Highway tunnel surrounding rock deformation prediction method

Abstract

The invention discloses a highway tunnel surrounding rock deformation prediction method, and relates to the technical field of geotechnical engineering monitoring, and the method comprises the steps: collecting existing data, determining a monitoring arrangement scheme, and collecting on-site monitoring data through a monitoring data real-time feedback technology; processing multiple types of monitoring data and screening proper input features; clustering analysis is carried out on the monitoring data of the multiple sections, and classification of the excavated sections is completed; extracting the shape center of each section classification, and comparing the shape center with early-stage monitoring data of a predicted section to realize classification of the predicted section; and establishing a prediction model according to the classification of the prediction sections, and performing long-term prediction on the tunnel surrounding rock deformation response. According to the method, fine processing of the monitoring data is realized through feature screening and hierarchical clustering, and classification and classification of the known section and the predicted section are realized by fully utilizing multiple types of monitoring data, so that the long-term response of tunnel section surrounding rock deformation is predicted more accurately.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geotechnical engineering monitoring, in particular to a highway tunnel surrounding rock deformation prediction method. BACKGROUND

[0002] With the advancement of large-scale highway tunnel engineering, surrounding rock deformation control has become a core problem for construction safety and operation maintenance. Early researches mostly rely on empirical formulas and empirical curves (such as depth-settlement relationship), supplemented by finite element numerical simulation to conduct static analysis on the excavation-supporting process. Although these methods provide a reference for the initial design stage of the tunnel, they are difficult to reflect the dynamic response of the on-site construction disturbance and complex geological conditions in real time. After entering the Internet of Things era, the maturity of sensor network and wireless data transmission technology enables the online and continuous acquisition of multi-parameter monitoring such as surrounding rock displacement, anchor rod axial force and arch top settlement. At the same time, time series analysis and machine learning gradually emerge in the field of civil engineering, developing from a single time series prediction model to a hybrid method integrating clustering analysis, multi-dimensional dynamic time warping (DTW) and deep neural network, laying a technical foundation for high-precision and real-time prediction of surrounding rock deformation.

[0003] However, the existing technology still has significant deficiencies in multi-section information fusion and long-term prediction capability. First, traditional prediction models mostly focus on short-term trends, lack a rolling update mechanism for changes within the construction period, and are difficult to ensure the sustained stability of prediction accuracy; second, the preprocessing of multi-source monitoring data mostly stays at the level of simple denoising or interpolation, without fully screening key input features, resulting in limited model training efficiency and prediction reliability. More importantly, section classification and prediction model are often designed separately, lacking a section matching and classification process based on shape center, thus failing to achieve accurate classification and long-term deformation response prediction of newly excavated sections. SUMMARY

[0004] The present application aims to solve the problem that existing methods only perform short-term prediction on a single construction section or adjacent sections, and mostly use single feature time series monitoring data as input, resulting in too short warning response time and difficulty in timely adjustment.

[0005] Technical solution: In a first aspect, the present application provides a highway tunnel surrounding rock deformation prediction method, which includes collecting existing data and determining a monitoring arrangement scheme, collecting on-site monitoring data through real-time feedback technology; processing multi-type monitoring data and screening suitable input features; performing clustering analysis on the monitoring data of multiple sections to complete the classification of excavated sections; extracting the shape center of each section classification and comparing it with the pre-monitoring data of the prediction section to realize the classification of the prediction section; and establishing a prediction model according to the classification of the prediction section to perform long-term prediction of the tunnel surrounding rock deformation response.

[0006] Further, the prediction method collects existing data and determines a monitoring arrangement scheme, collects field monitoring data in real time through monitoring data feedback technology, including the following steps: investigating the geological conditions of the tunnel excavation area, the self-stability of the rock mass and the initial support structure, checking the surrounding rock classification, and recording relevant information; based on the completion of the geological and support investigation, combined with the actual situation on site, the monitoring points are arranged; the surrounding displacement and vault subsidence monitoring is carried out, and online monitoring is adopted to realize the collection of monitoring data.

[0007] Further, the prediction method processes multiple types of monitoring data and selects appropriate input features, including the following steps: interpolating the monitoring data to generate equidistant time series monitoring data; and selecting features from the equidistant time series monitoring data to calculate the importance of each feature.

[0008] Further, the prediction method performs cluster analysis on multi-section monitoring data to complete the classification of the excavated section, including the following steps: based on the importance of each feature, the distance between the initial independent clusters in the hierarchical aggregation clustering is calculated, and the GroupAverage algorithm is used to merge the cluster classes; the clustering tree is divided using an evaluation method to divide the clustering results.

[0009] Further, the hierarchical aggregation clustering specifically includes the following steps: first, initialize each feature vector as an independent cluster; calculate the cumulative distance between all clusters according to the cluster distance formula; merge the two clusters with the closest distance; repeat the steps until only one cluster is left.

[0010] Further, the prediction method extracts the shape center of each section classification and compares it with the pre-monitoring data of the predicted section to realize the classification of the predicted section, including the following steps: using the barycenter average dynamic time warping to calculate the shape center of the target feature in each cluster; using dynamic time warping to compare the similarity of the equidistant time series monitoring data of the target feature of the predicted section and the corresponding equidistant time series monitoring data in the shape center, and classifying the predicted section into the known classification.

[0011] Further, the prediction method establishes a prediction model, including setting a rolling window length according to the shortest interval of adjacent section excavation, and dividing it into a training set and a test set in proportion; based on the training set, a bidirectional LSTM network is constructed, wherein the input is historical window data and the output is the deformation prediction value of the future maximum monitoring days; in the test stage, real-time fusion of new monitoring data is realized to update the training set with a rolling window and dynamically adjust the model parameters to realize continuous prediction of surrounding rock deformation.

[0012] Further, the monitoring data includes surrounding displacement, vault subsidence, anchor rod axial force and step displacement.

[0013] In a second aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the computer program, when executed by the processor, implements any step of the highway tunnel surrounding rock deformation prediction method according to the first aspect of the present application.

[0014] In a third aspect, the present application provides a computer readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements any step of the highway tunnel surrounding rock deformation prediction method according to the first aspect of the present application.

[0015] Advantages: Compared with the prior art, the significant advantages of the present application are that: the prediction method utilizes multiple sets of completed monitoring section data, realizes section classification through hierarchical aggregation clustering, and based on the dynamic time warping principle, predicts and identifies the deformation data type of the prediction section. This method can effectively fuse multiple sources of information and multiple types of monitoring data, thereby greatly improving the accuracy of tunnel surrounding rock deformation prediction and the applicable range of the model. A rolling long-term prediction mechanism is adopted to realize continuous and dynamic prediction of the target section deformation. This mechanism significantly prolongs the warning response time, providing a strong guarantee for engineering safety management and timely adjustment. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 Flowchart of the highway tunnel surrounding rock deformation prediction method in the embodiment;

[0017] Figure 2 Schematic diagram of the result of using a random forest to filter monitoring data features in the embodiment;

[0018] Figure 3 Shape center extraction result of each cluster classification provided in the embodiment;

[0019] Figure 4 Ablation experiment result of the surrounding rock crown settlement displacement time series long-term prediction in the embodiment;

[0020] Figure 5 Long-term prediction result of the surrounding rock crown settlement displacement time series of part of the prediction sections provided in the embodiment. DETAILED DESCRIPTION

[0021] The technical solutions of the present application will be further described in detail below with reference to the accompanying drawings.

[0022] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. The "in one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or alternative embodiment that excludes other embodiments.

[0023] Embodiment 1

[0024] With reference to Figure 1 , the embodiment provides a highway tunnel surrounding rock deformation prediction method, comprising the following steps:

[0025] S1. Collect existing data and determine the monitoring arrangement scheme, collect field monitoring data through monitoring data real-time feedback technology.

[0026] Carry out engineering geology and initial support investigation. Investigate the geological conditions of the tunnel excavation area, the self-stability of the rock mass and the initial support structure, check the surrounding rock classification, and record the relevant information.

[0027] Specifically, drill hole sequences are arranged along the tunnel axis, and the obtained core is subjected to core integrity index (RQD) test; the surrounding rock is divided into six grades according to the “Rock Self-stability Classification Atlas”; the initial support form (such as the thickness of sprayed concrete, the diameter of anchor rod, the spacing of steel frame) that has been arranged on site is measured and recorded; the drilling data, in-situ test results and project archives (stratigraphic section, underground water level) are cross-checked; a “Surrounding Rock Classification and Initial Support Parameter Summary Table” is formed, which is used as the basic data for monitoring point arrangement.

[0028] On the basis of completing the geological and support investigation, combined with the actual situation on site, the monitoring points are reasonably arranged.

[0029] Carry out surrounding displacement and vault subsidence monitoring. Specifically, vault subsidence measuring points and net change measuring points are arranged at each monitoring section, the vault subsidence measuring points are arranged near the vault axis, the measuring point positions of each section are kept consistent and symmetrical as much as possible to ensure the comparability and accuracy of the monitoring data.

[0030] At the same time, the Internet of Things online monitoring technology is used to realize real-time, continuous and complete data acquisition; the monitoring data includes but is not limited to surrounding displacement, vault subsidence, anchor rod axial force and step displacement, etc.

[0031] S2. Process multiple types of monitoring data and select appropriate input features.

[0032] Interpolate the multiple types of monitoring data to generate equidistant time series monitoring data; wherein the interpolation method includes but is not limited to cubic spline interpolation method, specifically:

[0033] Read the original monitoring record to form a time stamp-monitoring value pair; sort in ascending order of time, and calculate the time interval of adjacent nodes.

[0034] In the form of piecewise cubic polynomial, the spline function formula on the interval [x i ,x i+1 ] is as follows:

[0035] S i (x)=βi +b i (xx i )+c i (xx i ) 2 +d i (xx i ) 3 (1);

[0036]

[0037] M0=0,M n =0(4);

[0038]

[0039] Among them, S i (x) is defined in the interval [x i ,x i+1 ] on the cubic polynomial function; S i (x i+1 ), S i+1 (x i+1 ) is the point x between the i-th segment and the i+1-th segment i+1 The function value at should be equal to the known value y i+1 β i is a constant term, equal to node x i The function value at β i =y i ; b i ,c i ,d i are the coefficients of the first, second and third terms; x is the independent variable and belongs to the interval [x i ,x i+1 ];x i is the position of the i-th node; M i For node x i The second-order derivative value at needs to be solved by the equation system; h i is the interval width, value is x i+1 -x i ;M0,M n is the second-order derivative of the first and last nodes, which is forced to be zero at the natural boundary; y i For node x i The known function value at ;

[0040] It should be noted that in the case of equally spaced time series monitoring data, cubic spline interpolation ensures the continuity of function values, first-order and second-order derivatives in adjacent intervals through the connection condition of formula (2). The second-order derivative values ​​M of all nodes are solved by the three-moment equation of formula (3) and the natural boundary condition of formula (4). iThe third-order polynomial b is calculated again through formula (5) i i i Finally, the interpolation result of any equidistant target point is generated by the segmented expression of formula (1).

[0041] The feature screening is performed on the equidistant time series monitoring data, the importance of each feature is calculated, and appropriate features are selected for subsequent clustering and prediction.

[0042] Specifically, the method for calculating the importance of features includes but is not limited to the random forest feature extraction method, and the formula for calculating the importance of features is:

[0043]

[0044] Wherein, N is the number of decision trees; t is the index of the decision tree, ranging from (1, 2, …, N); T t (x) is the prediction result of the input feature by the tth tree; j is the index of the feature, corresponding to the monitoring item; S j is the set of nodes using feature j to split; node is a specific node in the decision tree; n node is the number of samples participating in splitting in the node; ΔGin inode is the reduction of Gini impurity after node splitting; G i n inode is the Gini impurity of the node; N samples is the total number of training samples; α is the time decay coefficient, used to suppress the importance of deep nodes; depth node is the depth of the node in the tree; Importance j is the unnormalized importance value of feature j; R is the total number of categories of the target variable; p r is the proportion of samples of the rth category in the node.

[0045] It should be noted that when calculating the importance of features of multiple equidistant time series monitoring data, the outputs of multiple decision trees are first aggregated through the prediction function (6) of the random forest, then the importance of features is calculated based on formulas (7) and (8), and finally the importance of features such as arch settlement is mapped to a relative proportion through the normalization formula.

[0046] S3. Clustering analysis is performed on the equidistant time series monitoring data of multiple sections to complete the classification of the excavated sections.

[0047] Based on the importance of each feature, the distance between the initial independent clusters in hierarchical aggregation clustering is calculated, and the cluster classes are merged using the Group Average algorithm.

[0048] ​​Specifically, the calculated feature importance is normalized as the weight coefficient of the dimension distance calculation, ensuring that high importance features play a dominant role in clustering; the range normalization (divided by the maximum value - minimum value) is performed on each feature dimension to eliminate the dimension difference.

[0049] The equal-interval time series monitoring data of two sections A and B are used to calculate the cumulative distance between the initial independent clusters by multi-dimensional dynamic time warping, and the calculation formula is as follows:

[0050]

[0051] Wherein, D(i,j) represents the cumulative distance of time series A and B to point (i,j); represents the distance between the kth dimension and ; m is the number of dimensions; respectively, the value of the kth dimension; min{... is the minimum cumulative distance selected from three possible pre-alignment states.

[0052] After completing the DTW distance calculation between all sections, a Z×Z distance matrix (Z is the total number of sections) is generated as the input of the hierarchical aggregation clustering.

[0053] Further, the clustering method is hierarchical aggregation clustering, specifically:

[0054] First, initialize each feature vector as an independent cluster;

[0055] According to the cluster distance formula, the cumulative distance between all clusters is calculated;

[0056] The two clusters with the closest distance are merged; then repeat step until only one cluster is left.

[0057] The method of merging clusters includes but is not limited to GroupAverage algorithm, specifically:

[0058]

[0059] Wherein, ||x-y||2 represents the Euclidean distance between points x and y; X, Y are sets composed of several multi-dimensional points (or time series vectors), corresponding to two clusters; |X|, |Y| represents the number of points contained in the cluster, i.e. the size of the cluster; x∈X, y∈Y is to traverse each vector in the cluster;

[0060] The evaluation method is used to divide the clustering tree to divide the clustering results.

[0061] Specifically, the evaluation method includes but is not limited to Gap Statistic evaluation method, the expression is:

[0062] Gap(K) = E(log D K )-log D K (11);

[0063]

[0064] where Gap(K) is the optimal cluster number; D K is the total sum of squared deviations of all samples v i with respect to their cluster centers u j ; E(log D K ) is the expectation of D K for multiple random data sets at a given cluster number K; log D K is the D K calculated under real data; l is the total number of Kth cluster centers; n is the total number of data points in the Kth cluster (or the total number of points as a whole after summing all clusters); ||v i -u j || is the Euclidean distance or other distance measure between the sample and the center.

[0065] S4. Extract the shape center of each section classification and compare it with the pre-monitoring data of the predicted section to realize the classification of the predicted section;

[0066] The shape center of the target feature in each cluster is calculated using the barycenter average dynamic time warping, such as the shape center of the surrounding rock vault settlement time series data;

[0067] Specifically, the barycenter average dynamic time warping algorithm is as follows:

[0068]

[0069] where, is the updated average sequence (i.e., the shape center); Z is a normalization factor; M e is the set of aligned time points in the current iteration step; is the value taken by the average sequence at position i and the jth intra-cluster sequence alignment in the e-th iteration.

[0070] The similarity between the equidistant time series monitoring data of the target feature of the predicted section and the corresponding equidistant time series monitoring data in the shape center is compared using dynamic time warping, and the predicted section is classified into a known classification.

[0071] Specifically, the dynamic time warping algorithm is as follows:

[0072]

[0073] where, d(a i ,bj ) is the instantaneous distance between time points a i and b j , one-dimensional absolute difference |a i ,b j | or multi-dimensional Euclidean distance can be used.

[0074] It should be noted that dynamic time warping is a method for measuring and establishing the similarity between two time series, which is particularly effective in processing sequences with nonlinear time scaling or misalignment characteristics, and can more accurately capture the similarity of complex time series characteristics while having strong synchronization error resistance and good robustness. The barycenter average dynamic time warping is an algorithm for generating an average of multiple time series, given multiple time series, which first aligns these time series through dynamic time warping and is regarded as a point set in a multi-dimensional space. Then, an indicative average time series is calculated step by step to ensure that the average sequence can reflect the main trend and characteristics of all input sequences.

[0075] S5. Establishing a prediction model according to the classification of the predicted section to make long-term prediction of the tunnel surrounding rock deformation response.

[0076] The monitoring data is set with a rolling window length according to the shortest interval of adjacent section excavation, and is proportionally divided into a training set and a test set; specifically, the window length W is calculated according to the shortest excavation time interval (days) of adjacent sections:

[0077]

[0078] Where Δt is the equidistant time series sampling interval (e.g. 1 day or 1 hour).

[0079] For the shape center or the section sequence to be predicted in each cluster, samples are generated from time point 1 to T-W-H+1 (T is the total time series length and H is the prediction days); 80% of all samples in time order are used for training, and the remaining 20% are used for testing (verification) to ensure the subsequent evaluation of the model generalization performance.

[0080] Based on the training set, a bidirectional LSTM network is constructed, wherein the input is the historical window data and the output is the deformation prediction value for the maximum monitoring days in the future; specifically, the network structure includes: an input layer that accepts time series features with a shape of (W, F), where F is the feature dimension after screening; a bidirectional LSTM layer with d units in each direction, and the output is concatenated to form a vector with a length of 2d; a fully connected output layer that is linearly mapped to a prediction vector with a length of L.

[0081] In the test phase, new monitoring data is fused in real time to update the training set with a rolling window and dynamically adjust the model parameters to realize continuous prediction of the surrounding rock deformation;

[0082] Specifically, as the construction monitoring data continuously arrives, the latest day (or hour) of observation value is added to the time sequence, and the earliest day (or hour) is deleted; the new equidistant time sequence monitoring data is combined with the previous sequence to generate a new sample, which is appended to the end of the training set, while the earliest sample is removed to keep the number of windows unchanged; on the updated training set, the prediction model is fine-tuned with a smaller learning rate and a small number of iterations (such as 1-2 epochs) to quickly capture the changes in new data characteristics; using the fine-tuned prediction model, the latest window is used to predict the deformation response in the future H days; whenever new data arrives, repeat the steps of "real-time fusion → training set update → model fine-tuning → prediction" to form a continuous rolling prediction; finally, a long-term prediction sequence for each rolling period is obtained to provide dynamic early warning and decision support for the construction party.

[0083] The embodiment also provides a computer device suitable for the highway tunnel surrounding rock deformation prediction method, which comprises a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the highway tunnel surrounding rock deformation prediction method proposed in the above embodiment.

[0084] The computer device can be a terminal, and the computer device comprises a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device. In addition, the input device can be an external keyboard, touchpad or mouse, etc.

[0085] The embodiment also provides a storage medium on which a computer program is stored, the program being executed by a processor to implement the method for predicting deformation of surrounding rock of a highway tunnel as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk, or an optical disk.

[0086] To sum up, the method for predicting deformation of surrounding rock of a highway tunnel is implemented by: using a plurality of groups of completed monitoring section data, implementing section classification through hierarchical aggregation clustering, and predicting and identifying a deformation data type of a prediction section based on a dynamic time warping principle. The method can effectively fuse multi-source information and multi-type monitoring data, thereby greatly improving the accuracy of tunnel surrounding rock deformation prediction and the applicable range of a model. A rolling long-term prediction mechanism is adopted to realize continuous and dynamic prediction of deformation of a target section. The mechanism significantly prolongs the early warning response time, thereby providing a strong guarantee for engineering safety management and timely adjustment.

[0087] Embodiment 2

[0088] Reference Figures 2 to 5 To further verify the technical solution of the present application, the prediction method process of the method for predicting deformation of surrounding rock of a highway tunnel is given in the second embodiment of the present application.

[0089] The specific embodiment is a Xiaoyan tunnel project in Taishun, Zhejiang.

[0090] Firstly, the engineering geology and initial support of Zhejiang Taishun Xiaoyan tunnel project are investigated, including the geological conditions of tunnel excavation area, rock mass self-stability, and the investigation and surrounding rock classification check of initial support structure; after completing the geological survey, the monitoring points are reasonably arranged combined with the actual situation, the arrangement method of the monitoring points is: before the excavation of the tunnel portal section, concrete piles and leveling base points are buried at the top of the tunnel and along the tunnel center line longitudinally every 10-20 meters, the monitoring points are arranged transversely combined with the terrain and avoiding the slope excavation area; then the surrounding displacement and vault subsidence monitoring are carried out, the specific implementation mode is: the vault subsidence measuring points and the clearance change measuring points are arranged at each monitoring section, the vault subsidence measuring points are arranged near the vault axis, and the positions of the measuring points at each section are kept consistent and symmetrical as much as possible; at the same time, the intelligent online monitoring system is used to collect the surrounding displacement, vault subsidence, anchor rod axial force and step displacement data in real time, and continuous and complete monitoring data collection is realized.

[0091] Secondly, through the intelligent online monitoring system, real-time monitoring data (in this embodiment, mainly the vault subsidence and the displacement of the lower step) are collected, and the monitoring data are equidistantly processed by using cubic spline interpolation.

[0092] Then, the monitoring data are selected based on the random forest theory, taking the 29 groups of section data of the left line of the excavated tunnel as an example, the feature selection standard of hierarchical agglomerative clustering is that the importance is greater than 0.1, and the feature selection standard of time series prediction is that the importance is greater than 0.01. The features meeting these standards will be retained, and the remaining features will be discarded, because hierarchical agglomerative clustering relies on significant features to ensure the discrimination between data points.

[0093] The distance between the initial independent clusters in hierarchical agglomerative clustering is calculated by using the multi-dimensional dynamic time warping algorithm, the clusters are merged by using the Group Average algorithm, the hierarchical clustering tree is divided by using the Gap Statistic evaluation method, and the clear clustering result is divided. The shape center of the vault subsidence feature of each cluster is extracted by using the barycenter average dynamic time warping algorithm, and the similarity of the prediction section and the previous data of the shape center is compared by using dynamic time warping, so as to realize the classification of the prediction section.

[0094] Finally, the surrounding rock deformation prediction and stability analysis are carried out. The bidirectional long short-term memory network rolling prediction model is trained by using the left line monitoring data in each cluster, the corresponding bidirectional long short-term memory network model is used for prediction according to the cluster divided by the prediction section, and the surrounding rock stability analysis of the right line prediction section is carried out.

[0095] Figure 2The results of calculating the feature importance using the random forest method. It can be seen that the features of the cluster analysis and the time series prediction input are different in this example, and at the same time, it is detected that there are multiple input features in the project which have less influence on the vault settlement. Therefore, this method can screen important features to improve the stability and accuracy of model prediction.

[0096] Figure 3 The shape centers of each cluster class extracted by the center of gravity average dynamic time warping algorithm after the division of the 29 groups of left line sections that have been excavated using the Gap Statistic evaluation method. It can be seen that this method clearly distinguishes the key features such as surrounding rock stability time, early deformation rate and final settlement value. Therefore, this method can effectively cluster and extract the shape center of the monitoring data of multiple sections.

[0097] Figure 4 The ablation experiment results of 13 groups of right line section monitoring data, and the results after removing each step. It can be seen that the model combines the feature importance analysis of random forest and hierarchical aggregation clustering, which significantly improves the prediction performance, and its prediction effect is the best. Remove the random forest feature importance extraction part, and its prediction effect is second. This shows that the advantage of random forest in feature selection affects the final prediction result. After removing the hierarchical clustering part, its prediction effect is third, which shows that clustering is crucial to improve the stability and accuracy of the model. The bidirectional long short-term memory network model with single feature input performs weakly in capturing the features of time series data, and its prediction effect is the worst. Therefore, it can be shown that the model combining feature selection and clustering can effectively improve the prediction accuracy of the model.

[0098] Figure 5 The long-term prediction results of individual sections. It can be seen that in terms of predicting the final value of settlement, fitting degree and surrounding rock basic stability time, the results of using feature selection and hierarchical clustering are the best. It can be shown that in this embodiment, the model combining feature selection and clustering can effectively improve the prediction accuracy of the model.

Claims

1. A method for predicting deformation of surrounding rock in a highway tunnel, characterized by: include, Collect existing information and determine the monitoring layout plan, and collect on-site monitoring data through real-time feedback technology; Processing multiple types of monitoring data and screening suitable input features; Conduct cluster analysis on monitoring data of multiple sections to complete the classification of excavated sections; Extract the shape center of each cross-section classification and compare it with the previous monitoring data of the predicted cross-section to achieve the classification of the predicted cross-section; A prediction model is established based on the classification of prediction sections to make long-term predictions on the deformation response of tunnel surrounding rock.

2. The method for predicting deformation of surrounding rock in a highway tunnel according to claim 1, wherein: The process of collecting existing data and determining a monitoring arrangement plan, and collecting on-site monitoring data through real-time monitoring data feedback technology, includes the following steps: Investigate the geological conditions, rock mass stability, and initial support structure of the tunnel excavation area, verify the surrounding rock classification, and record relevant information; After completing the geological and support surveys, arrange monitoring points in accordance with the actual situation on site; Carry out monitoring of surrounding displacement and arch subsidence, and adopt online monitoring to collect monitoring data.

3. The method for predicting deformation of surrounding rock in a highway tunnel according to claim 2, wherein: The processing of multiple types of monitoring data and screening of appropriate input features includes the following steps: Perform interpolation processing on the monitoring data to generate equally spaced time series monitoring data; Perform feature screening on equally spaced time series monitoring data and calculate the importance of each feature.

4. The method for predicting deformation of surrounding rock in a highway tunnel according to claim 3, wherein: The cluster analysis of the monitoring data of multiple sections to complete the classification of the excavated sections includes the following steps: Based on the importance of each feature, the distance between the initial independent clusters in the hierarchical clustering is calculated, and the clusters are merged using the GroupAverage algorithm; Use the evaluation method to split the cluster tree and divide the clustering results.

5. The method for predicting deformation of surrounding rock in a highway tunnel according to claim 4, wherein: The hierarchical aggregation clustering is specifically as follows: First, initialize, each eigenvector is treated as an independent cluster; According to the formula of inter-cluster distance, calculate the cumulative distance between all clusters; Merge the two closest clusters; Repeat until only one cluster remains.

6. The method for predicting deformation of surrounding rock in a highway tunnel according to claim 5, wherein: The process of extracting the shape center of each cross-section classification and comparing it with the previous monitoring data of the predicted cross-section to achieve the classification of the predicted cross-section includes the following steps: Use centroid-averaged dynamic time warping to calculate the shape center of the target feature in each cluster; Dynamic time warping is used to compare the similarity between the equally spaced time series monitoring data of the target features of the predicted section and the corresponding equally spaced time series monitoring data in the shape center, and the predicted section is classified into the known classification.

7. The method for predicting deformation of surrounding rock in a highway tunnel according to claim 6, wherein: The establishment of the prediction model includes: The monitoring data is divided into training set and test set in proportion by setting the rolling window length according to the shortest interval between adjacent section excavations. A bidirectional LSTM network is constructed based on the training set, where the input is the historical window data and the output is the deformation prediction value of the maximum number of monitoring days in the future; During the testing phase, new monitoring data are integrated in real time, the training set is updated with a rolling window, and the model parameters are dynamically adjusted to achieve continuous prediction of surrounding rock deformation.

8. The method for predicting deformation of surrounding rock in a highway tunnel according to claim 7, wherein: The monitoring data include peripheral displacement, arch crown subsidence, anchor rod axial force and step displacement.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for predicting deformation of surrounding rock of a highway tunnel according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting deformation of surrounding rock of a highway tunnel according to any one of claims 1 to 8 are implemented.

Images