Tunnel ellipticity dynamic monitoring and early warning method and system based on time-series point cloud

Abstract

The present invention relates to the technical field of tunnel ellipticity dynamic monitoring and early warning methods, in particular to a tunnel ellipticity dynamic monitoring and early warning method and system based on time-series point cloud, which obtains tunnel point cloud data; based on the tunnel point cloud data, performs point cloud registration, tunnel central axis extraction and fitting processing to obtain ellipticity; according to monitoring data of different periods, establishes a mathematical model of the relationship between the monitoring period and the ellipticity; based on the current deformation data, dynamically predicts the future ellipticity and draws the early warning development trajectory; according to the future ellipticity development trend, performs deformation early warning. The method of the present invention firstly realizes the precise alignment of point cloud data of different periods through point cloud registration technology, laying the foundation for subsequent analysis. The innovative ellipticity calculation method takes into account the tunnel central axis and contour features, and can more accurately reflect the geometric deformation of the tunnel. By establishing a mathematical model of the relationship between the monitoring period and the ellipticity, the method realizes the dynamic prediction of the future ellipticity.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel ellipticity dynamic monitoring and early warning methods, in particular to a tunnel ellipticity dynamic monitoring and early warning method and a system thereof based on a time-series point cloud. Background Art

[0002] With the acceleration of urbanization and the increasing use of underground space, tunnel engineering plays an increasingly important role in modern infrastructure construction. However, as a complex underground structure, the safety and stability of tunnels have always been the focus of attention in the engineering community. In particular, the deformation of tunnels not only directly affects the service life of tunnels, but may also endanger driving safety and the surrounding environment. Therefore, monitoring and early warning of tunnel deformation have become a key link in tunnel operation and maintenance.

[0003] Traditional tunnel deformation monitoring methods mainly rely on manual measurement and fixed sensors. Although these methods can reflect the deformation of the tunnel to a certain extent, they have many limitations. First, manual measurement methods are labor-intensive and inefficient, and it is difficult to achieve large-scale, high-frequency monitoring. Secondly, although fixed sensors can achieve continuous monitoring, the installation locations are limited and cannot fully reflect the overall deformation of the tunnel. In addition, these methods can often only obtain discrete point data, making it difficult to accurately describe the overall geometric changes of the tunnel.

[0004] In recent years, with the development of 3D laser scanning technology, point cloud data has been widely used in the field of tunnel monitoring. Point cloud data can collect geometric information of tunnels with high precision and high density, providing new possibilities for tunnel deformation analysis. However, there are still some problems with the existing point cloud-based tunnel monitoring methods. First, most methods only focus on the processing of single scan data and lack in-depth analysis of time series changes. Secondly, in terms of ellipticity calculation, existing methods often use simple geometric fitting algorithms, which are difficult to cope with complex deformation situations. Furthermore, existing early warning systems are mostly based on static thresholds and cannot adapt to the dynamic characteristics of tunnel deformation.

[0005] In addition, with the rapid development of artificial intelligence technology, machine learning methods have shown great potential in the field of structural health monitoring. However, in tunnel ellipticity monitoring, the application of advanced technologies such as deep learning and reinforcement learning is still in its infancy. How to effectively use these technologies to improve the accuracy of monitoring and the timeliness of early warning is an important challenge facing current research.

[0006] In view of the above problems, it is urgent to make full use of time-series point cloud data, combined with advanced data processing and artificial intelligence technology, to achieve a new method for dynamic monitoring and early warning of tunnel ellipticity. This method should be able to comprehensively and accurately capture the deformation characteristics of the tunnel, adaptively adjust the monitoring strategy, and provide reliable early warning information. Summary of the invention

[0007] The present invention aims at the above problems existing in the prior art and proposes a tunnel ellipticity dynamic monitoring and early warning method and system based on time-series point cloud. The method realizes high-precision dynamic monitoring and intelligent early warning of tunnel ellipticity by innovatively combining point cloud processing technology, time-series analysis and deep learning algorithm.

[0008] The present invention proposes a tunnel ellipticity dynamic monitoring and early warning method based on time-series point cloud, comprising:

[0009] The acquisition steps include:

[0010] Obtain tunnel point cloud data;

[0011] The processing steps include:

[0012] Based on the tunnel point cloud data, point cloud registration, tunnel centerline extraction and fitting processing are performed to obtain ellipticity;

[0013] According to the monitoring data of different periods, a mathematical model of the relationship between monitoring period and ellipticity is established;

[0014] Based on the current deformation data, the future ellipticity is dynamically predicted and the early warning development trajectory is drawn;

[0015] The output steps include:

[0016] According to the future development trend of ovality, deformation warning is carried out.

[0017] Preferably, the point cloud registration in the processing step specifically includes:

[0018] The optimal transformation matrix is ​​determined by minimizing the root mean square error and the following two objective functions:

[0019]

[0020] Among them, R i is the rotation angle of the transformation matrix of adjacent periods, t is the movement vector, p i and q i is the registered point cloud of the adjacent period, M is the number of corresponding points of the registered point cloud, O i is the point sequence of the overlapping area of ​​the point cloud in two adjacent periods, n i O i The normal vector, w i is the corresponding point with O i The scalar value obtained by multiplying the normal vector of i For p i The coordinate difference with the corresponding point.

[0021] Preferably, the ellipticity calculation in the processing step specifically includes:

[0022] Project each point in the point column of the tunnel centerline onto the xy plane to obtain a point column, and compare it with the tunnel contour point column. Calculate the average value of the absolute value of the horizontal coordinate difference between the corresponding points of the two point columns and multiply it by the number of point columns to obtain the absolute deviation of the lateral offset;

[0023] The absolute deviation of the lateral offset is judged by a threshold value, and the points with an absolute deviation greater than the threshold value are determined as outliers, and the outliers are removed from the point list;

[0024] Substitute the point sequence after removing the outliers into the ellipse equation to calculate the tunnel ellipticity.

[0025] Preferably, the mathematical model establishment in the processing step specifically includes:

[0026] Calculate the ellipticity vs. time to form a logarithmic function of the fitted data:

[0027] y=aln(t)+b+ε

[0028] Among them, y is the tunnel ellipticity series, that is, the ellipticity data of the tunnel at different periods, t is the monitoring time series, that is, the time data of the tunnel at different periods, a and b are the parameters to be fitted, and ε is the relative error of the tunnel ellipticity and time fitting.

[0029] Preferably, the dynamic prediction in the processing step specifically includes:

[0030] Based on the calculated slope b, the deformation rate is calculated, and based on this, the ellipticity value at the future time t is predicted:

[0031] y t =aln(t)+b

[0032] Among them, y t is the predicted ellipticity value at time t;

[0033] Calculate the residual error of the deformation based on the calculated ovality value at time t;

[0034] Draw the deformation development trend line to obtain the deformation development trend.

[0035] Preferably, the deformation warning in the output step specifically includes:

[0036] The warning message is obtained by comparing the threshold value of the warning ellipticity obtained by historical fitting with the measured ellipticity;

[0037] The residual error calculated by the historical fitting curve and the measured ellipticity is used to make a threshold judgment on the future ellipticity development trend;

[0038] If the ellipticity change rate exceeds the warning threshold in the future, a warning message will be sent to the management personnel, otherwise no warning message will be sent, and the feedback of the warning results and the real-time update of the ellipticity will be completed.

[0039] Preferably, the warning threshold is dynamically modified, specifically including:

[0040] Through continuous training, the ellipticity curve based on different deformation rates is obtained;

[0041] If the measured deformation and predicted residuals are both around 10% after monitoring time T, the ellipticity curve is compared with the measured value and the threshold is corrected;

[0042] The warning threshold of ellipticity prediction uses the following formula to judge the threshold of ellipticity change rate:

[0043]

[0044] Among them, ε i is the ellipticity corresponding to the predicted time point i, ε i+1 is the ellipticity corresponding to the predicted time point i+1, and λ is the warning threshold.

[0045] As a preference, it also includes:

[0046] Train the deep reinforcement learning model, and the ellipticity of the model at the initial moment is the measured ellipticity at the monitoring time t0;

[0047] During model training, the model predicts from time t0 to time tend based on the reward model and calculates the reward:

[0048] R is the total reward value, n is the total number of monitoring frequencies, i is the index of the monitoring frequency, from 1 to n, w i is the weight of the i-th monitoring frequency, f is the reward function, a i is the monitoring frequency adopted by the model at the i-th time, e i is the relative error between the predicted ellipticity and the measured ellipticity for the i-th time, v i is the change rate of ellipticity predicted for the i-th time, c i is the monitoring frequency currently adopted by the model during the i-th prediction;

[0049] The rewards obtained are used to train the deep reinforcement learning model to learn to predict the ellipticity value, monitor the frequency and obtain the model.

[0050] Preferably, the obtaining step further comprises:

[0051] Get the ellipticity parameters of the tunnel vault;

[0052] A dataset based on time series deformation is constructed for each ellipticity parameter;

[0053] Input a sample of the dataset into the pre-trained deep reinforcement learning model to determine the monitoring frequency of the sample;

[0054] A scan is performed at the monitoring frequency to obtain the next data set sample in the data set.

[0055] The tunnel ellipticity dynamic monitoring and early warning system based on time-series point cloud includes:

[0056] A point cloud data acquisition module is used to scan the tunnel surface at different parts of the tunnel to obtain corresponding point cloud data, and then obtain several dome point cloud data groups;

[0057] An ellipticity calculation module is used to calculate corresponding ellipticity indices for a plurality of dome cloud data groups based on a preset ellipticity calculation method, and form a corresponding ellipticity index sequence;

[0058] The ellipticity monitoring neural network module is used to calculate the corresponding ellipticity index sequence according to the point cloud data of the vault to be measured, and to perform error correction on the sequence;

[0059] The reinforcement learning model module is used to adjust the monitoring frequency of the point cloud data acquisition module according to the timestamp corresponding to the ellipticity index sequence;

[0060] The time series analysis module is used to analyze the local deformation trend of the tunnel based on the monitoring frequency corresponding to the ellipticity index sequence and construct the ellipticity trend function;

[0061] The early warning module is used to judge the change rate of the ellipticity index in any point cloud obtained in the time series analysis module, and issue an early warning through the early warning terminal if it exceeds the preset threshold;

[0062] Wherein, the ellipticity index sequence and the ellipticity trend function are limited by the monitoring frequency corresponding to the ellipticity index sequence.

[0063] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0064] Specifically, the method of the present invention firstly realizes the precise alignment of point cloud data at different periods through point cloud registration technology, laying the foundation for subsequent analysis. The innovative ellipticity calculation method takes into account the central axis and contour features of the tunnel, and can more accurately reflect the geometric deformation of the tunnel. By establishing a mathematical model of the relationship between the monitoring period and the ellipticity, the method realizes the dynamic prediction of future ellipticity. It is particularly worth mentioning that the present invention introduces a deep reinforcement learning model, which not only improves the accuracy of the prediction, but also can adaptively adjust the monitoring frequency, thereby realizing the optimal configuration of monitoring resources.

[0065] The method and system of the present invention show significant advantages in practical applications. First, compared with traditional methods, the present invention can more comprehensively and accurately reflect the deformation state of the tunnel, effectively avoiding omissions that may be caused by local monitoring. Secondly, dynamic prediction and adaptive monitoring strategies greatly improve the flexibility and efficiency of the system, which not only ensures the monitoring quality but also reduces operating costs. Furthermore, the early warning mechanism based on deep learning can identify potential risks in a timely manner and provide reliable decision-making support for tunnel managers.

[0066] In addition, the method of the present invention has good scalability and versatility. Although it is mainly aimed at tunnel ellipticity monitoring, its core idea and technical framework can be easily extended to other types of structural health monitoring fields. This provides a new technical path for the intelligent and refined management of infrastructure safety.

[0067] In general, the tunnel ellipticity dynamic monitoring and early warning method and system based on time-series point cloud provided by the present invention not only solves many problems existing in the prior art, but also makes significant progress in monitoring accuracy, early warning timeliness and resource utilization efficiency. This is of great significance for improving tunnel operation safety, extending service life and reducing maintenance costs, and also provides a valuable reference for the development of intelligent infrastructure management. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 The figure is a flow chart of the overall method of the present invention.

[0069] Figure 2 It is a detailed logic block diagram of the point cloud data acquisition module of the present invention.

[0070] Figure 3 It is a logic block diagram of the ellipticity measurement module of the present invention.

[0071] Figure 4 It is a logic block diagram of the ellipticity monitoring neural network module of the present invention.

[0072] Figure 5 It is a logical block diagram of the reinforcement learning model module of the present invention.

[0073] Figure 6 It is a logic block diagram of the timing analysis module of the present invention.

[0074] Figure 7 It is a logic block diagram of the early warning module of the present invention. DETAILED DESCRIPTION

[0075] Please refer to Figure 1-7The present invention provides a method and system for dynamic monitoring and early warning of tunnel ellipticity based on time-series point cloud. The method obtains tunnel point cloud data, performs point cloud registration, tunnel central axis extraction and fitting processing, and obtains ellipticity, thereby realizing dynamic monitoring of tunnel deformation. Furthermore, the method also establishes a mathematical model of the relationship between the monitoring cycle and the ellipticity, dynamically predicts the future ellipticity based on the current deformation data, draws the early warning development trajectory, and finally performs deformation early warning according to the future ellipticity development trend.

[0076] Specifically, the method of the present invention first obtains tunnel point cloud data. These point cloud data usually come from equipment such as laser scanners, which can accurately capture the three-dimensional geometric information of the tunnel surface. Preferably, a mobile laser scanning system can be used to obtain high-density and high-precision tunnel point cloud data.

[0077] After acquiring the point cloud data, this method performs point cloud registration processing. Point cloud registration is the process of unifying point cloud data collected at different times into the same coordinate system, which is the basis for subsequent analysis. The present invention adopts an innovative registration method to determine the optimal transformation matrix by minimizing the root mean square error of two objective functions. The two objective functions are:

[0078] The optimal transformation matrix is ​​determined by minimizing the root mean square error and the following two objective functions:

[0079]

[0080] Among them, R i is the rotation angle of the transformation matrix of adjacent periods, t is the movement vector, p i and q i is the registered point cloud of the adjacent period, M is the number of corresponding points of the registered point cloud, O i is the point sequence of the overlapping area of ​​the point cloud in two adjacent periods, n i O i The normal vector, w i is the corresponding point with O i The scalar value obtained by multiplying the normal vector of i For p i The coordinate difference with the corresponding point. α and β are the coefficients of the objective function.

[0081] By minimizing the root mean square error of these two objective functions, the optimal transformation matrix can be obtained, thereby achieving high-precision point cloud registration. In practical applications, the iterative closest point (ICP) algorithm can be used to solve this optimization problem. Preferably, the values ​​of α and β can be adjusted according to the specific tunnel environment and point cloud characteristics. Usually, the value range of α is 0.6-0.8, and the value range of β is 0.2-0.4.

[0082] After completing the point cloud registration, the method extracts and fits the tunnel central axis. The central axis is an important feature that describes the geometry of the tunnel, and its extraction accuracy directly affects the accuracy of the subsequent ellipticity calculation. The present invention adopts a central axis extraction method based on cross-section fitting. Specifically, the registered point cloud data is first sliced ​​along the longitudinal direction of the tunnel, and each slice is fitted with a circular or elliptical shape, and then all the fitting center points are connected to form an initial central axis. Then, the initial central axis is smoothed to obtain the final tunnel central axis.

[0083] Based on the extracted central axis, this method calculates the ellipticity. Ellipticity is an important indicator to measure the degree to which the tunnel section deviates from the ideal circle. The present invention proposes an innovative ellipticity calculation method, and the specific steps are as follows:

[0084] 1. Project each point in the point sequence of the tunnel centerline onto the xy plane to obtain a point sequence, which corresponds to the tunnel contour point sequence.

[0085] 2. Calculate the average of the absolute values ​​of the horizontal coordinate differences between the corresponding points in the two columns of point clouds, and multiply it by the number of point columns to obtain the absolute deviation of the lateral offset.

[0086] 3. Perform a threshold judgment on the absolute deviation of the lateral offset, determine the points whose absolute deviation is greater than the threshold as outliers, and remove these outliers from the point sequence. Preferably, the threshold can be set to 2-3 times the mean of the absolute deviation of the lateral offset.

[0087] 4. Substitute the point sequence after removing the outliers into the elliptic equation to calculate the tunnel ellipticity.

[0088] This ellipticity calculation method can effectively reduce the influence of abnormal points and improve the accuracy and stability of the calculation results. In practical applications, the threshold value can be adjusted according to the specific conditions of the tunnel. For example, for tunnel sections with more serious deformation, the threshold can be appropriately lowered to capture more deformation information.

[0089] Through the above steps, the method realizes the accurate calculation of the tunnel ellipticity. This provides a reliable data basis for subsequent deformation analysis and early warning. The method of the present invention can not only accurately reflect the current state of the tunnel, but also capture the dynamic trend of tunnel deformation through time series analysis, thereby providing important support for tunnel safety management. After obtaining the tunnel ellipticity data, the method of the present invention further establishes a mathematical model of the relationship between the monitoring period and the ellipticity. This step is crucial for accurately predicting future deformation trends. Specifically, the method uses a logarithmic function to fit the relationship between ellipticity and time, and its mathematical expression is as follows:

[0090] y=aln(t)+b+ε

[0091] Among them, y is the tunnel ellipticity series, that is, the ellipticity data of the tunnel at different periods, t is the monitoring time series, that is, the time data of the tunnel at different periods, a and b are the parameters to be fitted, and ε is the relative error of the tunnel ellipticity and time fitting.

[0092] The selection of logarithmic function as the fitting model has a profound engineering background. In tunnel engineering, deformation usually develops rapidly in the early stage of construction and then gradually stabilizes. The logarithmic function can just describe this nonlinear change trend. Preferably, the least squares method can be used to solve the parameters a and b to minimize the fitting error ε.

[0093] Based on the established mathematical model, the method of the present invention can dynamically predict the future ellipticity. Specifically, first, the deformation rate is calculated based on the calculated slope b. The deformation rate reflects the speed of tunnel deformation and is an important basis for prediction. Subsequently, the method predicts the ellipticity value at the future time t based on this, and the prediction formula is:

[0094] y t =aln(t)+b

[0095] Among them, y t is the predicted ellipticity value at time t

[0096] In order to evaluate the accuracy of the prediction, the method also calculates the residual error of the deformation. The residual error is defined as the difference between the predicted value and the measured value, which can be used to measure the performance of the prediction model. Preferably, the root mean square error (RMSE) can be used as a metric for the residual error.

[0097] After obtaining the future ellipticity prediction value, the method of the present invention draws a deformation development trend line to intuitively display the future trend of tunnel deformation. This trend line not only includes the existing monitoring data, but also extends to future time points, providing tunnel managers with a clear deformation development picture.

[0098] Based on the above prediction results, the method of the present invention finally realizes deformation early warning. The early warning process mainly includes the following steps:

[0099] 1. Get the warning message by comparing the threshold value of the historical fitting warning ellipticity with the measured ellipticity. The historical fitting warning ellipticity here is the ellipticity value obtained by fitting based on historical data, while the measured ellipticity is the latest monitoring result.

[0100] 2. The residual error calculated by the historical fitting curve and the measured ellipticity is used to determine the threshold value of the future ellipticity development trend. This step takes into account the uncertainty of the prediction model and improves the reliability of the early warning.

[0101] 3. If the future ellipticity change rate exceeds the warning threshold, a warning message is sent to the management personnel. Otherwise, no warning message is sent. The warning threshold here can be set according to the specific situation of the tunnel, usually 2-3 times the standard deviation of the ellipticity change rate.

[0102] 4. Complete the feedback of early warning results and real-time update of ellipticity. This step ensures the continuous and effective operation of the early warning system.

[0103] The method of the present invention realizes dynamic early warning of tunnel deformation in this way. Compared with the traditional static threshold early warning, this early warning method based on time series analysis and dynamic prediction can more accurately capture the development trend of tunnel deformation and identify potential risks in advance.

[0104] Preferably, the method of the present invention can also dynamically modify the warning threshold according to the actual monitoring results. For example, a monitoring period (such as 3 months) can be set, during which the measured deformation and the predicted residual are compared. If the difference between the two is about 10%, it is considered that the current warning model performs well; otherwise, the warning threshold needs to be adjusted. This adaptive mechanism further improves the accuracy and reliability of the warning system.

[0105] In general, the method for dynamic monitoring and early warning of tunnel ellipticity based on time-series point cloud provided by the present invention realizes comprehensive, accurate, dynamic monitoring and early warning of tunnel deformation by combining point cloud processing technology, time-series analysis and predictive modeling. This method can not only detect abnormal deformation of the tunnel in a timely manner, but also predict future deformation trends, providing strong technical support for the safe management and maintenance of the tunnel. In a preferred embodiment of the present invention, the dynamic correction of the early warning threshold is a key link, which enables the early warning system to continuously optimize itself over time. Specifically, the method obtains ellipticity curves based on different deformation rates through continuous training. These curves reflect the changing pattern of tunnel ellipticity at different deformation speeds.

[0106] Preferably, the method sets a monitoring time T, which can usually be selected as 3-6 months. During this time period, the system will continue to compare the measured deformation with the predicted residual. If after the monitoring time T, the measured deformation and the predicted residual are both around 10%, it is considered that the current prediction model performs well. At this time, the method will compare the ellipticity curve with the measured value and make appropriate corrections to the threshold. This approach ensures that the early warning system can adapt to the dynamic changes in the deformation characteristics of the tunnel.

[0107] The method of the present invention adopts an innovative ellipticity prediction and warning threshold judgment formula:

[0108]

[0109] Among them, εi is the ellipticity corresponding to the predicted time point i, ε i+1 is the ellipticity corresponding to the predicted time point i+1, and λ is the warning threshold. This formula calculates the relative change rate of the ellipticity. When this change rate exceeds the preset threshold λ, the system will trigger a warning.

[0110] Preferably, the value of λ can be adjusted according to the specific conditions of the tunnel. For a newly built tunnel, a smaller λ value, such as 0.05-0.1, can be set in the initial stage to capture smaller deformations; for a stable tunnel that has been in operation for many years, the λ value can be appropriately increased, such as 0.1-0.2, to reduce false alarms.

[0111] The method of the present invention also introduces a deep reinforcement learning model to further improve the accuracy of prediction and warning. In the initial stage of the model, the ellipticity is taken as the measured ellipticity at the monitoring time t0. This provides a reliable starting point for the model.

[0112] During the model training process, this method predicts from time t0 to time tend based on the reward model. The reward calculation formula is as follows:

[0113]

[0114] R is the total reward value, n is the total number of monitoring frequencies, i is the index of the monitoring frequency, from 1 to n, w i is the weight of the i-th monitoring frequency, f is the reward function, a i is the monitoring frequency adopted by the model at the i-th time, e i is the relative error between the predicted ellipticity and the measured ellipticity for the i-th time, v i is the change rate of ellipticity predicted for the i-th time, c i is the monitoring frequency currently adopted by the model during the i-th prediction.

[0115] This reward function takes into account prediction accuracy, rationality of monitoring frequency, and resource utilization efficiency. By maximizing this reward function, the deep reinforcement learning model can learn the optimal prediction strategy and monitoring frequency.

[0116] Preferably, a deep Q network (DQN) or a policy gradient method can be used to implement this deep reinforcement learning model. The input of the model includes historical ellipticity data, current monitoring frequency and other features, and the output is the next monitoring frequency selection and ellipticity prediction value.

[0117] Through repeated training and optimization, the deep reinforcement learning model can continuously improve its ability to predict ellipticity values ​​and select the optimal monitoring frequency. Compared with the traditional fixed-interval monitoring method, this method can allocate monitoring resources more flexibly, effectively reducing monitoring costs while ensuring the accuracy of early warnings.

[0118] The method of the present invention realizes high-precision dynamic monitoring and early warning of tunnel ellipticity by combining traditional mathematical modeling and advanced machine learning technology. This method can not only accurately capture the slight changes in tunnel deformation, but also adaptively adjust the monitoring strategy, providing strong technical support for tunnel safety management. In another preferred embodiment of the present invention, the acquisition step of the method is further refined to improve the accuracy and efficiency of data acquisition. First, the method obtains the ellipticity parameters of the tunnel vault. These parameters generally include major axis length, minor axis length, inclination angle, etc., which together describe the geometric characteristics of the tunnel cross section. Preferably, a high-precision three-dimensional laser scanner can be used for data acquisition, and the scanning accuracy can reach the millimeter level to ensure the accuracy of the acquired ellipticity parameters.

[0119] For each ellipticity parameter, this method constructs a dataset based on time series deformation. This dataset contains the values ​​of the ellipticity parameters at different time points, forming a time series. Preferably, the time span of the dataset should cover at least a complete seasonal cycle to capture possible seasonal changes. For example, 12 consecutive months of data can be selected as a complete dataset.

[0120] Next, the method inputs the sample of the data set into a pre-trained deep reinforcement learning model. This model is one of the core components of the present invention, which can intelligently determine the optimal monitoring frequency based on historical data. The input of the model includes historical ellipticity parameter sequences, environmental factors (such as temperature, humidity) and other features, and the output is the optimal monitoring frequency for the sample.

[0121] Preferably, the deep reinforcement learning model can adopt a double Q network (Double DQN) structure to reduce the over-estimation problem of Q value. The reward function of the model can be designed as a weighted sum of monitoring accuracy and monitoring cost to strike a balance between accuracy and efficiency.

[0122] According to the monitoring frequency output by the model, this method scans at this frequency to obtain the next data set sample in the data set. This adaptive data collection strategy can minimize unnecessary monitoring while ensuring data quality, thereby improving the operating efficiency of the entire system.

[0123] The present invention also provides a tunnel ellipticity dynamic monitoring and early warning system based on time-series point cloud corresponding to the above method. The system includes multiple functional modules, each module is responsible for a specific task, and together realizes the dynamic monitoring and early warning of the tunnel ellipticity.

[0124] The point cloud data acquisition module 1 is used to scan the tunnel surface at different parts of the tunnel to obtain corresponding point cloud data. Preferably, the module can use a vehicle-mounted mobile laser scanning system to achieve fast and high-precision scanning of the tunnel. Through multiple scans, the point cloud data acquisition module 1 can obtain several vault vertex cloud data groups to provide basic data for subsequent analysis.

[0125] Based on the preset ellipticity calculation method, the ellipticity calculation module 2 calculates the corresponding ellipticity index for multiple arch vertex cloud data groups. The ellipticity index here can include parameters such as major axis length, minor axis length, flattening, etc. By analyzing multiple data groups, the ellipticity calculation module 2 constructs a corresponding ellipticity index sequence, which reflects the changing trend of the tunnel ellipticity over time.

[0126] The ellipticity monitoring neural network module 3 is one of the core components of the system. This module uses deep learning technology to calculate the corresponding ellipticity index sequence based on the point cloud data of the vault to be measured. Compared with traditional methods, the neural network model can automatically learn the characteristics of point cloud data and improve the accuracy and robustness of ellipticity calculation. In addition, the ellipticity monitoring neural network module 3 can also perform error correction on the calculated sequence to further improve the reliability of the monitoring results.

[0127] The reinforcement learning model module 4 is responsible for dynamically adjusting the monitoring frequency of the point cloud data acquisition module 1 according to the timestamp corresponding to the ellipticity index sequence. This adaptive monitoring strategy can optimize resource utilization and improve the overall efficiency of the system while ensuring the monitoring quality. Preferably, the reinforcement learning model can adopt a deep Q network (DQN) or an advantage action commentator (A2C) algorithm to better handle continuous state space.

[0128] The time series analysis module 5 analyzes the deformation trend of the local tunnel based on the ellipticity index sequence and the corresponding monitoring frequency, and constructs the ellipticity trend function. This module uses time series analysis techniques, such as the ARI MA model or the long short-term memory network (LSTM), which can effectively capture the long-term trend and short-term fluctuation of tunnel deformation.

[0129] Finally, the early warning module 6 is responsible for judging the change rate of the ellipticity index in any point cloud obtained in the time series analysis module 5. If the change rate exceeds the preset threshold, the early warning module 6 will issue an alarm through the early warning terminal. The design of this module fully considers the dynamic characteristics of tunnel deformation and can detect potential safety hazards in a timely manner.

[0130] It is worth noting that the ellipticity index sequence and ellipticity trend function in this system are limited by the monitoring frequency corresponding to the ellipticity index sequence. This design ensures the close connection between the various modules of the system, making the entire monitoring and early warning process a closed loop, thereby achieving efficient, accurate, and dynamic monitoring and early warning of tunnel ellipticity.

[0131] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A tunnel ellipticity dynamic monitoring and early warning method based on time-series point cloud, characterized by ,include: The acquisition steps include: Obtain tunnel point cloud data; The processing steps include: Based on the tunnel point cloud data, point cloud registration, tunnel centerline extraction and fitting processing are performed to obtain ellipticity; According to the monitoring data of different periods, a mathematical model of the relationship between monitoring period and ellipticity is established; Based on the current deformation data, the future ellipticity is dynamically predicted and the early warning development trajectory is drawn; The output steps include: According to the future development trend of ovality, deformation warning is carried out.

2. The method for dynamic monitoring and early warning of tunnel ellipticity based on time-series point cloud according to claim 1 is characterized in that ,The point cloud registration in the processing step specifically includes: The optimal transformation matrix is ​​determined by minimizing the root mean square error and the following two objective functions: Among them, R i is the rotation angle of the transformation matrix of adjacent periods, t is the movement vector, p i and q i is the registered point cloud of the adjacent period, M is the number of corresponding points of the registered point cloud, O i is the point sequence of the overlapping area of ​​the point cloud in two adjacent periods, n i O i The normal vector, w i is the corresponding point with O i The scalar value obtained by multiplying the normal vector of i For p i The coordinate difference with the corresponding point.

3. The method for dynamic monitoring and early warning of tunnel ellipticity based on time-series point cloud according to claim 1 is characterized in that , the ellipticity calculation in the processing step specifically includes: Project each point in the point column of the tunnel centerline onto the xy plane to obtain a point column, and compare it with the tunnel contour point column. Calculate the average value of the absolute value of the horizontal coordinate difference between the corresponding points of the two point columns and multiply it by the number of point columns to obtain the absolute deviation of the lateral offset; The absolute deviation of the lateral offset is judged by a threshold value, and the points with an absolute deviation greater than the threshold value are determined as outliers, and the outliers are removed from the point list; Substitute the point sequence after removing the outliers into the ellipse equation to calculate the tunnel ellipticity.

4. The method for dynamic monitoring and early warning of tunnel ellipticity based on time-series point cloud according to claim 1 is characterized in that , the mathematical model establishment in the processing step specifically includes: Calculate the ellipticity vs. time to form a logarithmic function of the fitted data: y=aln(t)+b+ε Among them, y is the tunnel ellipticity series, that is, the ellipticity data of the tunnel at different periods, t is the monitoring time series, that is, the time data of the tunnel at different periods, a and b are the parameters to be fitted, and ε is the relative error of the tunnel ellipticity and time fitting.

5. The method for dynamic monitoring and early warning of tunnel ellipticity based on time-series point cloud according to claim 4 is characterized in that , the dynamic prediction in the processing step specifically includes: Based on the calculated slope b, the deformation rate is calculated, and based on this, the ellipticity value at the future time t is predicted: y t =aln(t)+b Among them, y t is the predicted ellipticity value at time t; Calculate the residual error of the deformation based on the calculated ovality value at time t; Draw the deformation development trend line to obtain the deformation development trend.

6. The method for dynamic monitoring and early warning of tunnel ellipticity based on time-series point cloud according to claim 1 is characterized in that ,The deformation warning in the output step specifically includes: The warning message is obtained by comparing the threshold value of the warning ellipticity obtained by historical fitting with the measured ellipticity; The residual error calculated by the historical fitting curve and the measured ellipticity is used to make a threshold judgment on the future ellipticity development trend; If the ellipticity change rate exceeds the warning threshold in the future, a warning message will be sent to the management personnel, otherwise no warning message will be sent, and the feedback of the warning results and the real-time update of the ellipticity will be completed.

7. The method for dynamic monitoring and early warning of tunnel ellipticity based on time-series point cloud according to claim 6 is characterized in that , the warning threshold is dynamically modified, specifically including: Through continuous training, the ellipticity curve based on different deformation rates is obtained; If the measured deformation and predicted residuals are both around 10% after monitoring time T, the ellipticity curve is compared with the measured value and the threshold is corrected; The warning threshold of ellipticity prediction uses the following formula to judge the threshold of ellipticity change rate: Among them, ε i is the ellipticity corresponding to the predicted time point i, ε i+1 is the ellipticity corresponding to the predicted time point i+1, and λ is the warning threshold.

8. The method for dynamic monitoring and early warning of tunnel ellipticity based on time-series point cloud according to claim 1 is characterized in that , also includes: Train the deep reinforcement learning model, and the ellipticity of the model at the initial moment is the measured ellipticity at the monitoring time t0; During model training, the model predicts from time t0 to time tend based on the reward model and calculates the reward: Where R is the total reward value, n is the total number of monitoring frequencies, and i is the index of the monitoring frequency, from 1 to n, w i is the weight of the i-th monitoring frequency, f is the reward function, a i is the monitoring frequency adopted by the model at the i-th time, e i is the relative error between the predicted ellipticity and the measured ellipticity for the i-th time, v i is the change rate of ellipticity predicted for the i-th time, c i is the monitoring frequency currently adopted by the model during the i-th prediction; The rewards obtained are used to train the deep reinforcement learning model to learn to predict the ellipticity value, monitor the frequency and obtain the model.

9. The method for dynamic monitoring and early warning of tunnel ellipticity based on time-series point cloud according to claim 1 is characterized in that , the obtaining step further includes: Get the ellipticity parameters of the tunnel vault; A dataset based on time series deformation is constructed for each ellipticity parameter; Input a sample of the dataset into the pre-trained deep reinforcement learning model to determine the monitoring frequency of the sample; A scan is performed at the monitoring frequency to obtain the next data set sample in the data set.

10. The tunnel ellipticity dynamic monitoring and early warning system based on time-series point cloud is characterized by ,include: A point cloud data acquisition module is used to scan the tunnel surface at different parts of the tunnel to obtain corresponding point cloud data, and then obtain several dome point cloud data groups; An ellipticity calculation module is used to calculate corresponding ellipticity indices for a plurality of dome cloud data groups based on a preset ellipticity calculation method, and form a corresponding ellipticity index sequence; The ellipticity monitoring neural network module is used to calculate the corresponding ellipticity index sequence according to the point cloud data of the vault to be measured, and to perform error correction on the sequence; The reinforcement learning model module is used to adjust the monitoring frequency of the point cloud data acquisition module according to the timestamp corresponding to the ellipticity index sequence; The time series analysis module is used to analyze the local deformation trend of the tunnel based on the monitoring frequency corresponding to the ellipticity index sequence and construct the ellipticity trend function; The early warning module is used to judge the change rate of the ellipticity index in any point cloud obtained in the time series analysis module, and issue an early warning through the early warning terminal if it exceeds the preset threshold; Wherein, the ellipticity index sequence and the ellipticity trend function are limited by the monitoring frequency corresponding to the ellipticity index sequence.

Images