Surrounding rock deformation prediction method for tunnel construction in soft rock area

Abstract

The invention discloses a surrounding rock deformation prediction method for tunnel construction in a soft rock area, belongs to the technical field of surrounding rock deformation prediction, and can solve the problem of low accuracy of surrounding rock deformation prediction in the soft rock area in the prior art. The method comprises the following steps: S1, according to a surrounding rock monitoring network of a tunnel, respectively adding adaptive weight items to a plurality of loss functions in a target function of a gradient boosting decision tree algorithm to obtain an improved gradient boosting decision tree algorithm; the plurality of loss functions are in one-to-one correspondence with a plurality of monitoring points in the monitoring network; s2, using surrounding rock monitoring data collected by a surrounding rock monitoring network to train an improved gradient boosting decision tree algorithm to obtain a tunnel surrounding rock deformation prediction model; and S3, predicting the surrounding rock deformation condition of the tunnel by using the surrounding rock deformation prediction model. The method is used for predicting surrounding rock deformation.

Description

Technical Field

[0001] The invention relates to a surrounding rock deformation prediction method for tunnel construction in soft rock areas, belonging to the technical field of surrounding rock deformation prediction. Background Art

[0002] Tunnel construction in soft rock areas is like carving a work of art in soft sand. Every step requires caution and is full of challenges. The geological conditions in soft rock areas are like a complex and changeable maze. The interaction between rock layers, the flow of groundwater, the distribution of ground stress and other factors are intertwined, making the deformation of the surrounding rock elusive and full of uncertainty. Existing surrounding rock deformation prediction methods, such as estimation methods based on empirical formulas and simulation methods based on simplified calculation models, are often like blind men touching an elephant in soft rock areas. It is difficult to fully, real-time and accurately reflect the actual deformation of the surrounding rock. This not only brings great safety risks to construction projects in soft rock areas, but also seriously affects the engineering progress and cost control of construction projects. Summary of the invention

[0003] The invention provides a surrounding rock deformation prediction method for tunnel construction in soft rock areas, which can solve the problem of insufficient accuracy and real-time performance of surrounding rock deformation prediction in soft rock areas in the prior art.

[0004] The present invention provides a surrounding rock deformation prediction method for tunnel construction in soft rock areas, the method comprising: S1. According to the surrounding rock monitoring network of the tunnel, adaptive weight terms are added to the multiple loss functions in the objective function of the gradient boosting decision tree algorithm to obtain an improved gradient boosting decision tree algorithm; the multiple loss functions correspond one-to-one to the multiple monitoring points in the monitoring network; S2. Using the surrounding rock monitoring data collected by the surrounding rock monitoring network, the improved gradient boosting decision tree algorithm is trained to obtain a surrounding rock deformation prediction model of the tunnel; S3. Predicting the surrounding rock deformation of the tunnel using the surrounding rock deformation prediction model.

[0005] Optionally, S2 is specifically: S21, dividing the surrounding rock monitoring data into a training set and a test set, and using the training set to train the improved gradient boosting decision tree algorithm to obtain a trained improved algorithm; S22. Using the test set, multiple adaptive weight items in the trained improved algorithm are adjusted to obtain a surrounding rock deformation prediction model of the tunnel.

[0006] Optionally, the S22 is specifically: According to the test set, and using the trained improved algorithm, the deformation of multiple monitoring points is predicted to obtain the deformation prediction effect of each monitoring point; According to the deformation prediction effect, multiple adaptive weight items in the trained improved algorithm are adjusted to obtain the surrounding rock deformation prediction model of the tunnel.

[0007] Optionally, in S21, the surrounding rock monitoring data is divided into a training set and a test set, specifically: Extracting characteristic data of surrounding rock deformation from the surrounding rock monitoring data; The characteristic data of surrounding rock deformation are divided into a training set and a test set.

[0008] Optionally, the step of training the improved gradient boosting decision tree algorithm using the training set in S21 is as follows: The improved gradient boosting decision tree algorithm is trained by using the training set and adjusting the learning rate.

[0009] Optionally, before S1, the method further includes: A surrounding rock monitoring network is deployed in the tunnel, and the surrounding rock monitoring data is collected using the surrounding rock monitoring network.

[0010] Optionally, the surrounding rock monitoring network of the tunnel is specifically arranged as follows: A plurality of monitoring points are arranged at different positions in the axial direction and the circumferential direction of the tunnel, and one or more of a displacement sensor, a stress sensor and a temperature sensor are arranged at each monitoring point.

[0011] Optionally, before S2, the method further includes: Preprocessing the surrounding rock monitoring data collected by the surrounding rock monitoring network; Correspondingly, the surrounding rock monitoring data in S2 is pre-processed surrounding rock monitoring data.

[0012] Optionally, the preprocessing of the surrounding rock monitoring data collected by the surrounding rock monitoring network is specifically: The surrounding rock monitoring data collected by the surrounding rock monitoring network are subjected to data cleaning, denoising, format conversion and standardization.

[0013] The beneficial effects that the present invention can produce include: The present invention optimizes and improves the algorithm by adding an adaptive weight term to the objective function of the gradient boosting decision tree algorithm. The surrounding rock deformation prediction model obtained in this way can re-evaluate the importance of each feature according to the deformation prediction effect of each monitoring point, thereby realizing dynamic adjustment of the model to balance the prediction accuracy and complexity of the model. This improvement enables the model to have better performance and generalization ability when processing complex and changeable soft rock tunnel data, which is significantly better than the single or simple prediction model that may be used in the prior art.

[0014] The present invention realizes all-round and real-time monitoring of surrounding rock deformation by constructing a high-precision surrounding rock monitoring network, which can capture tiny deformation signals and ensure the comprehensiveness, real-timeness and accuracy of data. This makes data collection no longer limited to local or periodic monitoring, but realizes continuous and real-time data flow, providing a richer and more sophisticated data foundation for subsequent data processing and analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic diagram of a surrounding rock deformation prediction method for tunnel construction in soft rock areas provided by an embodiment of the present invention; Figure 2 A flow chart of preprocessing surrounding rock monitoring data provided by an embodiment of the present invention; Figure 3 A flow chart for constructing a surrounding rock deformation prediction model provided by an embodiment of the present invention; Figure 4 A flow chart for predicting surrounding rock deformation provided by an embodiment of the present invention; Figure 5 A schematic diagram of the structure of a deformation prediction system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] The present invention is described in detail below in conjunction with embodiments, but the present invention is not limited to these embodiments.

[0017] The embodiment of the present invention provides a surrounding rock deformation prediction method for tunnel construction in soft rock areas, such as Figure 1 As shown, the method includes: S1. According to the surrounding rock monitoring network of the tunnel, adaptive weight terms are added to the multiple loss functions in the objective function of the gradient boosting decision tree algorithm to obtain an improved gradient boosting decision tree algorithm; the multiple loss functions correspond one-to-one to the multiple monitoring points in the monitoring network; S2. Using the surrounding rock monitoring data collected by the surrounding rock monitoring network, the improved gradient boosting decision tree algorithm is trained to obtain a surrounding rock deformation prediction model for the tunnel; S3. Use the surrounding rock deformation prediction model to predict the surrounding rock deformation of the tunnel.

[0018] In this embodiment, before S1, the method may further include: Lay out the tunnel's surrounding rock monitoring network and use it to collect surrounding rock monitoring data.

[0019] Furthermore, the surrounding rock monitoring network of the tunnel may specifically be: A plurality of monitoring points are arranged at different positions in the axial direction and the circumferential direction of the tunnel, and one or more of a displacement sensor, a stress sensor and a temperature sensor are arranged at each monitoring point.

[0020] Specifically, the displacement sensor can be a laser rangefinder, the stress sensor includes a strain gauge or a fiber grating sensor, and the temperature sensor can be a thermocouple or an infrared thermometer. These sensors are arranged at monitoring points on the top and side walls of the tunnel and other key areas that may be affected by the construction to ensure that the deformation dynamics of the surrounding rock can be captured in all directions and in real time. When making specific arrangements, the geometry of the tunnel, geological conditions, and construction progress should be considered, and a three-dimensional spatial coordinate system should be used for precise positioning to ensure that the surrounding rock monitoring network can accurately reflect the actual deformation of the surrounding rock.

[0021] For example, suppose a tunnel is located in a soft rock area with complex geological conditions, and the tunnel section is a circle with a diameter of 10 meters. In order to achieve real-time monitoring and accurate prediction of surrounding rock deformation, a comprehensive surrounding rock monitoring network needs to be deployed at key locations around the tunnel.

[0022] Specifically, a high-precision laser rangefinder is arranged at the monitoring point located at the center of the top of the tunnel to monitor the sinking of the vault. The sensor is 1 meter away from the vault and can capture the tiny displacement changes of the vault in real time. Considering the symmetry of the tunnel and the uncertainty of geological conditions, three displacement sensors are arranged at the upper, middle and lower monitoring points of the tunnel sidewall. The heights of the three monitoring points are 1 meter, 3 meters and 5 meters from the tunnel floor, respectively, and are evenly spaced to ensure comprehensive monitoring of the sidewall deformation.

[0023] Displacement sensors are arranged at the monitoring point at the central axis of the tunnel floor to monitor the uplift of the floor. At the same time, displacement sensors are also arranged at the monitoring points 3 meters away from the central axis on both sides of the central axis to monitor the lateral displacement of the floor. The arrangement of these sensors helps to monitor the stability of the tunnel floor and give early warning of possible floor deformation.

[0024] In order to monitor the stress condition of the key supporting structure of the tunnel, a monitoring point is set every 5 meters near the anchor rod, shotcrete layer and other locations, and stress sensors are arranged at these monitoring points. These stress sensors can withstand the harsh underground environment and are very sensitive to stress changes, which can provide important data support for the stability monitoring of the tunnel. In addition, in order to consider the impact of ambient temperature on the stability of the surrounding rock, a temperature sensor can be arranged at each of the five monitoring points at different locations in the tunnel. These sensors can monitor the changes in ambient temperature in real time and respond to the changes in the physical properties of the surrounding rock caused by temperature.

[0025] When arranging the monitoring points, a three-dimensional spatial coordinate system is required for precise positioning to ensure that each monitoring point can be accurately reflected in the three-dimensional model of the tunnel. Through the Geographic Information System (GIS) technology, the monitoring points are combined with the geological model of the tunnel to facilitate the analysis and interpretation of the surrounding rock monitoring data. In terms of data collection, all sensors are connected to a central data acquisition system via wired or wireless means. The system can receive data from the sensors in real time and transmit it to the data processing center for analysis. The data processing center is equipped with powerful servers and professional data analysis software, which can quickly process and analyze large amounts of data and provide accurate input for the surrounding rock deformation prediction model.

[0026] The tunnel's surrounding rock monitoring network can collect surrounding rock monitoring data and conduct all-round real-time monitoring of surrounding rock deformation during tunnel construction in soft rock areas. This not only improves the safety of tunnel construction, but also provides timely deformation warning and decision-making support for engineering and technical personnel, thereby optimizing the construction process, improving construction efficiency and reducing construction costs.

[0027] Then, this embodiment improves the Gradient Boosting Decision Tree (GBDT) algorithm according to the surrounding rock monitoring network of the tunnel to obtain an improved Gradient Boosting Decision Tree algorithm (improved GBDT algorithm).

[0028] The improved GBDT algorithm introduces a feature importance re-evaluation strategy based on the GBDT algorithm, where the features are key indicators that have an important impact on surrounding rock deformation, including historical deformation, deformation rate, temperature change, stress state, etc. The feature importance re-evaluation strategy iteratively re-evaluates the importance of each feature and adjusts the selection and weight of each feature accordingly to ensure that the final surrounding rock deformation prediction model can capture the most critical information for the prediction results.

[0029] Specifically, this embodiment optimizes and improves the objective function of the GBDT algorithm. The objective function of the GBDT algorithm usually includes two parts: a loss function and a regularization term. This embodiment improves this objective function by adding an adaptive weight term to each loss function in the objective function of the GBDT algorithm, so that it can better balance the prediction accuracy and complexity of the model.

[0030] The objective function of the improved GBDT algorithm is expressed as: (1) In formula (1), It is the objective function of the improved GBDT algorithm; It is a monitoring point The loss function is used to measure the monitoring point The difference between the predicted deformation and the actual deformation. The smaller the difference, the more accurate the prediction of the model. It is a monitoring point The actual deformation amount; It is a monitoring point The predicted deformation amount; It is the monitoring point The predicted deformation has the characteristics of influence; is the total number of monitoring points; It is a monitoring point The adaptive weight term is based on the monitoring point Dynamically adjust the prediction difficulty; is a regularization term used to control the complexity of the surrounding rock deformation prediction model; and is the regularization parameter; is the total number of features; It is a feature Importance weight of is the indicator function, when the feature When selected, The value of is 1, otherwise The value of is 0.

[0031] Specifically, for regression problems, common loss functions include mean square error (MSE); for classification problems, common loss functions include cross entropy loss.

[0032] By introducing the adaptive weight term , the surrounding rock deformation prediction model can pay more attention to those monitoring points with greater prediction difficulty. Before the model training starts, this embodiment assigns an adaptive weight item to each monitoring point , its initial value can be set according to the category, distribution location, prediction difficulty and prediction demand of the monitoring point. For example, when dealing with an unbalanced data set, a higher weight can be assigned to the monitoring points corresponding to the minority class samples.

[0033] During the model training process, the adaptive weight items are dynamically adjusted based on the model's prediction performance at each monitoring point. If the prediction error of a monitoring point is large, its weight value can be increased so that the model pays more attention to these monitoring points in subsequent iterations. This dynamic adjustment mechanism helps to improve the model's prediction accuracy for monitoring points that are more difficult to predict, thereby improving the overall prediction performance of the model.

[0034] Specifically, common regularization terms include L1 regularization (the sum of the absolute values ​​of the weights) and L2 regularization (the sum of the squares of the weights).

[0035] Regularization term It can control the complexity of the model and prevent the model from overfitting. Overfitting means that the model performs well on the training data, but performs poorly on new, unseen data. Regularization improves the generalization ability of the model by limiting the complexity of the model. and It is necessary to determine its optimal value through methods such as cross-validation to balance the model's fitting ability and generalization ability.

[0036] In summary, the objective function of the improved GBDT algorithm can better balance the prediction accuracy and complexity of the model, and improve the prediction performance and generalization ability of the model in the deformation prediction task of tunnel construction in soft rock areas.

[0037] This embodiment collects surrounding rock monitoring data through a surrounding rock monitoring network, and uses the surrounding rock monitoring data to train an improved GBDT algorithm to obtain a surrounding rock deformation prediction model for the tunnel.

[0038] Before training, that is, before S2, this embodiment also needs to pre-process the surrounding rock monitoring data collected by the surrounding rock monitoring network; Correspondingly, the surrounding rock monitoring data in S2 is the preprocessed surrounding rock monitoring data.

[0039] Furthermore, the above-mentioned preprocessing of the surrounding rock monitoring data collected by the surrounding rock monitoring network may specifically be: The surrounding rock monitoring data collected by the surrounding rock monitoring network are cleaned, denoised, format converted and standardized.

[0040] Specifically, the data preprocessing process of this embodiment is as follows: Figure 2As shown in the figure. The data processing center receives the surrounding rock monitoring data collected by the surrounding rock monitoring network. First, the surrounding rock monitoring data is cleaned to remove abnormal values ​​or missing values ​​caused by equipment failure, signal interference, etc. Then, digital filtering technology (such as Kalman filtering) is used to denoise the cleaned data to reduce the impact of random errors on subsequent analysis. Finally, the denoised data is format converted and standardized to ensure that the format of data collected by different sensors can be unified for subsequent data processing and model training.

[0041] Specifically, the recursive formula of the Kalman filter includes a prediction equation and an update equation.

[0042] The prediction equation can be expressed as: (2) (3) In formula (2) and formula (3), is based on The estimated time Prediction of the state at a given moment, where “state” refers to the current condition of the system, such as displacement, velocity, acceleration, etc. For example; represent The displacement vector at the moment; express The covariance matrix of the moment; is the state transition matrix; is the control matrix; is the control input; is the forecast error covariance matrix; is the process noise covariance matrix.

[0043] The update equation can be expressed as: (4) (5) (6) In formula (4) to formula (6), is the Kalman gain; is the observation matrix; yes The observed value at time; is the observation noise covariance matrix; is the updated state estimate; is the updated error covariance matrix.

[0044] In this embodiment, S2 may specifically be: S21, dividing the pre-processed surrounding rock monitoring data into a training set and a test set, and using the training set to train the improved gradient boosting decision tree algorithm to obtain a trained improved algorithm; S22. Using the test set, multiple adaptive weight items in the trained improved algorithm are adjusted to obtain a surrounding rock deformation prediction model for the tunnel.

[0045] Furthermore, in S21, the pre-processed surrounding rock monitoring data is divided into a training set and a test set, which can be specifically: Extract characteristic data of surrounding rock deformation from the preprocessed surrounding rock monitoring data; divide the characteristic data of surrounding rock deformation into a training set and a test set.

[0046] Specifically, this embodiment extracts feature data that has an important impact on surrounding rock deformation from the preprocessed surrounding rock monitoring data through feature engineering, such as historical deformation data, deformation rate data, temperature change data, stress state data, etc., to form a data set. Then, the data set is divided into a training set and a test set using cross-validation and other techniques, which are used for training and verification of the surrounding rock deformation prediction model, respectively.

[0047] Furthermore, in S21, the improved gradient boosting decision tree algorithm is trained using the training set to obtain an improved algorithm after training, which may be specifically: The improved gradient boosting decision tree algorithm is trained by using the training set and adjusting the learning rate to obtain the improved algorithm after training.

[0048] This improved gradient boosting decision tree algorithm can dynamically adjust the learning rate according to the rate of decrease of the loss function during training, thereby accelerating convergence and reducing the risk of overfitting.

[0049] Specifically, this embodiment dynamically adjusts the learning rate according to the training process to optimize the model performance. Adjusting the learning rate includes: Gradient accumulation: In each iteration, the squares of all past gradients are accumulated and used to adjust the learning rate. For example, in the AdaGrad algorithm, the learning rate is adjusted inversely proportional to the square root.

[0050] Attenuation coefficient: In the RMSprop algorithm, the attenuation coefficient is introduced to prevent the accumulation of the sum of squared gradients from being too extreme, thereby avoiding a sharp decrease in the learning rate.

[0051] Momentum and variance estimation: In the Adam algorithm, the advantages of RMSprop and Momentum are combined, and the first-order moment estimation and second-order moment estimation of the gradient are used to adjust the learning rate.

[0052] Further, S22 may specifically be: According to the test set, the deformation of multiple monitoring points is predicted using the improved algorithm after training to obtain the deformation prediction effect of each monitoring point; According to the deformation prediction effect, multiple adaptive weight items in the trained improved algorithm are adjusted to obtain the surrounding rock deformation prediction model of the tunnel.

[0053] Specifically, this embodiment uses a test machine to verify the prediction effect of the improved algorithm after training at each monitoring point, and dynamically adjusts the adaptive weight items and other parameters according to the prediction effect to optimize the algorithm parameters and model structure. If the prediction error of a certain monitoring point is large, its weight value can be increased so that the surrounding rock deformation prediction model finally established pays more attention to these monitoring points in subsequent iterations. This dynamic adjustment mechanism helps to improve the prediction accuracy of the surrounding rock deformation prediction model for monitoring points that are more difficult to predict, thereby improving the prediction performance of the surrounding rock deformation prediction model. The model construction process of this embodiment is as follows: Figure 3 shown.

[0054] After the surrounding rock deformation prediction model is built, this embodiment inputs the surrounding rock monitoring data collected in real time into the prediction model to predict the surrounding rock deformation. The prediction model will calculate and output key indicators such as the deformation amount and deformation rate of the surrounding rock based on the input data to guide the construction. In order to ensure the real-time prediction, this embodiment uses a stream processing framework (such as Apache Kafka or Spark Streaming) to process and analyze the real-time data stream.

[0055] In summary, the surrounding rock deformation prediction flow chart of this embodiment is as follows: Figure 4 shown.

[0056] In terms of data collection, the data collection and processing algorithm of this embodiment realizes all-round and real-time monitoring of surrounding rock deformation through a high-precision surrounding rock monitoring network, and can capture tiny deformation signals to ensure the comprehensiveness, real-time and accuracy of data. This makes data collection no longer limited to local or periodic monitoring, but realizes continuous and real-time data flow, providing a richer and more sophisticated data foundation for subsequent data processing and analysis.

[0057] In terms of data processing, this embodiment uses advanced data cleaning, denoising and feature extraction technologies, such as digital filtering technologies such as Kalman filtering, to reduce the impact of random errors on subsequent analysis and ensure that data collected by different sensors can be in a unified format to facilitate subsequent data processing and model training. This refined data processing process can more effectively extract valuable information and provide more reliable input for the prediction model compared to the simple data processing method in the prior art.

[0058] In terms of prediction model construction, the prediction model of this solution is built based on big data and machine learning technology, which can comprehensively consider the influence of various factors such as geological conditions, construction methods, support measures, etc. on surrounding rock deformation to achieve accurate prediction. In particular, this embodiment adopts an improved GBDT algorithm, introduces an adaptive learning rate mechanism and a feature importance re-evaluation strategy, and optimizes the objective function of the algorithm, so that it can better balance the prediction accuracy and complexity of the model. This improvement enables the model to have better performance and generalization ability when processing complex and changeable soft rock tunnel data, which is significantly better than the single or simple prediction model in the prior art.

[0059] In addition, the prediction model of this embodiment also has a strong adaptive capability and real-time feedback mechanism, which can dynamically adjust and optimize the prediction model according to real-time monitoring data to adapt to the complex and changeable geological conditions and changes in the construction environment. This adaptive and real-time feedback capability makes the prediction model of this solution not only superior to the existing technology in terms of prediction accuracy, but also significantly improves its real-time and adaptability.

[0060] It can be seen that the improvements in the data acquisition and processing algorithm and the prediction model algorithm of this embodiment not only improve the comprehensiveness, real-time and accuracy of data acquisition, but also improve the refinement level of data processing and the accuracy, real-time and adaptability of the prediction model, thereby providing a more scientific and effective solution for the prediction of surrounding rock deformation during tunnel construction in soft rock areas. These improvements make this embodiment significantly different from the existing technology in terms of technical implementation and application effect, and provide a stronger guarantee for the safety and engineering quality of tunnel engineering.

[0061] Another embodiment of the present invention provides a deformation prediction system based on the surrounding rock deformation prediction method for tunnel construction in soft rock areas, such as Figure 5 shown.

[0062] The deformation prediction system includes a data acquisition module, a data transmission module, a data processing module, a prediction model module and a result output module. The data acquisition module uses the surrounding rock monitoring network to collect real-time data of the surrounding rock; the data transmission module transmits the collected data to the data processing module through the data transmission system; the data processing module preprocesses and extracts features of the data; the prediction model module predicts deformation based on machine learning algorithms; the result output module visualizes the prediction results, generates deformation warning information and construction guidance information, and outputs them to engineering and technical personnel.

[0063] The deformation prediction system uses advanced sensor technology and data analysis algorithms to achieve real-time monitoring and accurate prediction of surrounding rock deformation. Through the surrounding rock monitoring network arranged around the tunnel, the system can collect surrounding rock deformation data in real time and transmit the data to the data center for processing and analysis through high-speed data transmission technology. The data analysis algorithm will combine geomechanics theory, construction experience and real-time monitoring data to establish an accurate prediction model to accurately predict the deformation trend of the surrounding rock.

[0064] In addition, the deformation prediction system also has strong adaptive capabilities and real-time feedback mechanisms. The system can dynamically adjust and optimize the prediction model based on real-time monitoring data to adapt to complex and changing geological conditions and changes in the construction environment. At the same time, the system can also provide deformation warning information and construction guidance information to engineering and technical personnel in real time, helping them to take effective measures in a timely manner to deal with potential safety risks.

[0065] Therefore, the deformation prediction system not only improves the accuracy and real-time performance of the prediction results, but also enhances the adaptability and practicality of the system. Its emergence will provide a more scientific and effective solution to the deformation prediction problem during tunnel construction in soft rock areas, and promote the continuous progress and development of tunnel engineering technology.

[0066] The above are only a few embodiments of the present application and do not constitute any form of limitation to the present application. Although the present application is disclosed as above with preferred embodiments, it is not intended to limit the present application. Any technician familiar with the profession, without departing from the scope of the technical solution of the present application, using the technical content disclosed above to make slight changes or modifications are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. A method for predicting surrounding rock deformation for tunnel construction in soft rock areas, characterized in that: The method comprises: S1. According to the surrounding rock monitoring network of the tunnel, adaptive weight terms are added to the multiple loss functions in the objective function of the gradient boosting decision tree algorithm to obtain an improved gradient boosting decision tree algorithm; the multiple loss functions correspond one-to-one to the multiple monitoring points in the monitoring network; S2. Using the surrounding rock monitoring data collected by the surrounding rock monitoring network, the improved gradient boosting decision tree algorithm is trained to obtain a surrounding rock deformation prediction model of the tunnel; S3, predicting the surrounding rock deformation of the tunnel using the surrounding rock deformation prediction model; The objective function of the improved gradient boosting decision tree algorithm is expressed as: ， In the formula, is the objective function of the improved gradient boosting decision tree algorithm, For monitoring point The loss function is For monitoring point The actual deformation of For monitoring point The predicted deformation For monitoring points The predicted deformation has the characteristics of influencing, is the total number of monitoring points, For monitoring point The adaptive weight term of the loss function, is the regularization term, and is the regularization parameter, is the total number of features, Features The importance weight of is the indicator function.

2. The method according to claim 1, characterized in that The S2 is specifically: S21, dividing the surrounding rock monitoring data into a training set and a test set, and using the training set to train the improved gradient boosting decision tree algorithm to obtain a trained improved algorithm; S22. Using the test set, multiple adaptive weight items in the trained improved algorithm are adjusted to obtain a surrounding rock deformation prediction model of the tunnel.

3. The method according to claim 2, characterized in that The S22 is specifically: According to the test set, and using the trained improved algorithm, the deformation of multiple monitoring points is predicted to obtain the deformation prediction effect of each monitoring point; According to the deformation prediction effect, multiple adaptive weight items in the trained improved algorithm are adjusted to obtain the surrounding rock deformation prediction model of the tunnel.

4. The method according to claim 2, characterized in that: In S21, the surrounding rock monitoring data is divided into a training set and a test set, specifically: Extracting characteristic data of surrounding rock deformation from the surrounding rock monitoring data; The characteristic data of surrounding rock deformation are divided into a training set and a test set.

5. The method according to claim 2, characterized in that: In S21, the improved gradient boosting decision tree algorithm is trained using the training set, specifically: The improved gradient boosting decision tree algorithm is trained by using the training set and adjusting the learning rate.

6. The method according to claim 1, characterized in that Before S1, the method further includes: A surrounding rock monitoring network is deployed in the tunnel, and the surrounding rock monitoring data is collected using the surrounding rock monitoring network.

7. The method according to claim 6, characterized in that The surrounding rock monitoring network of the tunnel is specifically arranged as follows: A plurality of monitoring points are arranged at different positions in the axial direction and the circumferential direction of the tunnel, and one or more of a displacement sensor, a stress sensor and a temperature sensor are arranged at each monitoring point.

8. The method according to claim 1, characterized in that: Before S2, the method further includes: Preprocessing the surrounding rock monitoring data collected by the surrounding rock monitoring network; Correspondingly, the surrounding rock monitoring data in S2 is pre-processed surrounding rock monitoring data.

9. The method according to claim 8, characterized in that The preprocessing of the surrounding rock monitoring data collected by the surrounding rock monitoring network is specifically as follows: The surrounding rock monitoring data collected by the surrounding rock monitoring network are subjected to data cleaning, denoising, format conversion and standardization.

Images