impermeability and increase the risk of high-pressure water inrush. It will also significantly increase support costs, delay construction schedules, and adversely affect the long-term stability of the project. Therefore, achieving real-time monitoring and intelligent control of large deformations in the surrounding rock of deep projects is a key technical problem that urgently needs to be overcome in the current field of geotechnical engineering.

[0003] At present, deep underground projects still face the following prominent problems in the monitoring and control of large deformations: 1) The causes of large deformation of surrounding rocks are complex, involving the coupling of multiple fields such as stress field, seepage field, temperature field and construction disturbance field, such as the weakening of rock strength by high water pressure, stress concentration caused by blasting disturbance, and rock volume effect caused by temperature change. However, existing monitoring methods are mostly limited to the collection of a single physical quantity, which makes it difficult to fully reflect the true state of the surrounding rock and easily leads to early warning delays or misjudgments; 2) Existing prediction models are mostly data-driven and lack the integration of physical mechanisms such as rock mechanics and seepage mechanics. In the context of complex deep geology and a lack of monitoring data, such models are prone to overfitting, and the prediction results often deviate from the actual physical laws and lack reliability; 3) In actual projects, the interpretation of monitoring data and subsequent control strategies mostly rely on engineering experience, lack quantitative decision support and real-time feedback mechanisms, and the effectiveness of the implementation of control measures is difficult to objectively evaluate.

[0004] In summary, existing technologies have yet to achieve the deep integration of multi-physics information and the effective embedding of mechanical mechanisms, and also lack the ability to achieve closed-loop control based on real-time data. Therefore, it is necessary to develop an intelligent method that integrates real-time multi-source information perception, physical mechanism drive, and dynamic closed-loop regulation to accurately predict and actively control large deformations of surrounding rock in deep underground engineering projects. Summary of the Invention

[0005] In response to the shortcomings of existing technologies, the present invention proposes a real-time monitoring and control method for large deformation of deep underground projects based on deep learning. By combining deep learning methods and closed-loop control strategies, real-time, accurate, and intelligent perception and control of large deformation conditions of surrounding rock during the construction of deep underground projects are achieved, which is conducive to improving construction safety, efficiency, quality, and intelligence.

[0006] The deep learning-based real-time monitoring and control method for large deformation of deep underground engineering includes the following steps: S1. Deployment of monitoring system: A distributed monitoring system is deployed in the surrounding rock area of ​​deep underground projects, including a distributed fiber optic sensing module and a microseismic monitoring module. The distributed fiber optic sensing module uses a fiber grating array as a sensing element and is arranged in a grid pattern along the surrounding rock surface and deep boreholes to collect surrounding rock strain, temperature, and seepage pressure data. The microseismic monitoring module is composed of several three-component acceleration sensors arranged in the area affected by the project excavation. It is used to monitor construction-induced microseismic events in real time and record the occurrence time, disturbance energy density, and three-dimensional spatial coordinates. S2. Preprocessing of monitoring data: The raw data collected in step S1 are preprocessed, including noise reduction, normalization, time-space alignment and coupling correction, and then the surrounding rock strain is converted to e ( x , y , z , t ),temperature T ( x , y , z , t ), seepage pressure P ( x , y , z , t ) and disturbance energy density E ( x , y , z , t ) are combined on a unified spatiotemporal grid point to construct a spatiotemporally aligned multidimensional feature vector; The dimension of this vector is equal to the sum of the quantities of all physical field parameters contained in it at a specific time and space point, and is directly used as the input of the neural network model in step S3; Preferably, the raw data is preprocessed in step S2, specifically comprising the following steps: S201. Data Noise Reduction and Quality Improvement: Filter and de-noise the raw data to eliminate high-frequency noise introduced by construction vibration and electromagnetic interference, thereby improving the data signal-to-noise ratio. Furthermore, based on statistical principles, abnormal data is identified and replaced, outliers that significantly deviate from the data distribution pattern are removed, and interpolation algorithms are used to fill in data gaps to ensure the continuity of the time series. S202, data normalization: Map the values ​​of the original data to a unified [0,1] interval, thereby eliminating the scale differences caused by different physical dimensions and ensuring that all monitoring parameters have comparable weights in the deep learning model; the normalized parameter values X norm The calculation expression is:

[0007] in, X is the original parameter value, X max and X min are the global maximum and minimum values ​​of the parameter recorded in the historical engineering database and the current monitoring period respectively; S203, time-space alignment and data fusion: the monitoring data after noise reduction, repair and normalization are aligned according to their precise time-space coordinates ( x , y , z , t ) performs spatial registration and temporal synchronization to ensure that data from different sensor sources correspond to each other at the same spatial location and the same time point, forming a temporally and spatially consistent multidimensional data set; S204, data coupling correction: The surrounding rock strain after processing in steps S201 to S203 is e 0 ( x , y , z , t ) is corrected, and the corrected surrounding rock strain e ( x , y , z , t ) satisfies the following expression:

[0008] in, β 1 、 β 2 and β 3are the coupling coefficients that characterize the contribution of temperature, seepage pressure, and disturbance energy density to the strain of the surrounding rock. These coefficients need to be calibrated through indoor physical and mechanical tests on representative rock samples. S205 , constructing a multidimensional feature vector: combining the data processed in steps S201 to S204 on a unified spatiotemporal grid point to form the multidimensional feature vector.

[0009] S3. Construction of neural network model: A neural network model based on deep learning and physical constraints is constructed to fuse multi-physics field monitoring data, predict surrounding rock conditions, and generate control instructions; the neural network model includes an input layer, a feature extraction layer, a physical constraint layer, and an output layer connected in sequence; The input layer is configured to receive the multidimensional feature vector generated in step S2, adopts a fully connected structure, and the number of neurons is consistent with the dimension of the multidimensional feature vector, and is used to map the input data to a high-dimensional feature space; The feature extraction layer is configured to automatically learn and fuse the complex correlation features of the surrounding rock state in spatial and temporal dimensions from the multidimensional feature vector; this layer adopts a hybrid neural network architecture, including a multimodal spatial feature extractor, a time evolution feature extractor and a feature fusion module; the multimodal spatial feature extractor is implemented by a three-dimensional convolutional neural network, and is used to extract spatial correlation features and coupling patterns between monitoring points at different spatial positions, as well as between multi-physical field parameters such as strain, temperature, seepage pressure, and disturbance energy; the time evolution feature extractor is implemented by a long short-term memory network, and is used to capture the dynamic evolution law and historical dependency of the multidimensional feature vector in the time series; the feature fusion module is implemented by an attention mechanism, and is used to perform adaptive weighted fusion of the spatial features output by the multimodal spatial feature extractor and the temporal features output by the time evolution feature extractor, and to enhance the attention weight of feature information in high-risk areas; The physical constraint layer is configured to embed the basic principles of geotechnical mechanics into the model's learning process, ensuring that the model's prediction results conform to physical laws and enhance generalization. This layer is located between the feature extraction layer and the output layer, and implements physical constraints by introducing physical equation-based constraints into the model training loss function. The physical equation constraints include: constraints based on the effective stress principle to ensure that the predicted surrounding rock stress state and pore water pressure conform to the effective stress relationship; constraints based on the rock constitutive relationship to ensure that the predicted stress-strain path conforms to the basic mechanical behavior of the rock mass; and constraints based on the energy conservation principle to ensure that the energy released by microseismic events is consistent with the energy consumed by surrounding rock deformation and failure. The output layer is configured to generate the final prediction results and control parameters based on the fusion features extracted by the feature extraction layer and the requirements of the physical constraint layer. This layer uses a linear activation function, and its neuron output values ​​correspond to: the predicted maximum deformation value of the key position of the surrounding rock within the next 24 hours; the stability assessment index that characterizes the overall stability of the current surrounding rock; the classification probability indicating the potential failure mode of the current surrounding rock, including compression type, shear type, and seepage induced type; and the quantitative control parameters used to guide subsequent control measures. S4. Training of neural network model: The total loss function of the neural network model L total Due to data loss L data Loss of physical constraints L phys Composition, that is, satisfying the following expression:

[0010] in, L data It is the mean square error between the model prediction value and the actual monitoring value, which ensures the model's ability to fit the monitoring data; L phys The constraint terms are constructed by introducing the effective stress principle, rock constitutive relations and energy conservation principle as the weighted sum of physical constraint error terms, and are incorporated into the total loss function in the form of a weighted sum to ensure that the model output conforms to physical common sense. m 1 and m 2 is the weight coefficient, and m 1 + m 2 Satisfy = 1, the specific value is determined by cross-validation; The training sample set consists of historical engineering measured data and virtual samples based on physical mechanism simulation to enhance the model's generalization ability under extreme working conditions. The model training uses the Adam optimization algorithm, with an initial learning rate set to 0.001 and a learning rate decay strategy introduced to prevent overfitting. S5. Output of prediction results: The neural network model trained in step S4 performs forward calculation on the multi-dimensional feature vector processed in step S2, and outputs deformation prediction value, stability evaluation index and surrounding rock failure mode; The deformation prediction value is the maximum deformation prediction value of the key position of the surrounding rock in the next 24 hours. U max ; The stability evaluation index I s (x , y , z , t ) is a quantitative index reflecting the stability of the surrounding rock, which satisfies the following expression:

[0011] in, e p is the peak strain of the rock mass; P p is the critical seepage pressure; The surrounding rock failure mode is output through the Softmax classifier deployed in the output layer, which outputs the probability distribution of three types of failure modes: extrusion failure, shear failure and seepage induced failure, which are respectively P sq 、 P sh and P se ; Accordingly, the probability of the highest probability failure mode is P max =max{ P sq , P sh , P se}; S6. Generation of control instructions: when U max ≥[ U max ]、 I s ( x , y , z , t )≤[ I s ]or P max ≥[ P ], triggers the generation of control instructions; among them, [ U max ] is the preset critical deformation threshold, which is set according to the engineering geological report; [ I s ] is the stability safety threshold, which is determined according to the safety requirements of deep underground engineering design; [ P ] is the probability threshold of the failure mode; Preferably, in step S6, when P max = P sqWhen , it reflects that the dominant type of surrounding rock failure mode is extrusion failure, and the control parameters generated by the neural network model are conducive to improving the radial constraint force of the surrounding rock; when P max = P sh When , it reflects that the dominant type of surrounding rock failure mode is shear failure, and the control parameters generated by the neural network model are beneficial to suppressing the structural surface slip; when P max = P se When , it reflects that the dominant type of surrounding rock failure mode is seepage induced failure, and the neural network model generates control parameters that are beneficial to reducing pore water pressure; the control parameters that are beneficial to improving the radial constraint force of the surrounding rock include the increase in anchor preload, the increase in grouting pressure and the increase in support structure layout density; the control parameters that are beneficial to suppressing structural surface slip include the increase in anchor locking force and the increase in the thickness of the surrounding rock surface spray layer; the control parameters that are beneficial to reducing pore water pressure include the increase in drainage hole flow and the increase in waterstop grouting pressure.

[0012] S7. Quantitative evaluation of control effect: After implementing the control measures according to the control instructions of step S6, the surrounding rock strain, temperature, seepage pressure and microseismic energy density data are re-collected and pre-processed according to steps S1 and S2 to construct the control effect index. I t , the calculation expression is as follows:

[0013] Among them, Δ e is the strain change value, which indicates the change in strain after the implementation of the control measures relative to the strain before the implementation of the control measures; Δ E is the change in microseismic energy density, which represents the change in the average microseismic energy density within 2 hours after the implementation of the control measures relative to the average microseismic energy density within 2 hours before the implementation of the control measures; Δ P is the seepage pressure change value, which means the change value of the seepage pressure after the implementation of the control measures relative to the seepage pressure before the implementation of the control measures; e pre 、 E pre 、 P pre are the surrounding rock strain, microseismic energy density and seepage pressure before the implementation of control measures; α 1 、 α 2 and α 3is the weight coefficient, and its specific value is predetermined by the analytic hierarchy process or expert scoring method based on engineering geological conditions, dominant type of surrounding rock failure mode and engineering practice experience, and α 1 + α 2 + α 3 =1.

[0014] Preferably, in step S7, when the control effect index I t When <-0.5, it is determined that the current surrounding rock control measures have achieved the expected effect. At this time, the multi-source monitoring data and the executed control instructions in this period are integrated into sample pairs and stored in the model training database for subsequent incremental learning. During the incremental learning process, in order to maintain the stability of the existing features of the model, the parameters of the convolutional neural network and the long short-term memory network layer are fixed unchanged, and only the fully connected weights of the output layer are fine-tuned and iteratively updated.

[0015] In summary, compared with the existing technology, the present invention has the following advantages: to address the problems existing in the accurate prediction and active control of large deformation of surrounding rock in deep underground projects, such as low accuracy of monitoring data interpretation, inconsistent prediction results of surrounding rock deformation with physical laws, subjective estimation of control parameters and lack of closed-loop control, a deep learning-based real-time monitoring and control method for large deformation in deep underground projects is proposed. The method includes the deployment of the monitoring system, preprocessing of monitoring data, construction of a neural network model, training of the neural network model, output of prediction results, generation of control instructions and quantitative evaluation of control effects, achieving the following breakthrough improvements: 1) Achieved deep integration of multi-physics field monitoring data: By deploying distributed fiber optic sensing and microseismic monitoring modules, multi-source information such as surrounding rock strain, temperature, seepage pressure, and disturbance energy density is simultaneously collected. After preprocessing and coupling correction, a multi-dimensional feature vector with unified time and space is constructed. This overcomes the limitations of traditional single-physics field monitoring and provides a comprehensive and reliable data foundation for surrounding rock state identification and deformation prediction. 2) Improving the predictive reliability of neural network models: By introducing a deep learning model based on physical mechanism constraints, the reliability and generalization of prediction results are significantly improved. By embedding the basic principles of geotechnical mechanics into the neural network, the model is not only data-driven but also conforms to physical laws, effectively avoiding overfitting or unintuitive prediction outputs caused by data scarcity under extreme working conditions. 3) It achieves intelligent identification of surrounding rock stability assessment and failure modes: the model can output future key deformation, stability index, and classification probability of typical failure modes, providing a quantitative, multi-dimensional basis for control decision-making, overcoming the shortcomings of traditional methods that rely on manual experience and lack quantitative support; 4) Achieved precise matching of control measures: A closed-loop intelligent control mechanism was constructed that automatically generates control instructions based on prediction results and quantitatively evaluates implementation effects. For different dominant damage modes, quantitative control parameters with clear physical meaning and engineering guidance value were output, and the control effects were objectively and quantitatively evaluated. 5) The model has online learning and optimization capabilities: By introducing the control effect index and incremental learning mechanism, the system can accumulate on-site control samples and continuously optimize the output layer weights, thereby adapting to geological conditions and construction dynamic changes, and enhancing the long-term applicability and stability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of the method for real-time monitoring and control of large deformation in deep underground engineering based on deep learning described in the present invention. DETAILED DESCRIPTION

[0017] The following is a detailed description of the specific embodiments of the present invention in conjunction with the accompanying drawings. It should be understood that the specific embodiments given here are only used to illustrate and explain the present invention and cannot be used to limit the present invention.

[0018] This application discloses Figure 1 The deep learning-based real-time monitoring and control method for large deformation of deep underground engineering shown in the figure includes the following steps: S1. Deployment of monitoring system: A distributed monitoring system is deployed in the surrounding rock area of ​​deep underground engineering, including a distributed fiber optic sensing module and a microseismic monitoring module; the distributed fiber optic sensing module uses a corrosion-resistant and electromagnetic interference-resistant fiber grating array as a sensing element, and is arranged in a grid manner along the surrounding rock surface and deep boreholes to collect surrounding rock strain, temperature and seepage pressure data; the microseismic monitoring module is composed of a number of three-component acceleration sensors arranged in the engineering excavation influence area, which is used to monitor construction-induced microseismic events in real time and record the time of occurrence, disturbance energy density and its three-dimensional spatial coordinates; the monitoring range of the distributed monitoring system covers the key deformation areas of the surrounding rock of deep underground engineering, the engineering excavation influence area and the potential risk sections identified in the geological survey report, including but not limited to the arch, side wall, the range of 1 times the tunnel diameter in front of the tunnel face and the potential slip section of the surrounding rock.

[0019] S2. Preprocessing of monitoring data: Preprocess the raw data collected in step S1, including noise reduction, normalization, time-space alignment and coupling correction, and then convert the surrounding rock strain e ( x , y , z , t ),temperature T ( x , y , z ,t ), seepage pressure P ( x , y , z , t ) and disturbance energy density E ( x , y , z , t ) are combined at a unified space-time grid point to construct a space-time aligned multidimensional feature vector; the dimension of this vector is equal to the sum of the number of all physical field parameters contained at a specific space-time point, which is directly used as the input of the neural network model in step S3. In specific implementation, the raw data is preprocessed in step S2, including the following steps: S201. Data noise reduction and quality improvement: Filter and reduce noise on the original data to eliminate high-frequency noise introduced by construction vibration and electromagnetic interference, and improve the data signal-to-noise ratio; at the same time, identify and replace abnormal data based on statistical principles, remove outliers that significantly deviate from the data distribution law, and use interpolation algorithms to fill in data gaps to ensure the continuity of the time series; specifically, according to 3 s The criterion eliminates data that differs from the data mean by more than 3 standard deviations, and then uses time series linear interpolation to repair, that is, the average value of the data adjacent to the vacant data in the time series is used to fill the vacant data.

[0020] S202, data normalization: Map the values ​​of the original data to a unified [0,1] interval, thereby eliminating the scale differences caused by different physical dimensions and ensuring that all monitoring parameters have comparable weights in the deep learning model; the normalized parameter values X norm The calculation expression is: (1) in, X is the original parameter value, X max and X min They are the global maximum and minimum values ​​of the parameter recorded in the historical engineering database and the current monitoring period, respectively.

[0021] S203, time-space alignment and data fusion: the monitoring data after noise reduction, repair and normalization are aligned according to their precise time-space coordinates ( x , y , z , t) to perform spatial registration and temporal synchronization, ensuring that data from different sensor sources correspond to each other at the same spatial location and time point, forming a temporally and spatially consistent multidimensional dataset. Specifically, all sensor coordinates are unified to the deep underground engineering construction coordinate system, and the sensor data is interpolated to a unified grid point. Furthermore, the sensor data is unified to the same sampling rate based on the data acquisition server clock.

[0022] S204, data coupling correction: Considering the cross interference of temperature, seepage pressure, and disturbance energy density on the surrounding rock strain, the surrounding rock strain processed by steps S201 to S203 is corrected. e 0 ( x , y , z , t ) to perform physical field coupling correction, and the corrected surrounding rock strain e ( x , y , z , t ) satisfies the following expression: (2) in, β 1 、 β 2 and β 3 are the coupling coefficients that characterize the contribution of temperature, seepage pressure and disturbance energy density to the surrounding rock strain. The coefficients need to be calibrated through indoor physical and mechanical tests of representative rock samples. In the specific implementation, the calibration results are β 1 =0.05, β 2 =0.12, β 3 =0.08.

[0023] S205. Constructing a multidimensional feature vector: The data processed in steps S201 to S204 are combined at a unified spatiotemporal grid to form a multidimensional feature vector. The dimension of this vector is equal to the sum of the number of all physical field parameters at a specific spatiotemporal point, and is directly used as input to the neural network model in step S3.

[0024] S3. Construction of a neural network model: Construct a neural network model based on deep learning and physical constraints to fuse multi-physical field monitoring data and predict surrounding rock conditions and generate control instructions; the neural network model includes an input layer, a feature extraction layer, a physical constraint layer, and an output layer connected in sequence.

[0025] In a specific implementation, the input layer is configured to receive the multidimensional feature vector generated in step S2, adopts a fully connected structure, and the number of neurons is consistent with the dimension of the multidimensional feature vector, and is used to map the input data to a high-dimensional feature space. The feature extraction layer is configured to automatically learn and fuse the complex correlation features of the surrounding rock state in the spatial and temporal dimensions from the multidimensional feature vector; this layer adopts a hybrid neural network architecture, including a multimodal spatial feature extractor, a time evolution feature extractor, and a feature fusion module; the multimodal spatial feature extractor is implemented by a three-dimensional convolutional neural network, and is used to extract spatial correlation features and coupling patterns between monitoring points at different spatial locations, as well as between multiple physical field parameters such as strain, temperature, seepage pressure, and disturbance energy; the time evolution feature extractor is implemented by a long short-term memory network, and is used to capture the dynamic evolution law and historical dependency of the multidimensional feature vector in the time series; the feature fusion module is implemented by an attention mechanism, and is used to perform adaptive weighted fusion of the spatial features output by the multimodal spatial feature extractor and the temporal features output by the time evolution feature extractor, and to enhance the attention weight of feature information in high-risk areas. The physical constraint layer is configured to embed the basic principles of geotechnical mechanics into the learning process of the model to ensure that the model prediction results conform to physical laws and enhance generalization. This layer is located between the feature extraction layer and the output layer, and physical constraints are implemented by introducing constraints based on physical equations into the model training loss function. The physical equation constraints include: constraints based on the effective stress principle L 1 , ensuring that the predicted surrounding rock stress state and pore water pressure conform to the effective stress relationship; based on the constraints of rock constitutive relationship L 2 , ensuring that the predicted stress-strain path conforms to the basic mechanical behavior of the rock mass; constraints based on the principle of energy conservation L 3 , ensuring that the energy released by microseismic events is consistent with the energy consumed by deformation and destruction of the surrounding rock.

[0026] The output layer is configured to generate the final prediction results and control parameters based on the fusion features extracted by the feature extraction layer and the requirements of the physical constraint layer. This layer uses a linear activation function, and its neuron output values ​​correspond to: the predicted value of the maximum deformation at the key position of the surrounding rock within the next 24 hours; the stability assessment index that characterizes the overall stability of the current surrounding rock; the classification probability indicating the potential failure mode of the current surrounding rock, including compression type, shear type and seepage induced type; and the quantitative control parameters used to guide subsequent control measures.

[0027] S4. Training of the neural network model: the total loss function of the neural network model L total Due to data loss L dataLoss of physical constraints L phys Composition, that is, satisfying the following expression: (3) in, L data It is the mean square error between the model prediction value and the actual monitoring value, which ensures the model's ability to fit the monitoring data; L phys The constraint terms are constructed by introducing the effective stress principle, rock constitutive relations and energy conservation principle as the weighted sum of physical constraint error terms, and are incorporated into the total loss function in the form of a weighted sum to ensure that the model output conforms to physical common sense. m 1 and m 2 is the weight coefficient, and m 1 + m 2 Satisfy = 1, the specific value is determined by cross-validation; in the specific implementation, L phys = d 1 · L 1 + d 2 · L 2 + d 3 · L 3 ,in d 1 、 d 2 and d 3 are all weight coefficients and satisfy d 1 + d 2 + d 3 =1.

[0028] The training sample set consists of historical engineering measured data and virtual samples based on physical mechanism simulation to enhance the generalization ability of the model under extreme working conditions; the model training adopts the Adam optimization algorithm, the initial learning rate is set to 0.001, and a learning rate decay strategy is introduced to suppress overfitting.

[0029] S5. Output of prediction results: The neural network model trained in step S4 performs forward calculation on the multi-dimensional feature vector processed in step S2, and outputs deformation prediction value, stability assessment index and surrounding rock failure mode; The deformation prediction value is the maximum deformation prediction value of the key position of the surrounding rock in the next 24 hours. U max ; The stability evaluation index I s ( x , y , z , t ) is a quantitative index reflecting the stability of the surrounding rock, which satisfies the following expression: (4) in, e p is the peak strain of the rock mass; P p is the critical seepage pressure; I s The value range is [0,1]. The closer the value is to 1, the higher the stability of the surrounding rock is, and the closer the value is to 0, the worse the stability of the surrounding rock is. The surrounding rock failure mode is output through the Softmax classifier deployed in the output layer, which outputs the probability distribution of three types of failure modes: extrusion failure, shear failure and seepage induced failure, which are respectively P sq 、 P sh and P se ; Accordingly, the probability of the highest probability failure mode is P max =max{ P sq , P sh , P se}.

[0030] S6. Generation of control instructions: When U max ≥[ U max ]、 I s ( x , y , z , t )≤[ I s ]or P max ≥[ P ], the control instruction is generated; among them, [ U max ] is the preset critical deformation threshold, which is set according to the engineering geological report; [ I s] is the stability safety threshold, which is determined according to the safety requirements of deep underground engineering design; [ P ] is the failure mode probability threshold.

[0031] when P max = P sq When , it reflects that the dominant type of surrounding rock failure mode is extrusion failure, and the control parameters generated by the neural network model are conducive to improving the radial constraint force of the surrounding rock; when P max = P sh When , it reflects that the dominant type of surrounding rock failure mode is shear failure, and the control parameters generated by the neural network model are beneficial to suppressing the structural surface slip; when P max = P se , it reflects that the dominant type of surrounding rock failure mode is seepage-induced failure, and the control parameters generated by the neural network model are conducive to reducing the pore water pressure.

[0032] In the specific implementation, in step S6, the control parameters that are conducive to improving the radial constraint force of the surrounding rock include the increase in the anchor rod preload, the increase in the grouting pressure and the increase in the support structure arrangement density; the control parameters that are conducive to suppressing structural surface slip include the increase in the anchor cable locking force and the increase in the thickness of the sprayed layer on the surrounding rock surface; the control parameters that are conducive to reducing the pore water pressure include the increase in the drainage hole flow rate and the increase in the water-stop grouting pressure.

[0033] S7. Quantitative evaluation of control effect: After implementing the control measures according to the control instructions of step S6, re-collect the surrounding rock strain, temperature, seepage pressure and microseismic energy density data according to steps S1 and S2 and perform pre-processing to construct a control effect index. I t , the calculation expression is as follows: (5) Among them, Δ e is the strain change value, which indicates the change in strain after the implementation of the control measures relative to the strain before the implementation of the control measures; Δ E is the change in microseismic energy density, which represents the change in the average microseismic energy density within 2 hours after the implementation of the control measures relative to the average microseismic energy density within 2 hours before the implementation of the control measures; Δ P is the seepage pressure change value, which means the change value of the seepage pressure after the implementation of the control measures relative to the seepage pressure before the implementation of the control measures; e pre 、 E pre 、 Ppre are the surrounding rock strain, microseismic energy density and seepage pressure before the implementation of control measures; α 1 、 α 2 and α 3 is the weight coefficient, and its specific value is predetermined by the analytic hierarchy process or expert scoring method based on engineering geological conditions, dominant type of surrounding rock failure mode and engineering practice experience, and α 1 + α 2 + α 3 =1.

[0034] In specific implementation, when the control effect index I t When <-0.5, it is determined that the current surrounding rock control measures have achieved the expected effect. At this time, the multi-source monitoring data and the executed control instructions in this period are integrated into sample pairs and stored in the model training database for subsequent incremental learning. During the incremental learning process, in order to maintain the stability of the existing features of the model, the parameters of the convolutional neural network and the long short-term memory network layer are fixed unchanged, and only the fully connected weights of the output layer are fine-tuned and iteratively updated.

[0035] The above is a description of one or more embodiments of the present invention, and while the description is relatively specific and detailed, it should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the spirit of the present invention, and these modifications and improvements fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A deep learning-based real-time monitoring and control method for large deformation in deep underground engineering, characterized by: The following steps are involved: S1. Deployment of monitoring system: Deploy a distributed monitoring system in the surrounding rock area of ​​deep underground projects to collect surrounding rock strain, temperature, and seepage pressure data. It also monitors construction-induced microseismic events in real time and records the time of occurrence, disturbance energy density, and its three-dimensional spatial coordinates. S2. Preprocessing of monitoring data: Preprocess the raw data collected in step S1, including noise reduction, normalization, time-space alignment and coupling correction, and then convert the surrounding rock strain ε ( x , y , z , t ),temperature T ( x , y , z , t ), seepage pressure P ( x , y , z , t ) and disturbance energy density E ( x , y , z , t ) are combined on a unified spatiotemporal grid point to construct a spatiotemporally aligned multidimensional feature vector; S3. Construction of neural network model: Build a neural network model based on deep learning and physical constraints to integrate multi-physics field monitoring data, predict surrounding rock conditions, and generate control instructions; S4. Training of the neural network model: training the neural network model, and the total loss function L total Due to data loss L data Loss of physical constraints L phys composition; S5. Output of prediction results: The neural network model trained in step S4 performs forward calculation on the multi-dimensional feature vector processed in step S2, and outputs deformation prediction value, stability assessment index and surrounding rock failure mode; S6. Generation of control instructions: When U max ≥[ U max ]、 I s ( x , y , z , t )≤[ I s ]or P max ≥[ P ], triggers the generation of control instructions; among them, [ U max ] is the preset critical deformation threshold, which is set according to the engineering geological report; [ I s ] is the stability safety threshold, which is determined according to the safety requirements of deep underground engineering design; [ P ] is the probability threshold of the failure mode; S7. Quantitative evaluation of control effect: After implementing the control measures according to the control instructions of step S6, re-collect the surrounding rock strain, temperature, seepage pressure and microseismic energy density data according to steps S1 and S2 and perform pre-processing to construct a control effect index. I t .

2. The method for real-time monitoring and control of large deformation in deep underground engineering based on deep learning according to claim 1 is characterized in that: In step S1, the distributed monitoring system includes a distributed fiber optic sensing module and a microseismic monitoring module; the distributed fiber optic sensing module uses a fiber grating array as a sensing element, which is arranged in a grid manner along the surface of the surrounding rock and deep boreholes to collect surrounding rock strain, temperature and seepage pressure data; the microseismic monitoring module is composed of a number of three-component acceleration sensors arranged in the area affected by the engineering excavation, which is used to monitor construction-induced microseismic events in real time and record the occurrence time, disturbance energy density and its three-dimensional spatial coordinates.

3. The method for real-time monitoring and control of large deformation in deep underground engineering based on deep learning according to claim 1 is characterized in that: In step S2, the raw data is preprocessed, which specifically includes the following steps: S201. Data Noise Reduction and Quality Improvement: Filter and de-noise the raw data to eliminate high-frequency noise introduced by construction vibration and electromagnetic interference, thereby improving the data signal-to-noise ratio. Furthermore, based on statistical principles, abnormal data is identified and replaced, outliers that significantly deviate from the data distribution pattern are removed, and interpolation algorithms are used to fill in data gaps. S202, data normalization: Map the values ​​of the original data to a unified [0,1] interval, thereby eliminating the scale differences caused by different physical dimensions and ensuring that all monitoring parameters have comparable weights in the deep learning model; the normalized parameter values X norm The calculation expression is: in, X is the original parameter value, X max and X min are the global maximum and minimum values ​​of the parameter recorded in the historical engineering database and the current monitoring period respectively; S203, time-space alignment and data fusion: the monitoring data after noise reduction, repair and normalization are aligned according to their precise time-space coordinates ( x , y , z , t ) performs spatial registration and temporal synchronization to ensure that data from different sensor sources correspond to each other at the same spatial location and the same time point, forming a temporally and spatially consistent multidimensional data set; S204, data coupling correction: The surrounding rock strain after processing in steps S201 to S203 is ε 0 ( x , y , z , t ) is corrected, and the corrected surrounding rock strain ε ( x , y , z , t ) satisfies the following expression: in, β 1 、 β 2 and β 3 are the coupling coefficients that characterize the contribution of temperature, seepage pressure, and disturbance energy density to the strain of the surrounding rock. These coefficients need to be calibrated through indoor physical and mechanical tests on representative rock samples. S205 , constructing a multidimensional feature vector: combining the data processed in steps S201 to S204 on a unified spatiotemporal grid point to form the multidimensional feature vector.

4. The method for real-time monitoring and control of large deformation in deep underground engineering based on deep learning according to claim 1 is characterized in that: In step S3, the neural network model includes an input layer, a feature extraction layer, a physical constraint layer, and an output layer connected in sequence; the input layer is configured to receive the multidimensional feature vector generated in step S2, adopts a fully connected structure, and the number of neurons is consistent with the dimension of the multidimensional feature vector, and is used to map the input data to a high-dimensional feature space; The feature extraction layer is configured to automatically learn and fuse the complex correlation features of the surrounding rock state in spatial and temporal dimensions from the multidimensional feature vector; this layer adopts a hybrid neural network architecture, including a multimodal spatial feature extractor, a time evolution feature extractor and a feature fusion module; the multimodal spatial feature extractor is implemented by a three-dimensional convolutional neural network, and is used to extract spatial correlation features and coupling patterns between monitoring points at different spatial positions, as well as between multi-physical field parameters such as strain, temperature, seepage pressure, and disturbance energy; the time evolution feature extractor is implemented by a long short-term memory network, and is used to capture the dynamic evolution law and historical dependency of the multidimensional feature vector in the time series; the feature fusion module is implemented by an attention mechanism, and is used to perform adaptive weighted fusion of the spatial features output by the multimodal spatial feature extractor and the temporal features output by the time evolution feature extractor, and to enhance the attention weight of feature information in high-risk areas; The physical constraint layer is configured to embed the basic principles of geotechnical mechanics into the model's learning process, ensuring that the model's prediction results conform to physical laws and enhance generalization. This layer is located between the feature extraction layer and the output layer, and implements physical constraints by introducing physical equation-based constraints into the model training loss function. The physical equation constraints include: constraints based on the effective stress principle to ensure that the predicted surrounding rock stress state and pore water pressure conform to the effective stress relationship; constraints based on the rock constitutive relationship to ensure that the predicted stress-strain path conforms to the basic mechanical behavior of the rock mass; and constraints based on the energy conservation principle to ensure that the energy released by microseismic events is consistent with the energy consumed by surrounding rock deformation and failure. The output layer is configured to generate the final prediction results and control parameters based on the fusion features extracted by the feature extraction layer and the requirements of the physical constraint layer. This layer uses a linear activation function, and its neuron output values ​​correspond to: the predicted value of the maximum deformation at the key position of the surrounding rock within the next 24 hours; the stability assessment index that characterizes the overall stability of the current surrounding rock; the classification probability indicating the potential failure mode of the current surrounding rock, including compression type, shear type and seepage induced type; and the quantitative control parameters used to guide subsequent control measures.

5. The method for real-time monitoring and control of large deformation in deep underground engineering based on deep learning according to claim 1 is characterized in that: In step S4, the total loss function L total Satisfies the following expression: in, L data It is the mean square error between the model prediction value and the actual monitoring value, which ensures the model's ability to fit the monitoring data; L phys It is the weighted sum of physical constraint error terms, and the constraint terms are constructed by introducing the effective stress principle, rock constitutive relation and energy conservation principle; μ 1 and μ 2 is the weight coefficient, and μ 1 + μ 2 Satisfies = 1, and the specific value is determined by cross-validation; the training sample set consists of historical engineering measured data and virtual samples based on physical mechanism simulation to enhance the generalization ability of the model under extreme working conditions; the model training adopts the Adam optimization algorithm, the initial learning rate is set to 0.001, and the learning rate decay strategy is introduced to suppress overfitting.

6. The method for real-time monitoring and control of large deformation in deep underground engineering based on deep learning according to claim 1 is characterized in that: In step S5, the predicted deformation value is the maximum predicted deformation value of the key position of the surrounding rock within the next 24 hours. U max ; The stability evaluation index I s ( x , y , z , t ) is a quantitative index reflecting the stability of the surrounding rock, which satisfies the following expression: in, ε p is the peak strain of the rock mass; P p is the critical seepage pressure; The surrounding rock failure mode is output through the Softmax classifier deployed in the output layer, which outputs the probability distribution of three types of failure modes: extrusion failure, shear failure and seepage induced failure, which are respectively P sq 、 P sh and P se ; Accordingly, the probability of the highest probability failure mode is P max =max{ P sq , P sh , P se }.

7. The method for real-time monitoring and control of large deformation in deep underground engineering based on deep learning according to claim 1 is characterized in that: In step S6, when P max = P sq When , it reflects that the dominant type of surrounding rock failure mode is extrusion failure, and the control parameters generated by the neural network model are conducive to improving the radial restraint force of the surrounding rock; when P max = P sh When , it reflects that the dominant type of surrounding rock failure mode is shear failure, and the control parameters generated by the neural network model are conducive to suppressing the structural plane slip; when P max = P se When , it reflects that the dominant type of surrounding rock failure mode is seepage induced failure, and the neural network model generates control parameters that are beneficial to reducing pore water pressure; the control parameters that are beneficial to improving the radial constraint force of the surrounding rock include the increase in anchor preload, the increase in grouting pressure and the increase in support structure layout density; the control parameters that are beneficial to suppressing structural surface slip include the increase in anchor locking force and the increase in the thickness of the surrounding rock surface spray layer; the control parameters that are beneficial to reducing pore water pressure include the increase in drainage hole flow and the increase in waterstop grouting pressure.

8. The method for real-time monitoring and control of large deformation in deep underground engineering based on deep learning according to claim 1 is characterized in that: In step S7, the control effect index I t The calculation expression is as follows: Among them, Δ ε is the strain change value, which indicates the change in strain after the implementation of the control measures relative to the strain before the implementation of the control measures; Δ E is the change in microseismic energy density, which represents the change in the average microseismic energy density within 2 hours after the implementation of the control measures relative to the average microseismic energy density within 2 hours before the implementation of the control measures; Δ P is the seepage pressure change value, which means the change value of the seepage pressure after the implementation of the control measures relative to the seepage pressure before the implementation of the control measures; ε pre 、 E pre 、 P pre are the surrounding rock strain, microseismic energy density and seepage pressure before the implementation of control measures; α 1 、 α 2 and α 3 is the weight coefficient, and its specific value is predetermined by the analytic hierarchy process or expert scoring method based on engineering geological conditions, dominant type of surrounding rock failure mode and engineering practice experience, and α 1 + α 2 + α 3 =1; when the control effect index I t When <-0.5, it is determined that the current surrounding rock control measures have achieved the expected effect. At this time, the multi-source monitoring data and the executed control instructions in this period are integrated into sample pairs and stored in the model training database for subsequent incremental learning. During the incremental learning process, in order to maintain the stability of the existing features of the model, the parameters of the convolutional neural network and the long short-term memory network layer are fixed unchanged, and only the fully connected weights of the output layer are fine-tuned and iteratively updated.

Images