Deep underground engineering large deformation prevention and control method and system

Abstract

The invention discloses a deep underground engineering large deformation prevention and control method and system, and relates to the field of underground engineering safety prevention and control, and the method comprises the following steps: arranging a displacement sensor in an underground engineering structure, and obtaining a displacement time sequence; carrying out normalization processing on the displacement time sequence; extracting a displacement change rate sequence of each monitoring point in each time window; taking the displacement change rate sequence of each monitoring point as a feature vector, and performing trend clustering on the monitoring points by adopting a clustering algorithm based on time sequence similarity to obtain a plurality of trend clustering class clusters; and when the number of the monitoring points contained in a certain trend clustering class cluster exceeds a preset proportion threshold value of the total point number and the average displacement change rate of the monitoring points continuously rises in the current time window, identifying an area corresponding to the class cluster as a trend abnormal accumulation area, and outputting early warning information of the area. According to the invention, the problems that the traditional threshold alarm cannot identify the tiny cooperative deformation precursor and the rate of missed alarm and false alarm is high are solved.

Description

Technical Field

[0001] The present invention relates to the field of safety prevention and control of underground engineering, and more specifically, to a method and system for preventing and controlling large deformations in deep underground engineering. Background Art

[0002] Deep underground engineering is widely used in the construction of major infrastructure such as tunnels, mines, underground energy storage, and power plant buildings. It usually has the characteristics of large burial depth, high in-situ stress of surrounding rock, and complex geological environment. In such projects, the stress state of the surrounding rock is prone to drastic changes. If not identified and intervened in time, it is extremely easy to induce large deformations or even sudden instability disasters, seriously threatening construction safety and structural stability.

[0003] At present, in the field of underground engineering, various displacement monitoring systems have been widely deployed to monitor and warn of the deformation of surrounding rock or structures. Traditional monitoring methods mainly rely on the over-limit alarm mechanism of single-point displacement values, that is, an alarm is triggered when the displacement of a certain monitoring point exceeds a preset threshold. However, this "absolute value threshold method" has obvious limitations: on the one hand, the early large deformation process often shows small, slow continuous deformations, and may have entered the dangerous evolution stage before reaching the alarm threshold; on the other hand, this method is difficult to reveal the co-deformation relationship between multiple points and cannot effectively capture the overall instability trend of local areas.

[0004] In addition, when analyzing the deformation trend in the prior art, it generally relies on manual experience judgment or simple linear fitting, and it is difficult to accurately identify potential group abnormal trends in a large amount of monitoring data at multiple points and multiple time periods. Especially in the micro-deformation stage, affected by factors such as monitoring errors and geological disturbances, traditional methods often cannot distinguish between "local random fluctuations" and "precursors of structural evolution", resulting in false alarms or missed alarms.

[0005] Therefore, there is an urgent need for a technical solution that can automatically identify the co-evolution of multi-point trends from time series data and accurately perceive the deformation aggregation trend of local areas, so as to improve the sensitivity, accuracy, and response timeliness of large deformation risk identification in deep underground engineering. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method and system for preventing and controlling large deformations in deep underground engineering to solve the problems mentioned in the background art.

[0007] To achieve the above object, the present invention adopts the following technical solutions: A method for preventing and controlling large deformations in deep underground engineering includes the following steps: S1: Arrange a plurality of displacement sensors in the underground engineering structure to obtain the displacement time series of multiple monitoring points within a continuous time period; S2: Normalize the displacement time series to eliminate the initial displacement deviation of each monitoring point and retain the trend change characteristics; S3: Perform a sliding time window operation on the normalized displacement time series to extract the displacement change rate series of each monitoring point within each time window; S4: Use the displacement change rate series of each monitoring point as the feature vector, and adopt a clustering algorithm based on time series similarity to perform trend clustering on the monitoring points to obtain multiple trend clustering clusters; S5: When the number of monitoring points included in a certain trend clustering cluster exceeds the preset proportion threshold of the total number of points, and its average displacement change rate continues to rise within the current time window, identify the area corresponding to this cluster as a trend abnormal aggregation area and output the warning information of this area.

[0008] Preferably, step S2 includes the following sub-steps: S21: Segment the original displacement time series of each monitoring point according to the monitoring period to obtain equal-length displacement sample series; S22: Calculate the mean and standard deviation of the displacement sample series for each monitoring point; S23: Adopt the Z-score normalization method to divide each value in the displacement sample series minus the mean of the displacement sample series by its standard deviation to obtain the normalized standardized displacement series.

[0009] Preferably, step S3 includes the following sub-steps: S31: Set a sliding time window with a fixed length, slide it on the normalized displacement time series according to the set step size, and extract the continuous data segments within each time window; S32: For each monitoring point, within each time window, calculate the displacement difference between the current moment and the previous moment divided by the time interval to obtain the displacement change rate series within this window; S33: Concatenate the displacement change rate series within all time windows to form a complete displacement change rate series.

[0010] Preferably, step S4 includes the following sub-steps: S41: Use the displacement change rate series of each monitoring point as the input feature vector to construct a time series similarity matrix between the monitoring points; S42: Adopt the dynamic time warping algorithm to calculate the similarity of the displacement change rate series between each monitoring point, obtain the temporal distance between each pair of monitoring points, and fill it into the time series similarity matrix; S43: Using the time sequence distance matrix as input, construct a similarity graph between monitoring points, and based on the similarity graph, use the density clustering algorithm to perform clustering analysis on the monitoring points to determine the set of monitoring points with similar time series change trends, and obtain multiple clustering clusters with similar trends.

[0011] Preferably, in the similarity graph, each node in the graph corresponds to a monitoring point, and the edge weight between nodes is the time sequence distance between each pair of monitoring points.

[0012] Preferably, the warning information output in step S5 includes the monitoring point numbers and spatial positions within the clustering cluster.

[0013] Preferably, the displacement sensor is a laser displacement sensor or an optical fiber displacement sensor.

[0014] The present invention also discloses a large deformation prevention and control system for deep underground engineering, including: A data acquisition module for arranging a plurality of displacement sensors in an underground engineering structure to obtain the displacement time series of a plurality of monitoring points within a continuous time period; A normalization processing module for normalizing the displacement time series to eliminate the initial displacement deviation of each monitoring point and retain the trend change characteristics; A change rate extraction module for performing a sliding time window operation on the normalized displacement time series to extract the displacement change rate series of each monitoring point within each time window; A trend clustering analysis module for using the displacement change rate series of each monitoring point as a feature vector and performing trend clustering on the monitoring points by using a clustering algorithm based on time series similarity to obtain multiple trend clustering clusters; A warning determination module for determining whether there is a trend abnormal aggregation area. When the number of monitoring points included in a certain trend clustering cluster exceeds a preset proportion threshold of the total number of points and its average displacement change rate continues to rise within the current time window, identify the area corresponding to this cluster as a trend abnormal aggregation area and output the warning information of this area.

[0015] Preferably, the normalization processing module includes: A sample sequence generation unit for splitting the original displacement time series of each monitoring point according to the monitoring period to obtain equal-length displacement sample sequences; A statistic calculation unit for calculating the mean value and standard deviation of the displacement sample sequence of each monitoring point; A Z-score normalization unit for using the Z-score normalization method to divide each value in the displacement sample sequence by its standard deviation after subtracting the mean value of the displacement sample sequence to obtain the normalized standardized displacement sequence.

[0016] Preferably, the trend clustering analysis module includes: A similarity calculation unit, configured to calculate the similarity of the displacement change rate sequences between monitoring points by using the dynamic time warping algorithm, and generate a time series distance matrix; A graph construction unit, configured to construct a similarity graph based on the time series distance matrix, where each node in the graph corresponds to a monitoring point, and the edge weight represents the time series distance between any two monitoring points; A clustering execution unit, configured to perform density clustering analysis based on the similarity graph, identify sets of monitoring points with similar trend change characteristics, and form multiple clustering clusters with similar trends.

[0017] The advantages of the present invention over the prior art are that the present invention proposes a large deformation prevention and control method and system for deep underground engineering, which can automatically identify the collaborative deformation trend of a local area based on the trend similarity between multiple monitoring points when the conventional displacement data has not reached the traditional warning threshold, so as to realize the early identification and accurate warning of large deformation risks, and significantly improve the foresight and reliability of underground engineering monitoring.

[0018] Specifically, the present invention eliminates the initial value difference between monitoring points through normalization processing, further extracts the displacement change rate sequence and constructs time series features, avoiding the dependence on "absolute displacement values", making the warning mechanism focus more on the change trend rather than static values, and having stronger adaptability and robustness.

[0019] The present invention adopts a time series similarity calculation method based on dynamic time warping, which can identify similar trend changes shown by multiple points at different amplitudes, and introduces a density clustering mechanism to perform trend clustering on monitoring points, realizing the automatic determination of "trend synchronization" abnormal groups. Compared with the traditional single-point alarm mechanism, the present invention can effectively capture the "group precursor signals" before structural instability and improve the sensitivity of abnormal identification.

[0020] In addition, the present invention realizes the accurate positioning of the trend abnormal aggregation area by introducing judgment conditions on the basis of clustering, such as the proportion of the number of clustering clusters and the continuous increase of the average displacement change rate, etc., effectively reducing the false alarm risk caused by noise or local disturbance.

[0021] In terms of the system solution, the present invention realizes a full-process closed-loop from data acquisition to abnormal identification and warning by constructing multiple functional modules including data acquisition, normalization processing, change rate extraction, similarity calculation, trend clustering, and warning determination, and has the technical advantages of high automation, strong engineering applicability, and being embeddable into the existing monitoring platform.

[0022] In summary, the present invention not only improves the sensitivity and accuracy of identifying deformation risks in deep underground engineering, but also provides technical support for realizing intelligent, real-time, and trend-driven large deformation prevention and control, and has broad engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is the overall flowchart of the present invention; Figure 2 is the data collection flowchart of the present invention; Figure 3 is the flowchart of generating the normalized sequence of the present invention; Figure 4 is the flowchart of time series clustering and anomaly identification of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] The following describes the specific embodiments of the present invention with reference to the drawings.

[0025] In deep underground engineering, large deformations of surrounding rocks or support structures often do not occur suddenly from a single point in isolation, but start from "small but synchronous" deformation trends in multiple regions within a similar time period. When multiple monitoring points simultaneously undergo displacement changes in the same direction (such as subsidence, convergence) at a similar rate within the same time window, it indicates that the rock mass or structural unit they are located in is being subjected to a common stress release or relaxation effect, which represents that the overall or local bearing capacity is weakening. This "trend synchronization" phenomenon cannot be judged by the absolute displacement value of a single point because the deformation of each point is very small in the early stage and is not yet large enough to exceed the traditional threshold. However, when these small changes occur collectively at multiple positions, it reveals the precursor of structural instability. Therefore, by performing time series clustering on the displacement change rates of different points and identifying those point groups with consistent trends, the present invention can capture the coordinated changes in the entire area before a large deformation actually occurs, thereby realizing early warning of deep large deformation risks.

[0026] More specifically, as shown in Figure 1, the present invention discloses a method for preventing and controlling large deformations in deep underground engineering. By arranging multiple displacement sensors in the underground engineering structure, obtaining the displacement time series, and performing a series of steps such as normalization processing, sliding time window operation, trend clustering, and identification of trend anomaly aggregation areas, the monitoring and early warning of large deformations are realized. The specific steps are as follows: S1: Arrange multiple displacement sensors in the underground engineering structure to obtain the displacement time series of multiple monitoring points within a continuous time period; S2: Perform normalization processing on the displacement time series to eliminate the initial displacement deviation of each monitoring point and retain the trend change characteristics; S3: Perform a sliding time window operation on the normalized displacement time series to extract the displacement change rate series of each monitoring point within each time window; S4: Using the displacement change rate series of each monitoring point as feature vectors, perform trend clustering on the monitoring points using a clustering algorithm based on time series similarity to obtain multiple trend clustering clusters; S5: When the number of monitoring points included in a certain trend clustering cluster exceeds a preset proportion threshold of the total number of points, and its average displacement change rate continues to rise within the current time window, identify the area corresponding to this cluster as a trend anomaly aggregation area and output the warning information for this area.

[0027] In a more specific embodiment, the position selection of the sensor needs to be determined according to the key parts of the engineering structure and the areas where deformation may occur. For example, in tunnel engineering, the sensors can be installed at positions such as the crown, side walls, and floor to comprehensively capture the displacement changes of different parts. The type of sensor can be selected as high-precision devices such as laser displacement sensors or fiber optic displacement sensors to ensure the accuracy and reliability of the data.

[0028] As shown in Figure 2, considering that displacement is three-dimensional (i.e., including components in the x, y, and z directions), each monitoring point needs to be able to record three-dimensional displacement data. One implementation method is to install three mutually perpendicular one-way displacement sensors at each monitoring point to measure the displacements in the x, y, and z directions respectively; another method is to use a multi-axis sensor that can directly output a three-dimensional displacement vector. Regardless of which method is adopted, each monitoring point will generate a three-dimensional displacement vector [dx, dy, dz] at each time point.

[0029] Taking tunnel engineering as an example, assume that a monitoring point P1 is arranged at the crown position, and three one-way displacement sensors are used to measure the radial (x), tangential (y), and axial (z) displacements respectively. Continuously monitor for one week (the sampling frequency is once per hour), and the displacement time series of P1 can be expressed as a series of three-dimensional vectors, for example: Time t1: [0.1, 0.05, 0.02] (unit: millimeter); Time t2: [0.15, 0.07, 0.03]; Time t3: [0.2, 0.09, 0.04]; …… These data reflect the displacement changes of P1 in three directions and provide a basis for subsequent analysis.

[0030] In subsequent steps, the displacements in the three directions can be processed independently, and in some embodiments, only the displacement in a single direction needs to be considered. For example, for crown settlement monitoring, what the construction personnel are more concerned about is the displacement in the longitudinal direction, so it may be sufficient to focus on a single direction.

[0031] Since the initial displacement values and change amplitudes at different monitoring points may vary greatly, it is difficult to compare their change trends directly using the original data. Therefore, it is necessary to normalize the displacement time series to eliminate the influence of dimensions and initial deviations and highlight the dynamic change characteristics. This method uses the Z-score normalization method to process the displacement sequences in each direction separately.

[0032] As shown in Figure 3, the specific operations are divided into the following sub-steps: First, the original displacement time series of each monitoring point is segmented into equally long sample sequences according to the monitoring period. Assuming the monitoring period is 24 hours, each sequence contains 24 data points. Taking P1 as an example, its displacement sequence for one day may be [0.1, 0.15, 0.2, ..., 0.5] (which can be considered in a certain direction, and the same for other directions).

[0033] Next, calculate the mean and standard deviation of each sequence. For the sequence of P1, the mean μ can be calculated by the average value of all data. For example, μ = (0.1 + 0.15 + 0.2 + ... + 0.5) / 24; the standard deviation σ is calculated through statistical formulas, reflecting the degree of dispersion of the data.

[0034] Then, apply the Z-score normalization formula to each data point in the sequence, that is, (d - μ) / σ, where d is the original displacement value, μ is the mean, and σ is the standard deviation. Taking the first data point 0.1 of P1 as an example, if μ = 0.3 and σ = 0.1, the normalized value is (0.1 - 0.3) / 0.1 = -2. After processing, the normalized sequence of P1 may be [-2, -1.5, -1, ..., 2], which eliminates the influence of absolute values and retains the change trend.

[0035] The advantage of this normalization method is that it standardizes the data into a distribution with a mean of 0 and a standard deviation of 1, making the data of different monitoring points comparable and laying a foundation for subsequent trend analysis. Note that the value range of the data after Z-score normalization is not limited to between -1 and 1, but can cover the entire real number range, depending on the distribution of the original data and the size of the standard deviation.

[0036] The above normalization process can be independently performed for the displacement data in each direction.

[0037] To further analyze the dynamic change trend of the displacement, it is necessary to perform a sliding time window operation on the normalized displacement time series to extract the displacement change rate within each time window. This process can capture the speed and direction of displacement changes.

[0038] Specifically, first, the size and step length of the sliding time window need to be set. The window size determines the time span of the analysis, and the step length determines the frequency of window movement. For example, the window size can be set to 24 hours and the step length to 1 hour, which means analyzing one day's data each time and sliding forward in one-hour increments. The selection of the window size and step length should be adjusted according to the engineering characteristics and monitoring frequency. For example, for projects with rapid changes, the window size can be reduced to improve sensitivity.

[0039] Within each time window, calculate the displacement change rate. Taking the normalized sequence of P1 as an example, assume a window contains data [-2, -1.5, -1, -0.5], corresponding to time points [t1, t2, t3, t4]. Then the change rate is (the last value - the initial value) / the time interval, that is, (-0.5 - (-2)) / (t4 - t1) = 1.5 / (t4 - t1). This value reflects the average speed of displacement change within this window.

[0040] Subsequently, concatenate the change rates of all windows to form a complete displacement change rate sequence. For example, the change rate sequence of P1 may be [0.05, 0.06, 0.07,...], indicating its change rate in different time periods. In this way, a sequence reflecting the dynamic trend of each monitoring point is generated. Similarly, the above independent processing can be performed on the displacement data in each direction, resulting in a three-dimensional sequence.

[0041] As shown in Figure 4, next, time series clustering analysis needs to be performed on the displacement change rate sequences of each monitoring point, including the following steps: Represent the displacement change rate sequence extracted from each monitoring point in consecutive sliding time windows as a set of time series vectors, denoted as ; where represents the change rate sequence of the i th monitoring point, for the displacement change rates corresponding to time series points 1 to T, where T is the time window length.

[0042] Using the change rate sequences of all monitoring points as input, initialize an N×N similarity matrix D, where N is the number of monitoring points, represents the similarity or distance between point i and point j in the time series dimension.

[0043] For any two monitoring points i 、 j, the dynamic time warping algorithm is used to calculate its displacement change rate sequence and the shortest time alignment distance between: ; DTW finds the optimal match between two time series of unequal length or with phase shift through a non-linear alignment path, preserving their change trends and shape features. This value is the element value of the similarity matrix D.

[0044] Finally, a complete time series distance matrix is obtained: ; Using the time series distance matrix D as the input, a similarity map between monitoring points is constructed, where: Each node in the graph corresponds to a monitoring point; The edge weight between nodes is determined by their DTW distance. The smaller the distance, the more similar the trends; The distance matrix can be trimmed for adjacent edges by setting a distance threshold to keep only the edges with similarity higher than the threshold.

[0045] Apply the density clustering algorithm (such as DBSCAN) on this graph for trend clustering: Using the DTW distance as the "spatial distance metric", set the number of points MinPts in the minimum neighborhood and the maximum neighborhood distance , and find the set of density-connected points as the clustering clusters, determining them as groups of points with similar time series change trends. MinPts can generally be set to 5% - 10% of the total number of monitoring points. That is, if there are N monitoring points in total, then MinPts ≈ 0.05N - 0.1N; should be determined according to the DTW distance distribution experience. First, sort all the DTW distance values of point pairs, and take the distance at the inflection point of its distance curve or the distance at the 10% - 20% percentile as the initial , and then fine-tune in combination with the trial run clustering effect. Usually select in the range of 0.5 - 1.5 after normalization; in this way, ensure that each cluster contains at least MinPts points with highly similar change trends, and the distance between points is less than to form an effective trend clustering.

[0046] The clustering output is multiple clusters with similar trends, and each cluster represents a "set of monitoring points with synchronized trends", which is subsequently used to identify areas with abnormal trend aggregations.

[0047] In some embodiments, if only the displacement in a single direction is of concern, only the single-direction case needs to be considered. If the overall displacement is to be considered, the three-dimensional case needs to be considered. For the three-dimensional case, when calculating the time series similarity (such as DTW), the DTW distances can be calculated separately for the three directions and then weighted and fused as the distance between the two monitoring points and filled into the matrix. D , or the Vector-based DTW algorithm can be used to directly perform an overall matching on the three-dimensional vector sequence, so as to retain the correlation of deformations in each direction. The subsequent clustering process remains the same.

[0048] The last step is to analyze the clustering results, identify the trend anomaly aggregation areas that may have the risk of large deformation, and output early warning information.

[0049] For each clustering cluster, calculate the proportion of the number of monitoring points it contains to the total number of points, and compare it with the preset proportion threshold. Suppose there are a total of 10 monitoring points and the threshold is set to 50%. If a certain cluster contains 7 points, the proportion is 70%, exceeding the threshold. The proportion threshold can be selected from 30% to 70% and determined according to specific circumstances. More specifically, the setting of the preset proportion threshold can comprehensively consider factors such as the number of monitoring points, the requirements of the project for the sensitivity of early warning, and the geological conditions: when the number of monitoring points is small, the threshold can be set relatively high to avoid false alarms caused by occasional anomalies of individual points; if the number of monitoring points is large and evenly distributed, the threshold can be appropriately reduced to promptly capture small-scale coordinated deformations; at the same time, it is necessary to initially set the threshold around 50% through historical data or trial operation observations in the initial stage of the project, and then gradually adjust it to the range of 30% to 70% according to the false alarm and missed alarm situations, so as to ensure that both group anomalies can be sensitively detected and the early warning is not overly triggered.

[0050] Next, calculate the average displacement change rate of all monitoring points in the cluster and determine whether it is continuously rising. For example, take the average displacement change rate of the nearest 3 time windows. If they are 0.05, 0.06, and 0.07 respectively, it indicates that the trend is continuously strengthening. Selecting 3 consecutive windows here is to avoid the influence of short-term fluctuations and ensure the stability of the trend.

[0051] If a certain cluster simultaneously meets two conditions - the proportion of monitoring points exceeds the threshold and the average displacement change rate is continuously increasing - then mark the corresponding spatial area as a trend anomaly aggregation area. For example, assume that the cluster contains P1, P2, P4, P6, P7, P8, P9, with a proportion of 70% and the change rate continuously rising, then this area is identified as an abnormal area.

[0052] Finally, output the early warning information, including the numbers of the monitoring points within the cluster (such as P1, P2, etc.) and the spatial positions (such as "tunnel vault"), so that engineering personnel can take measures in a timely manner.

[0053] As described above, it is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes, should be covered within the protection scope of the present invention.

Claims

1. A method for preventing and controlling large deformations in deep underground engineering, characterized in that, It includes the following steps: S1: Deploy a plurality of displacement sensors in the underground engineering structure to obtain the displacement time series of multiple monitoring points within a continuous time period; S2: Perform normalization processing on the displacement time series to eliminate the initial displacement deviation of each monitoring point and retain the trend change characteristics; S3: Perform a sliding time window operation on the normalized displacement time series to extract the displacement change rate series of each monitoring point within each time window; S4: Using the displacement change rate series of each monitoring point as the feature vector, perform trend clustering on the monitoring points using a clustering algorithm based on time series similarity to obtain multiple trend clustering clusters; S5: When the number of monitoring points included in a certain trend clustering cluster exceeds the preset proportion threshold of the total number of points, and its average displacement change rate continues to rise within the current time window, identify the area corresponding to this cluster as the trend abnormal aggregation area and output the warning information of this area.

2. The large deformation prevention and control method for deep underground engineering according to claim 1, wherein Step S2 includes the following sub-steps: S21: Segment the original displacement time series of each monitoring point according to the monitoring period to obtain an equal-length displacement sample series; S22: Calculate the mean and standard deviation of the displacement sample series of each monitoring point; S23: Adopt the Z-score normalization method to divide each value in the displacement sample series minus the mean of the displacement sample series by its standard deviation to obtain the normalized standardized displacement series.

3. The large deformation prevention and control method for deep underground engineering according to claim 1, characterized in that, Step S3 includes the following sub-steps: S31: Set a sliding time window with a fixed length, slide it on the normalized displacement time series according to the set step size, and extract the continuous data segments within each time window; S32: For each monitoring point, within each time window, calculate the displacement difference between the current moment and the previous moment divided by the time interval to obtain the displacement change rate series within this window; S33: Concatenate the displacement change rate series within all time windows to form a complete displacement change rate series.

4. The large deformation prevention and control method for deep underground engineering according to claim 1, characterized in that Step S4 includes the following sub-steps: S41: Use the displacement change rate series of each monitoring point as the input feature vector to construct a time series similarity matrix between the monitoring points; S42: Adopt the dynamic time warping algorithm to calculate the similarity of the displacement change rate series between each monitoring point, obtain the temporal distance between each pair of monitoring points, and fill it into the time series similarity matrix; S43: Using the temporal distance matrix as the input, construct a similarity graph between the monitoring points, and perform clustering analysis on the monitoring points using the density clustering algorithm based on the similarity graph to determine the set of monitoring points with similar time series change trends, and obtain multiple clustering clusters with similar trends.

5. The large deformation prevention and control method for deep underground engineering according to claim 4, wherein, In the similarity graph, each node in the graph corresponds to a monitoring point, and the edge weight between the nodes is the temporal distance between each pair of monitoring points.

6. The method for preventing and controlling large deformations in deep underground engineering according to claim 1, wherein, The warning information output in step S5 includes the monitoring point numbers and spatial positions within the clustering cluster.

7. The large deformation prevention and control method for deep underground engineering according to claim 1, characterized in that The displacement sensor is a laser displacement sensor or an optical fiber displacement sensor.

8. A large deformation prevention and control system for deep underground engineering, characterized in that, It includes: A data acquisition module for deploying a plurality of displacement sensors in the underground engineering structure to obtain the displacement time series of multiple monitoring points within a continuous time period; A normalization processing module, which is used to perform normalization processing on the displacement time series to eliminate the initial displacement deviation of each monitoring point and retain the trend change characteristics; A change rate extraction module, which is used to perform a sliding time window operation on the normalized displacement time series and extract the displacement change rate series of each monitoring point within each time window; A trend clustering analysis module, which is used to use the displacement change rate series of each monitoring point as a feature vector and adopt a clustering algorithm based on time series similarity to perform trend clustering on the monitoring points to obtain multiple trend clustering clusters; An early warning determination module, which is used to determine whether there is a trend abnormal aggregation area. When the number of monitoring points included in a certain trend clustering cluster exceeds the preset proportion threshold of the total number of points and its average displacement change rate continues to rise within the current time window, identify the area corresponding to this cluster as a trend abnormal aggregation area and output the early warning information of this area.

9. The large deformation prevention and control system for deep underground engineering according to claim 8, characterized in that The normalization processing module includes: A sample sequence generation unit, which is used to divide the original displacement time series of each monitoring point according to the monitoring period to obtain equal-length displacement sample sequences; A statistic calculation unit, which is used to calculate the mean and standard deviation of the displacement sample sequence of each monitoring point; A Z-score normalization unit, which is used to adopt the Z-score normalization method to divide each value in the displacement sample sequence by its standard deviation after subtracting the mean of the displacement sample sequence to obtain the normalized standardized displacement sequence.

10. The large deformation prevention and control system for deep underground engineering according to claim 8, wherein The trend clustering analysis module includes: A similarity calculation unit, which is used to calculate the similarity of the displacement change rate series between each monitoring point by using the dynamic time warping algorithm to generate a time series distance matrix; A graph construction unit, which is used to construct a similarity graph based on the time series distance matrix, where each node in the graph corresponds to a monitoring point and the edge weight represents the time series distance between any two monitoring points; A clustering execution unit, which is used to perform density clustering analysis based on the similarity graph, identify the set of monitoring points with similar trend change characteristics, and form multiple clustering clusters with similar trends.

Images