Intelligent tunnel supporting method and system based on machine learning
By constructing a machine learning model and integrating multi-source data for tunnel support decision-making, the problem of mismatch between support schemes and actual working conditions in traditional methods has been solved, realizing the intelligentization and safety improvement of tunnel support.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-24
- Publication Date
- 2026-03-13
AI Technical Summary
Traditional tunnel support methods lack quantitative and intelligent decision support, resulting in a mismatch between support schemes and actual working conditions. They are unable to cope with rapid changes in surrounding rock deformation, and multi-source data has not been effectively mined and utilized, forming data silos and making it impossible to achieve deep integration and intelligent analysis of multi-source heterogeneous data.
By integrating multi-source information from the entire construction process, a machine learning model is constructed to collect and preprocess data in real time, generate support adjustment plans, and iteratively update the model through decision optimization algorithms and feedback data to achieve real-time prediction and dynamic optimization of support needs.
It improves the safety and economy of support, enables proactive prediction and in-process control of support decisions, reduces reliance on simplification assumptions and human experience, and enhances the ability to predict and control construction risks.
Smart Images
Figure CN121659792A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of tunnel and underground engineering construction technology, and in particular relates to a tunnel intelligent support method and system based on machine learning. Background Technology
[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.
[0003] Tunnel support is a core component ensuring the safety and long-term stability of tunnel construction. The scientific nature of its design and dynamic adjustment directly impacts project safety, construction efficiency, and cost control. Traditional tunnel support design primarily relies on engineering analogy, theoretical calculations, and manual experience. During the design phase, the support scheme is predetermined based on geological survey reports and surrounding rock classification results. Although feedback is obtained through monitoring and measurement during the construction phase, it is only used to assess the safety of the support structure. There is a significant lag in the dynamic adjustment of the original design scheme, and it relies excessively on manual experience, lacking quantitative and intelligent decision support.
[0004] As tunnel engineering develops towards deeper, longer tunnels with more complex geological conditions, the limitations of traditional support methods are becoming increasingly prominent: First, the complexity and uncertainty of geological conditions make it easy for pre-set support schemes to be mismatched with actual working conditions, resulting in either insufficient support leading to safety risks or excessive support causing resource waste; Second, dynamic adjustments based on human experience are highly subjective and have a slow response speed, making it difficult to cope with rapid changes in surrounding rock deformation; Third, the massive amounts of multi-source data generated during construction have not been effectively mined and utilized, forming "data silos" that cannot provide continuous support for support decision optimization.
[0005] In existing technologies, although some information technology methods have been attempted to be applied to support decision-making, such as early warning systems based on monitoring data and numerical simulation inverse analysis methods, most of them are post-event verification or early warning, lacking the ability to predict in advance and make dynamic decisions during the event; moreover, the models rely on a large number of assumptions and simplifications, making it difficult to accurately reflect complex geomechanical behavior; and they have failed to achieve deep integration and intelligent analysis of multi-source heterogeneous data. Summary of the Invention
[0006] To overcome the shortcomings of the existing technologies, this invention provides a machine learning-based intelligent tunnel support method and system. By integrating multi-source information from the entire construction process, a machine learning model that can accurately reflect the relationship between the surrounding rock response and the support effect is constructed, enabling real-time prediction of support requirements and dynamic optimization of support schemes, thereby improving the safety, economy and intelligence of the support.
[0007] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of this invention provides a machine learning-based intelligent tunnel support method; A machine learning-based intelligent tunnel support method includes: Multi-source data is collected in real time during tunnel construction and preprocessed to obtain standardized time-series feature vectors; The time-series feature vector is input into the trained surrounding rock-support system state prediction model to obtain the prediction result of the system state. Based on the prediction results and preset safety thresholds and economic indicators, a recommended support adjustment scheme is generated through a decision optimization algorithm. The generated support adjustment plan is pushed to the construction end, and the new monitoring data after the decision is executed is used as feedback to incrementally learn the state prediction model.
[0008] As a further technical solution, the multi-source data includes geological information, support structure response information, and construction condition information; The geological information includes at least one of the following: digital information of geological sketches of the tunnel face, results of advanced geological prediction, rock strength parameters, and information on rock mass structure surfaces; The support structure response information includes at least one of the following: crown settlement, horizontal convergence, support concrete stress, anchor bolt axial force, and steel arch strain. The construction condition information includes at least one of the following: excavation progress, support construction delay distance, blasting parameters, or TBM tunneling parameters.
[0009] As a further technical solution, the preprocessing includes data cleaning, temporal alignment and feature engineering of the collected multi-source data to form a standardized temporal feature vector.
[0010] As a further technical solution, in the process of training the state prediction model of the surrounding rock-support system, a training dataset is constructed based on historical engineering data or numerical simulation data, and trained using a machine learning algorithm, wherein the machine learning algorithm is a long short-term memory network, a gated recurrent unit, a temporal convolutional network, or a combination thereof.
[0011] As a further technical solution, the decision optimization algorithm is a rule-based reasoning expert system, a multi-objective optimization algorithm, or a reinforcement learning agent.
[0012] As a further technical solution, when the decision optimization algorithm is a reinforcement learning agent, The output of the state prediction model is used as the environmental state perception. The set of support and adjustment actions is used as the action space of the intelligent agent; The reward function is based on an indicator that comprehensively reflects the safety and economy of support. Through training, the reinforcement learning agent learns a mapping strategy from states to optimal actions.
[0013] As a further technical solution, the support adjustment scheme includes at least one of whether to strengthen the support, adjust the support parameters, and optimize the timing of the operation; when using TBM construction, the support adjustment scheme also includes adjusting the TBM tunneling parameters and / or triggering advanced support.
[0014] A second aspect of the present invention provides a machine learning-based intelligent tunnel support system.
[0015] A machine learning-based intelligent tunnel support system includes: The data acquisition and preprocessing module is configured to: acquire multi-source data in real time during tunnel construction and preprocess it to obtain standardized time-series feature vectors; The state prediction module is configured to input the time-series feature vector into the trained surrounding rock-support system state prediction model to obtain the prediction result of the system state. The intelligent decision-making module is configured to generate a recommended support adjustment plan based on the prediction results and preset safety thresholds and economic indicators through a decision optimization algorithm. The decision execution and feedback module is configured to push the generated support adjustment plan to the construction end and use the new monitoring data after the decision execution as feedback to incrementally learn the state prediction model.
[0016] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of a machine learning-based intelligent tunnel support method as described in the first aspect of the present invention.
[0017] A fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of a machine learning-based intelligent tunnel support method as described in the first aspect of the present invention.
[0018] The above one or more technical solutions have the following beneficial effects: (1) This invention uses a machine learning model to predict the future state of the surrounding rock-support system, transforming support decision-making from a passive post-event response to an active pre-event prediction and in-event control, significantly improving the ability to predict and control construction risks. By deeply integrating multi-source heterogeneous data from geology, monitoring, and construction, and using a data-driven model to capture complex nonlinear relationships, it reduces the reliance of traditional methods on simplification assumptions and human experience, making the decision-making process more objective and accurate.
[0019] (2) This invention continuously iterates and updates the prediction model using feedback data after decision execution, enabling the system to adapt to changes in geological conditions and adjustments in construction techniques, and possessing the ability to continuously learn and optimize. By comprehensively considering safety thresholds and economic indicators in the decision generation stage, and using optimization algorithms to find safe, reliable, and cost-effective support schemes, it helps to reduce engineering costs and avoid over-support while ensuring construction safety.
[0020] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0021] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0022] Figure 1 This is a flowchart of the method in the first embodiment.
[0023] Figure 2 This is a system structure diagram of the second embodiment. Detailed Implementation
[0024] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0025] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.
[0026] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.
[0027] Example 1 This embodiment discloses a machine learning-based intelligent tunnel support method. It involves real-time acquisition and preprocessing of multi-source construction data to train a rock-support system state prediction model. Combining safety and economic indicators, an optimization algorithm generates a dynamic support scheme, and the model is iteratively updated using feedback data. This method enables proactive prediction and control of support decisions, improving risk predictability and decision accuracy, and balancing construction safety and economy.
[0028] Specifically, such as Figure 1 As shown, a machine learning-based intelligent tunnel support method includes: Step S1: Collect multi-source data in real time during the tunnel construction process and preprocess it to obtain a standardized time-series feature vector.
[0029] By deploying sensor networks and data acquisition systems at the tunnel construction site, multi-source data information such as geological information, support structure response information, and construction conditions can be obtained in real time.
[0030] For geological information, the system uses a digital geological logging system to acquire information such as lithology of the tunnel face, joint and fracture development, rock weathering degree, and groundwater conditions, and quantifies these into feature vectors. Simultaneously, it integrates the inferences from advanced geological forecasts.
[0031] For the support structure response information, sensors are deployed on the initial support to monitor the arch settlement, horizontal convergence, surface stress of the support concrete, axial force of the anchor bolts, strain of the steel arch frame, etc. in real time.
[0032] For construction status information, the excavation cycle advance, the distance between the working face and the latest support (support lag distance), blasting parameters, and TBM tunneling parameters are obtained from the construction management system.
[0033] Furthermore, the collected multi-source data undergoes preprocessing, including data cleaning, temporal alignment, and feature engineering. Data cleaning addresses outliers caused by sensor malfunctions, while interpolation and filling in missing values with data from preceding and following time points are employed.
[0034] Since different data collection frequencies are used, time-series alignment is performed based on a unified timestamp. By resampling or aggregating the data, it is ensured that the feature vectors at the same time contain all the necessary information.
[0035] In addition to the raw data, more meaningful derived features are constructed by calculating deformation rate, cumulative deformation, deformation acceleration, and the ratio of stress to material strength in the support structure. Finally, all features at each time t are combined into a standardized (e.g., Z-score normalization) feature vector. .
[0036] Step S2: Input the time-series feature vector into the trained surrounding rock-support system state prediction model to obtain the prediction result of the system state.
[0037] For the surrounding rock-support system state prediction model, a long short-term memory network is used as the prediction model in this embodiment. Gated recurrent units, temporal convolutional networks, or combinations thereof can also be used. The model is trained using machine learning algorithms by constructing a training dataset based on historical engineering data or numerical simulation data. Specifically, during the construction of the training dataset, a large amount of complete construction data from completed tunnels is collected, including multi-source information X from excavation to secondary lining completion, and the corresponding surrounding rock deformation. Support stress Wait for the actual results. Each sample is a time series segment. and its corresponding states at k future time points When historical data is insufficient, refined numerical simulation is used to simulate the tunnel construction process under different geological conditions and support schemes, generating a large number of "data-result" pairs as supplementary training data.
[0038] The training process includes: constructing a multiple-input multiple-output (MIMO) LSTM network. The input layer receives a feature sequence of a historical time window of length n. The LSTM hidden layer is used to capture temporal dynamic features in a sequence. The output layer is connected to a fully connected network and outputs multiple state predictions for the next k time steps, for example: Where S can be a vector containing predicted values for crown settlement, horizontal convergence, and key point support stress. The loss function is typically the mean squared error loss, and the optimizer is Adam. The dataset is proportionally divided into training, validation, and test sets for model training and hyperparameter tuning until the model's prediction accuracy on the test set meets engineering requirements or the prediction error is within acceptable limits.
[0039] Step S3: Based on the prediction results and preset safety thresholds and economic indicators, a recommended support adjustment plan is generated through a decision optimization algorithm.
[0040] By utilizing a trained prediction model and combining it with optimization objectives, intelligent support decisions are generated. In this embodiment, two decision generation methods are employed: a rule-based method and a reinforcement learning-based method, with different methods suitable for different scenarios.
[0041] (1) Decision-making based on rules and multi-objective optimization First, at construction time t, the latest acquired multi-source data feature sequence X(t) is input into the prediction model. Using the trained prediction model, the surrounding rock deformation for the next k construction cycles is output. Support stress Prediction curves for isostatic indices, where i = 1, 2, ..., k.
[0042] Secondly, define the decision-making objectives and set a safety threshold for surrounding rock deformation. and the safety threshold of support stress As constraints on support safety, support material cost (C) and construction delay cost (D) are defined as economic indicators. Optional support adjustment measures are determined by constructing a decision space. For example, increasing the density of anchor bolts and increasing the thickness of shotcrete.
[0043] Assuming support adjustment actions are implemented By modifying the feature vector The corresponding elements are then input into the prediction model again. This yields the result when an action is taken. After that, future state indicators , The new prediction curve.
[0044] Finally, the optimal decision is output by constructing a comprehensive evaluation function, where the comprehensive evaluation function is:
[0045] in, The weighting coefficients for the corresponding indicators. and This represents the cost of the original plan.
[0046] Choose the evaluation function Support adjustment action to achieve minimum value This is the recommended support adjustment plan.
[0047] (2) Intelligent decision-making based on reinforcement learning State S(t) represents the predicted state of the tunnel surrounding rock-support system at time t. Here, state S(t) is a summary vector extracted from the output of the state prediction model, which can include key indicators such as predicted deformation and stress in the next k time steps.
[0048] Action A represents optional support adjustment measures, such as the set of actions defined in method A. ,in These represent different support adjustment schemes.
[0049] The reward R(t+1) is determined based on the deviation between the actual monitoring result S_real(t+1) and the predicted result S(t+1) at the next time step t+1, and the action taken. Based on the cost, the reward function can be defined as follows: ,in These are the weighting coefficients. In order to take action The cost.
[0050] The environment consists of a real tunnel construction process and a state prediction model. The agent interacts with the environment, obtains the state S(t), and executes actions. .
[0051] Furthermore, the agent is trained. Using historical construction data or a simulation environment, the agent is allowed to continuously try different support adjustment actions. Based on the reward R(t+1), the optimal decision-making strategy π(S(t)) is learned. Strategy π(S(t)) can directly output the optimal support adjustment action based on the current state S(t). .
[0052] A pre-trained policy network π(S(t)) is deployed for online decision-making. During actual construction, the real-time feature vector X(t) is input into the state prediction model to obtain the current state S(t). The optimal support adjustment action is directly generated through the policy network π(S(t)). This online decision-making method based on reinforcement learning offers fast decision-making speed and can learn complex long-term optimal strategies.
[0053] Step S4: The generated support adjustment plan is pushed to the construction end, and the new monitoring data after the decision is executed is used as feedback to incrementally learn the state prediction model.
[0054] The support decision recommendations (including specific measures, parameters, and basis) generated in step S3 are pushed to the construction manager and technicians in the tunnel in real time through the terminal device to guide them in adjusting the support operation.
[0055] After the decision is implemented, new monitoring data continues to be collected. The actual deformation and stress data are then compared with the model's predicted data.
[0056] 2) Model update strategy: ① Sliding window update: Always retain the data from the most recent N construction cycles, and periodically use this latest dataset to fine-tune the prediction model so that the model closely follows the geological characteristics of the current construction section.
[0057] ② Online learning: For models that support incremental learning, a small-step weight update is performed every time a new batch of (X,Y) data pairs is obtained.
[0058] ③ Reinforcement learning agent update: Based on the actual reward r_t obtained, continuously update the parameters of the policy network π to continuously improve its decision-making ability.
[0059] Through this closed loop of "decision-execution-feedback-update", the entire intelligent support system has the ability to adapt and continuously evolve.
[0060] Furthermore, considering the continuous and parameter-rich nature of TBM construction data streams, in step S2, a hybrid model combining a one-dimensional temporal convolutional network (1D-CNN) and a gated recurrent unit (GRU) is employed for prediction. 1D-CNN can efficiently extract local feature patterns from real-time TBM tunneling parameters (thrust, torque, rotational speed, penetration depth, etc.) and synchronous monitoring data. Its input is a time series. ,in The eigenvector at time t ( (where d is the dimension of the feature vector at a single time step); the convolutional layer uses a one-dimensional convolutional kernel. ( (where k is the length of the one-dimensional convolution kernel) The input sequence is convolved to obtain the feature map. : ,in This is the activation function.
[0061] 2) GRU time series modeling: GRU is good at capturing long-term dependencies in time series data.
[0062] GRU's hidden state From the current input and the hidden state at the previous moment A joint decision.
[0063] 3) Hybrid model prediction: Local feature sequences extracted by 1D-CNN As input to GRU, GRU further captures the evolution of features in the time dimension; the final output layer is connected to a fully connected network to output the prediction results for the next k time steps, such as surrounding rock deformation and support stress; this hybrid model can more sensitively reflect the instantaneous correlation between changes in TBM tunneling parameters and surrounding rock response.
[0064] In step S3, decision generation focuses on real-time optimization of TBM tunneling parameters and triggering of advanced support. Specifically: Based on the surrounding rock stability trend output by the prediction model, the decision-making system will suggest adjusting the tunneling parameters of the TBM, forming a rapid closed loop of "prediction-control".
[0065] When a significant deformation or instability risk is predicted to occur within a short distance ahead, the system will trigger an advance support decision in advance and provide specific support range and parameter suggestions, transforming traditional post-event emergency response into pre-event prevention.
[0066] Example 2 This embodiment discloses a machine learning-based intelligent tunnel support system; like Figure 2 As shown, a machine learning-based intelligent tunnel support system includes: The data acquisition and preprocessing module is configured to: acquire multi-source data in real time during tunnel construction and preprocess it to obtain standardized time-series feature vectors; The state prediction module is configured to input the time-series feature vector into the trained surrounding rock-support system state prediction model to obtain the prediction result of the system state. The intelligent decision-making module is configured to generate a recommended support adjustment plan based on the prediction results and preset safety thresholds and economic indicators through a decision optimization algorithm. The decision execution and feedback module is configured to push the generated support adjustment plan to the construction end and use the new monitoring data after the decision execution as feedback to incrementally learn the state prediction model.
[0067] Example 3 The purpose of this embodiment is to provide a computer-readable storage medium.
[0068] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a machine learning-based intelligent tunnel support method as described in Example 1.
[0069] Example 4 The purpose of this embodiment is to provide an electronic device.
[0070] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in a machine learning-based intelligent tunnel support method as described in Embodiment 1.
[0071] The steps and methods involved in the apparatuses of Embodiments 2, 3, and 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.
[0072] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.
[0073] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.
Claims
1. A machine learning-based intelligent tunnel support method, characterized in that, include: Multi-source data is collected in real time during tunnel construction and preprocessed to obtain standardized time-series feature vectors; The time-series feature vector is input into the trained surrounding rock-support system state prediction model to obtain the prediction result of the system state. Based on the prediction results and preset safety thresholds and economic indicators, a recommended support adjustment scheme is generated through a decision optimization algorithm. The generated support adjustment plan is pushed to the construction end, and the new monitoring data after the decision is executed is used as feedback to incrementally learn the state prediction model.
2. The intelligent tunnel support method based on machine learning as described in claim 1, characterized in that, The multi-source data includes geological information, support structure response information, and construction condition information; The geological information includes at least one of the following: digital information of geological sketches of the tunnel face, results of advanced geological prediction, rock strength parameters, and information on rock mass structure surfaces; The support structure response information includes at least one of the following: crown settlement, horizontal convergence, support concrete stress, anchor bolt axial force, and steel arch strain. The construction condition information includes at least one of the following: excavation progress, support construction delay distance, blasting parameters, or TBM tunneling parameters.
3. The intelligent tunnel support method based on machine learning as described in claim 1, characterized in that, The preprocessing includes data cleaning, temporal alignment, and feature engineering of the collected multi-source data to form standardized temporal feature vectors.
4. The intelligent tunnel support method based on machine learning as described in claim 1, characterized in that, In the process of training the state prediction model of the surrounding rock-support system, a training dataset is constructed based on historical engineering data or numerical simulation data, and trained using machine learning algorithms, wherein the machine learning algorithms are long short-term memory networks, gated recurrent units, temporal convolutional networks, or combinations thereof.
5. The intelligent tunnel support method based on machine learning as described in claim 1, characterized in that, The decision optimization algorithm is a rule-based reasoning expert system, a multi-objective optimization algorithm, or a reinforcement learning agent.
6. The intelligent tunnel support method based on machine learning as described in claim 5, characterized in that, When the decision optimization algorithm is a reinforcement learning agent, the output of the state prediction model is used as the environmental state perception; and the set of support and adjustment actions is used as the agent's action space. The reward function is a comprehensive indicator reflecting the safety and economy of support; the reinforcement learning agent is trained to learn a mapping strategy from state to optimal action.
7. The intelligent tunnel support method based on machine learning as described in claim 1, characterized in that, The support adjustment scheme includes at least one of whether to strengthen the support, adjust the support parameters, and optimize the timing of the operation; when using TBM construction, the support adjustment scheme also includes adjusting the TBM tunneling parameters and / or triggering advance support.
8. A machine learning-based intelligent tunnel support system, characterized in that, include: The data acquisition and preprocessing module is configured to: acquire multi-source data in real time during tunnel construction and preprocess it to obtain standardized time-series feature vectors; The state prediction module is configured to input the time-series feature vector into the trained surrounding rock-support system state prediction model to obtain the prediction result of the system state. The intelligent decision-making module is configured to generate a recommended support adjustment plan based on the prediction results and preset safety thresholds and economic indicators through a decision optimization algorithm. The decision execution and feedback module is configured to push the generated support adjustment plan to the construction end and use the new monitoring data after the decision execution as feedback to incrementally learn the state prediction model.
9. A computer-readable storage medium having a program stored thereon, characterized in that, When executed by a processor, the program implements the steps of a machine learning-based intelligent tunnel support method as described in any one of claims 1-7.
10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the machine learning-based intelligent tunnel support method as described in any one of claims 1-7.