Dynamic Optimization Layout Method for High Slope Monitoring Points

Abstract

The present invention belongs to the technical field of geotechnical engineering safety monitoring, and specifically discloses a dynamic optimization layout method for high slope monitoring points, including obtaining multi-source monitoring data of each monitoring point among multiple monitoring points in a high slope monitoring area; determining the neighborhood radius and feature density of each monitoring point according to the multi-source monitoring data, and partitioning the monitoring area according to the neighborhood radius and feature density; obtaining the partition characteristics of each partition, and inputting the partition characteristics and multi-source monitoring data into a deep learning model based on the attention mechanism to obtain the importance score and stability state of each partition; determining the optimization strategy for arranging monitoring points in each partition according to the importance score, stability state and feature density. The present invention can maximize the monitoring effect under the condition of a limited number of monitoring points, realize real-time monitoring and early warning of the deformation and stability of high slopes. This method has high scientificity, effectiveness and practicability, and can significantly improve the efficiency and accuracy of high slope monitoring.

Description

Technical Field

[0001] The present invention belongs to the technical field of geotechnical engineering safety monitoring, and discloses a dynamic optimization layout method for high slope monitoring points, in particular to a dynamic optimization layout method for high slope monitoring points integrating deep learning and dynamic optimization algorithms, which is applicable to the stability monitoring of high slopes in the fields of water conservancy, transportation, mines, etc. Background Technique

[0002] In various engineering constructions, as a common geological structure, the stability of high slopes is directly related to the safety and sustainable development of projects. With the continuous advancement of infrastructure construction, such as the vigorous development of large-scale projects like highways, railways, water conservancy and hydropower, the scale and complexity of high slopes are increasing day by day, and the requirement for their stability monitoring is becoming more and more urgent.

[0003] Traditional methods for arranging high slope monitoring points mainly rely on engineering experience and simple static arrangements, such as contour arrangement, equidistant arrangement, etc., and have many limitations. On the one hand, traditional methods often have difficulty in comprehensively considering the influence of various factors such as complex geological conditions, topography and landforms, and engineering construction on the slope stability, resulting in the lack of pertinence and rationality in the arrangement of monitoring points. For example, in some areas with complex geological conditions, simply arranging monitoring points according to experience may miss key deformation areas, resulting in monitoring blind spots and unable to comprehensively and accurately capture the deformation information of the slope. On the other hand, traditional arrangement methods lack the ability of dynamic adjustment. During the construction process and operation period of high slopes, when the state of the slope changes, it is difficult to optimize and adjust the monitoring points in a timely manner according to the actual situation, thus affecting the accuracy and effectiveness of monitoring data.

[0004] The method for arranging high slope monitoring points based on numerical simulation mainly relies on numerical simulation software to analyze the stress and strain distribution of the slope under different working conditions, so as to determine the arrangement position of the monitoring points. The principle of this method is based on the basic theory of geotechnical mechanics. The slope is regarded as a complex mechanical system, and a mathematical model is established to simulate the mechanical response of the slope under the action of factors such as self-weight, external loads, and groundwater. Although this method can relatively comprehensively consider various influencing factors of the slope and provide a relatively scientific basis for the arrangement of monitoring points. However, the accuracy of numerical simulation depends strongly on the accuracy of the model and the selection of parameters. If the physical and mechanical parameters of the rock and soil mass are inaccurately determined, or the boundary conditions of the model are set unreasonably, it will lead to a large deviation between the simulation results and the actual situation, thus affecting the accuracy of the monitoring point arrangement. The calculation amount in the numerical simulation process is large, and it requires a lot of time and computing resources. Especially for complex high slope models, the calculation time may be very long, which to a certain extent limits the application efficiency of this method.

[0005] The reasonable layout of high slope monitoring points directly affects the comprehensiveness of data collection and the timeliness of early warning. Existing methods for optimizing the layout of monitoring points mostly rely on clustering analysis (such as the Affinity Propagation algorithm) combined with optimization algorithms (such as the quantum particle swarm algorithm) to optimize the monitoring points. Although it can reduce the number of monitoring points to a certain extent, there are the following problems: (1) Static optimization relies on historical data and is difficult to adapt to the dynamic deformation characteristics of complex geological conditions, resulting in monitoring blind spots; (2) Clustering parameters need to be adjusted manually, with strong subjectivity, redundant or insufficient coverage of monitoring points, causing resource waste or early warning delays; (3) The spatio-temporal correlation of historical monitoring data is not fully exploited, and the spatio-temporal non-linear relationship between monitoring points is difficult to fully express through traditional mathematical models, resulting in the inability to predict potential risk areas. Therefore, there is an urgent need for a method that can dynamically optimize the layout of monitoring points in combination with the dynamic change characteristics of high slopes. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for dynamically optimizing the layout of high slope monitoring points to solve the technical problems of existing layout methods, such as monitoring blind spots, inability to predict potential risk areas, redundant or insufficient coverage of monitoring points.

[0007] The present invention provides a method for dynamically optimizing the layout of high slope monitoring points, including

[0008] Obtain multi-source monitoring data of each monitoring point among multiple monitoring points in the high slope monitoring area.

[0009] Determine the neighborhood radius and feature density of each monitoring point according to the multi-source monitoring data, and partition the monitoring area according to the neighborhood radius and feature density.

[0010] Obtain the partition features of each partition, and input the partition features and the multi-source monitoring data into a deep learning model based on the attention mechanism to obtain the importance score and stable state of each partition.

[0011] Determine the optimization strategy for arranging monitoring points in each partition according to the importance score, stable state and feature density.

[0012] Preferably, the multi-source monitoring data includes displacement spatio-temporal data and geological data. Then, determining the neighborhood radius and feature density of each monitoring point according to the multi-source monitoring data specifically includes:

[0013] Determine the deformation feature weight according to the standard deviation and mean of the displacement spatio-temporal data; determine the geological feature weight according to the standard deviation and mean of the geological data.

[0014] Determine the neighborhood radius of each monitoring point and the deformation feature density and geological feature density within the neighborhood radius.

[0015] Determine the feature density of each monitoring point according to the deformation feature weight, deformation feature density, geological feature weight, and geological feature density.

[0016] Preferably, determining the neighborhood radius of each monitoring point specifically includes:

[0017] Sort the distances between each monitoring point and other monitoring points from small to large, and record the first distance as the nearest neighbor distance of the corresponding monitoring point.

[0018] Determine the average value of the nearest neighbor distances corresponding to all monitoring points, and determine the standard deviation of the distances between monitoring points according to the average value.

[0019] Determine the neighborhood parameter according to the average value and the standard deviation of the distances between monitoring points k , and use the k th distance between each monitoring point and other monitoring points as the neighborhood radius of the monitoring point.

[0020] Preferably, partitioning the monitoring area according to the neighborhood radius and feature density specifically includes:

[0021] Determine whether each monitoring point is a core point according to the neighborhood radius and feature density, and determine the adaptive neighborhood radius of the core point.

[0022] All monitoring points within each core point and its adaptive neighborhood radius form the normal partition of the monitoring area; the monitoring points that do not belong to the normal partition are regarded as noise points, and the neighborhood radius of the noise points is used as the abnormal partition.

[0023] Preferably, determining whether each monitoring point is a core point according to the neighborhood radius and feature density specifically includes:

[0024] Determine the density threshold and the minimum number of points threshold.

[0025] Determine whether the number of monitoring points within the neighborhood radius of each monitoring point is greater than or equal to the minimum number of points threshold. If so, when the feature density of each monitoring point is greater than the density threshold, the corresponding monitoring point is considered a core point.

[0026] Preferably, determining the minimum number of points threshold specifically includes:

[0027] Determine the expected neighborhood radius according to the average value of the distances between all monitoring points.

[0028] Obtain the number of monitoring points within the expected neighborhood radius corresponding to each monitoring point, and record the median of the numbers corresponding to all monitoring points as the minimum number of points threshold.

[0029] Preferably, determining the density threshold specifically includes:

[0030] Determine the characteristic density of each monitoring point within the corresponding expected neighborhood radius, as well as the mean and standard deviation of all characteristic densities.

[0031] Determine the density threshold based on the mean and standard deviation of all characteristic densities.

[0032] Preferably, determining the adaptive neighborhood radius of the core point specifically includes:

[0033] Denote the quotient of the average of the characteristic densities of all monitoring points and the characteristic density of the core point as the first value.

[0034] Determine the adaptive neighborhood radius of the core point according to the product of the neighborhood radius and the first value.

[0035] Preferably, inputting the partition features and the multi-source monitoring data into the deep learning model based on the attention mechanism specifically includes:

[0036] Construct a fusion matrix according to the partition features and the multi-source monitoring data.

[0037] Input the fusion matrix into the convolutional neural network layer for spatial feature extraction to obtain a feature map.

[0038] Input the feature map into the CNN-LSTM model based on the attention mechanism.

[0039] Preferably, determine the optimization strategy for arranging monitoring points on each partition according to the importance score, stable state, and characteristic density, specifically including:

[0040] Determine the minimum score threshold and the maximum score threshold according to the mean and standard deviation of the importance scores of all partitions.

[0041] When the importance score of a partition is less than the minimum score threshold and it is unstable, the optimization strategy for this partition is to increase the number of monitoring points.

[0042] When the importance score of a partition is greater than the maximum score threshold and the characteristic density of this partition is greater than the partition density threshold, the optimization strategy for this partition is to reduce the number of monitoring points.

[0043] When the partition is an abnormal partition, the optimization strategy for this partition is to increase the number of monitoring points.

[0044] The dynamic optimization layout method for high slope monitoring points of the present invention has the following beneficial effects compared with the prior art:

[0045] By combining the deformation characteristics of high slopes and deep learning technology, the present invention dynamically optimizes the layout scheme of monitoring points, enabling the maximization of monitoring effects with a limited number of monitoring points and realizing real-time monitoring and early warning of the deformation and stability of high slopes. This method is highly scientific, effective, and practical, and can significantly improve the efficiency and accuracy of high slope monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is the overall framework of the dynamic optimization layout method for high slope monitoring points in the embodiments of the present invention.

[0047] Figure 2 It is the flow chart of dynamic regional division of the monitoring area in the embodiments of the present invention.

[0048] Figure 3 It is the flow chart of dynamic optimization of monitoring points in the embodiments of the present invention.

[0049] Figure 4 It is the monitoring area division result diagram of the specific embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] In the following description, specific details such as specific system structures and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present invention. However, those skilled in the art should clearly understand that the present invention can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from interfering with the description of the present invention.

[0051] The embodiments of the present invention provide a method for dynamically optimizing the layout of high slope monitoring points, as Figures 1 to 4 shown, including:

[0052] Step 1, real-time data acquisition, obtaining multi-source monitoring data of each monitoring point among multiple monitoring points in the high slope monitoring area.

[0053] The multi-source monitoring data in the embodiments of the present invention includes displacement spatio-temporal data and geological data, and may further include environmental quantity data.

[0054] Exemplarily, the embodiments of the present invention use sensors such as Global Navigation Satellite System (GNSS), displacement meters, and earth pressure gauges to collect displacement spatio-temporal data of the high slope monitoring area in real time; use sensors such as rain gauges and water level gauges to collect environmental quantity data in real time, and the two types of data together constitute the multi-source monitoring data of high slope displacement.

[0055] Let the monitoring data be , first perform standardization processing on it , is the mean of the th feature dimension, is the standard deviation of the th feature dimension. To eliminate the influence of dimension and order of magnitude and ensure the analysis of different features of multi-source monitoring data on the same scale.

[0056] According to the time step monitoring point feature dimension to construct a spatio-temporal feature matrix . Each row of the spatio-temporal feature matrix represents a time step, and each column represents a certain feature of a monitoring point. Specifically, for the th time step , the th monitoring point , the th feature dimension , the matrix element is the data after standardizing the feature of the monitoring point at this time step, that is .

[0057] Among them, the time step is determined according to the acquisition frequency and analysis requirements of the high slope monitoring data, and can be set to 1 hour, 1 day, etc., represents the total number of time step data collected; represents the total number of monitoring points arranged. Each monitoring point can collect various types of data, such as displacement, stress, rainfall, etc.; analyze the data types collected by each monitoring point to determine the total number of feature dimensions .

[0058] When determining the location of the monitoring point in the prior art, it often only focuses on the monitoring data itself and ignores the geological features, resulting in inaccurate results. To solve this problem in the embodiments of the present invention, geological data is also fused in the multi-source monitoring data, such as the internal friction angle of the rock and soil mass , cohesion , etc. The number of parameters is denoted as . These parameters provide more physical meanings and constraint conditions for the spatio-temporal feature matrix model. Taking them as additional feature dimensions and splicing them with the established spatio-temporal feature matrix to form an attribute-enhanced spatio-temporal feature matrix , where , the th geological parameter at the th monitoring area has a value of , then for all time steps define . For the enhanced spatio-temporal feature matrix Perform normalization to ensure that all features are within the same scale range. The normalization formula is: , where and are the minimum and maximum values of the corresponding dimensions in the spatio-temporal feature matrix, respectively.

[0059] Step 2: Determine the neighborhood radius and feature density of each monitoring point based on multi-source monitoring data, and partition the monitoring area according to the neighborhood radius and feature density. Specifically, it includes:

[0060] Step 2.1: Determine the neighborhood radius and feature density of each monitoring point based on multi-source monitoring data. Specifically, it includes:

[0061] Step 2.1.1: Determine the deformation feature weight according to the standard deviation and mean of the displacement spatio-temporal data; determine the geological feature weight according to the standard deviation and mean of the geological data.

[0062] In the embodiment of the present invention, in combination with the geological conditions of the slope (such as strata and geological structures, reflected in geological data) and deformation characteristics (displacement, stress, etc., reflected in displacement spatio-temporal data), different weights are assigned to different features when calculating the feature density, so that the subsequent clustering and partitioning results are more in line with the actual situation of the high slope.

[0063] Exemplarily, let the deformation feature weight be and the geological feature weight be . In the embodiment of the present invention, first calculate the standard deviation and mean of the displacement spatio-temporal data, as well as the standard deviation and mean of the geological data. Then, determine the fluctuation coefficient of the corresponding deformation feature and the fluctuation coefficient of the geological feature according to the above standard deviation and mean. Further, determine the corresponding deformation feature weight and geological feature weight according to the fluctuation coefficient of the deformation feature and the fluctuation coefficient of the geological feature, as shown in formula (1).

[0064] (1)

[0065] Step 2.1.2: Determine the neighborhood radius of each monitoring point, and the deformation feature density and geological feature density within the neighborhood radius.

[0066] Among them, determining the neighborhood radius of each monitoring point specifically includes:

[0067] Step S1: Sort the distances between each monitoring point and other monitoring points from small to large, and record the first distance as the nearest neighbor distance of the corresponding monitoring point.

[0068] In the embodiment of the present invention, the Euclidean distance is used to calculate the distance from a monitoring point to other monitoring points , generating a distance vector , is the data dimension, is the -th dimensional coordinate of the monitoring point. For each data point, the first distance value after sorting is taken as the nearest neighbor distance .

[0069] Step S2: Determine the average value of the nearest neighbor distances corresponding to all monitoring points , and determine the standard deviation of the distances between monitoring points according to the average value , as shown in formula (2).

[0070] (2)

[0071] In the formula, represents the average value of the distances from the data point to all other points.

[0072] Step S3: Determine the neighborhood parameter according to the average value and the standard deviation , and use the -th distance value between each monitoring point and other monitoring points as the neighborhood radius of this monitoring point, denoted as the initial neighborhood radius.

[0073] The above determination of the neighborhood parameter is specifically as shown in formula (3).

[0074] (3)

[0075] In the formula, is the adjustment factor , which is determined according to the distribution of the data and the requirements of the slope zoning . When the monitoring points are relatively evenly distributed, a smaller value can be taken; if the data distribution is complex and there are many outliers, a larger value can be taken.

[0076] After obtaining the neighborhood parameter using formula (3), the neighborhood radius of the monitoring pointcan be obtained .

[0077] Among them, the deformation feature density and geological feature density within the neighborhood radius of each monitoring point are determined, specifically as shown in formula (4).

[0078] (4)

[0079] Wherein, is the deformation feature density of the monitoring point , is the number of monitoring points within the neighborhood radius of the monitoring point whose deformation features are similar to those of , which is determined based on experience is the neighborhood volume is the geological feature density of the monitoring point , is the number of monitoring points within the neighborhood radius of the monitoring point whose geological features are similar to those of , which is determined based on experience

[0080] Step 2.1.3: Determine the feature density of each monitoring point according to the deformation feature weight, deformation feature density, geological feature weight, and geological feature density, as shown in formula (5).

[0081] (5)

[0082] Wherein, is the feature density of the monitoring point , denoted as the monitoring feature density is the deformation feature weight is the deformation feature density of the monitoring point , is the geological feature weight is the geological feature density of the monitoring point .

[0083] Step 2.2: Divide the monitoring area according to the neighborhood radius and feature density, specifically including:

[0084] Determine whether each monitoring point is a core point according to the neighborhood radius and feature density, and determine the adaptive neighborhood radius of the core point. All monitoring points within each core point and its adaptive neighborhood radius form the normal partition of the monitoring area; at the same time, the monitoring points that do not belong to the normal partition are regarded as noise points, and the neighborhood radius of the noise points is used as the abnormal partition

[0085] The above determination of whether each monitoring point is a core point according to the neighborhood radius and feature density specifically includes: determining the minimum number of points threshold and the density threshold ; determining whether the number of monitoring points within the neighborhood radius of each monitoring point is greater than or equal to the minimum number of points threshold (i.e., ), if so, determining whether the feature density of each monitoring point is greater than the density threshold (i.e., ), if so, then this monitoring point is a core point

[0086] Among them, determining the minimum number of points threshold specifically includes: according to the average value of the distances between all monitoring points determine the expected neighborhood radius , and then obtain the number of monitoring points within the expected neighborhood radius corresponding to each monitoring point , and record the median of the number of all monitoring points as the minimum number of points threshold .

[0087] Exemplarily, in the embodiment of the present invention, first calculate the average value of the distances of all monitoring points on the slope: , and then define a neighborhood radius , , the minimum number of points threshold take the median of the number of all monitoring points with as the neighborhood .

[0088] The embodiment of the present invention determines the density threshold , specifically: determine the characteristic density of each monitoring point within the corresponding expected neighborhood radius, and the mean and standard deviation of all characteristic densities; determine the density threshold according to the mean and standard deviation of all characteristic densities.

[0089] Exemplarily, for each monitoring point calculate the characteristic density with as the neighborhood: , where is the number of monitoring points within the range of of the monitoring point whose deformation characteristics are similar to , is the neighborhood volume with as the neighborhood radius. Further calculate the mean and standard deviation of all data points , and set the density adjustment coefficient according to the distribution of monitoring points, then . When the data distribution is relatively concentrated, can be appropriately taken as a small value; when the data distribution is relatively dispersed, is taken as a large value.

[0090] The embodiment of the present invention determines the adaptive neighborhood radius of the core point, specifically including: recording the quotient of the average value of the characteristic densities of all monitoring points and the characteristic density of the core point as the first value; determining the adaptive neighborhood radius of the core point according to the product of the neighborhood radius and the first value, as shown in formula (6).

[0091] (6)

[0092] In the formula, is the adaptive neighborhood radius, is the neighborhood radius of the monitoring point , is the average density of all monitoring points, is the characteristic density of the core point.

[0093] In the embodiment of the present invention, the density-based spatial clustering algorithm with noise application (Density-Based Spatial Clustering of Applications with Noise, DBSCAN) is used to partition the monitoring area and extended (EXDBSCAN). It divides the monitoring area with sufficient high density into clusters and determines the boundary of the partition according to the density of the clusters, as Figure 2 shown. The embodiment of the present invention introduces an adaptive neighborhood radius and improves the neighborhood radius , so as to better identify the clustering of different density regions.

[0094] In the embodiment of the present invention, the determined core point set is further traversed. For each core point , the set of all points within the area with the adaptive neighborhood radius as the radius is calculated. All points within the set (including ) are divided into a cluster . If there exists a point , satisfying and and , then the clusters and are merged.

[0095] After the clustering is generated, all data points are traversed. For a data point , if does not belong to the neighborhood of any core point, then is marked as a noise point. Taking the noise point as the core point, the area with the neighborhood of the noise point being is marked as an abnormal area.

[0096] The stop condition of the clustering in the embodiment of the present invention is as follows: The stop condition of the partition is double-constrained by no change in the core point and the maximum number of iterations. Let the core point set at the th iteration, represents the number of elements in the set . When is satisfied, and for any core point , be any point not in , it can be considered that the core point remains unchanged. At the same time, set the maximum number of iterations according to the calculation conditions. When the core point remains unchanged and , stop training and output the partition result.

[0097] Step 3: Obtain the partition features of each partition, and input the partition features and multi-source monitoring data into the deep learning model based on the attention mechanism to obtain the importance score and stable state of each partition. As Figure 3 shown, it specifically includes: constructing a fusion matrix according to the partition features and multi-source monitoring data; inputting the fusion matrix into the convolutional neural network layer (CNN layer) for spatial feature extraction to obtain a feature map ; inputting the feature map into the CNN-LSTM model based on the attention mechanism.

[0098] Based on the result of the high-slope deformation partition by the EXDBSCAN model in the embodiment of the present invention, a new monitoring point optimization method coupling EXDBSCAN-ATTCNN-LSTM is proposed, and the ATTCNN-LSTM monitoring point dynamic optimization model is further extended. ATTCNN is used to extract the spatial features in the spatio-temporal feature matrix and capture the spatial relationship between monitoring points; LSTM is used to process time series information and analyze the change trend of monitoring data over time, so as to determine the optimal location and quantity of the already deployed monitoring points. The specific steps are as follows:

[0099] (1) Partition feature encoding: One-hot encode the partition result output in Step 2. Assume that partition categories are output. For the th partition, the generation rule of its one-hot encoding is: , represents the th element of the vector , and takes values from 1 to .

[0100] (2) Model input: Expand the encoded partition features and the spatio-temporal feature matrix in the data feature dimension . First, expand the one-dimensional vector into a matrix of , and then splice and in the dimension to generate a fusion matrix , The number of characteristic dimensions is , and it is input into the improved ATTCNN-LSTM model feature extractor constructed.

[0101] (3)Feature extraction: Through the CNN layer, the spatio-temporal feature matrix is spatially extracted, and after convolutional layers, the feature map is obtained,[[]] which contains information such as the spatial distribution relationship between monitoring points and the feature differences at different positions.

[0102] (4)Introduce the attention mechanism (ATTENTION) to improve the CNN-LSTM model (ATTCNN-LSTM): By calculating the attention weights of each spatio-temporal monitoring position, determine the degree of attention of the model to the features at different monitoring positions. First, input the feature map into the fully connected layer for dimensionality reduction to obtain three key parameters in the attention mechanism: the query vector , the key vector , and the value vector , where , , are learnable weight matrices. Then calculate the attention score , is the transpose symbol, is and 's dimension. Finally, perform Softmax normalization on the attention score: , where is the total number of monitoring positions in the feature map.

[0103] (5)Feature weighting: Weight and fuse the attention weight matrix with the feature map to obtain the weighted feature map , where is the attention weight matrix.

[0104] (6)Dynamically adjust the LSTM structure: Input the weighted feature map into the LSTM layer for time series analysis. Introduce the Gradient-weighted Class Activation Mapping (Grad-CAM) method to output the importance score of the monitoring point features: , the higher the score, the greater the contribution of the monitoring point to the judgment of the slope state. In the optimization of monitoring points, this score can be used to determine whether to retain, adjust, or add monitoring points, so as to evaluate the importance of each monitoring point to the assessment of the slope safety state. According to the real-time monitoring data of the slope and the results of the zoning algorithm, judge whether the state of the slope has changed significantly. If the importance score of a certain monitoring point changes greatly, or the slope state changes significantly, dynamically adjust the number of hidden units of the LSTM layer according to the preset rules , and the adjustment range is , , which is the adjustment step set according to the actual situation.

[0105] (7) Result output: Output the prediction result of the slope state, mark the stable state as 1 and the unstable state as 0.

[0106] Step 4: Determine the optimization strategy for arranging monitoring points in each partition according to the importance score, stable state, and feature density, which specifically includes: determining the minimum score threshold and the maximum score threshold according to the mean and standard deviation of the importance scores of all partitions; when the importance score of a partition is less than the minimum score threshold and is unstable, the optimization strategy for this partition is to increase the number of monitoring points; when the importance score of a partition is greater than the maximum score threshold and the feature density of this partition is greater than the partition density threshold, the optimization strategy for this partition is to reduce the number of monitoring points; when the partition is an abnormal partition, the optimization strategy for this partition is to increase the number of monitoring points.

[0107] Exemplarily, sort the output importance scores and calculate the mean and standard deviation of the scores, and set the score threshold . Then for the feature density , calculate the average value and standard deviation of the density, and set the density threshold .

[0108] Dynamically optimize the monitoring points when the following three situations occur:

[0109] 1) When , and the partitions with unstable slope states should be densified with monitoring points.

[0110] 2) When but , retain the top 80% of the monitoring points with higher scores and reduce the number of monitoring points.

[0111] 3) When the partition is identified as an abnormal area, the monitoring points should be densified according to the geological conditions and the positions of the monitoring points should be optimized.

[0112] To verify the effects of the embodiments of the present invention, the following embodiments are verified.

[0113] (1) Project overview

[0114] A large slope located on the west bank of the Yellow River in China is adopted. The slope area is about 0.52 km 2 , the footwalk length is about 1086 m, and the topwalk length is about 312 m. Initially, 32 monitoring points are arranged, and the monitoring period is ten years from 2014 to 2024. Data is collected once a month, with a total of 12 time steps (T = 12). The monitoring data includes two characteristic dimensions of displacement (M1) and stress (M2), and at the same time, two geological characteristic dimensions of the in - situ internal friction angle (G1) and cohesion (G2) in the area are introduced ( ).

[0115] (2) Deformation zoning results based on EXDBSCAN

[0116] First, the displacement and stress monitoring data are standardized, and the normalized feature matrix with enhanced attributes is constructed by integrating geological information . The constructed EXDBSCAN algorithm is used to perform deformation zoning on the monitoring points. Traverse 32 monitoring points, and the training parameters are set as: neighborhood radius = 50 m, minimum sample number = 5, = 0.07 per m 2 , and the number of core points is 9. Starting from 8 core points, the expansion of density - connected points is carried out. After calculation, the slope is divided into 6 regions, namely P1, P2, P3, P4, P5, P6, P7, as Figure 4 shown. Among them, since the M15 measuring point does not belong to any cluster after calculation, the area where this point is located is marked as an abnormal area. The monitoring point numbers included in each partition are shown in Table 1.

[0117] Table 1 Calculation partition and optimization results

[0118]

[0119] (3) Dynamic optimal layout of high - slope monitoring points coupling EXDBSCAN - ATTCNN - LSTM

[0120] The matrix integrating partition features and spatio - temporal features It is divided into a training set, a validation set, and a test set according to the ratio of 8:1:1 and input into the improved ATTCNN-LSTM model. The ATTCNN part contains 3 convolutional layers, with the convolutional kernel sizes being 3×3, 5×5, and 7×7 respectively, and the stride being 1 for all; the LSTM part contains 2 LSTM layers, with the number of hidden units being 64 and 32 respectively. During the training process, the cross-entropy loss function and the Adam optimizer are used, and the learning rate is 0.001. After 50 epochs of training, the accuracy of the model on the test set reaches 87%. The output slope safety status is shown in Table 1, and the output monitoring point importance evaluation results are shown in Table 2.

[0121] Table 2 Monitoring Point Importance Evaluation Results

[0122]

[0123] According to the optimization results and the monitoring point importance evaluation results, the following dynamic optimization strategies are given:

[0124] 1) Add 1 monitoring point within a radius of 50 m around the M15 measuring point in the abnormal area M15 to collect data more densely and deeply understand the deformation trend and mechanical response of this area.

[0125] 2) Add 2 monitoring points near the M16 measuring point with a relatively low importance score (lower than the threshold of 0.5) in the actively deforming P6 area to make up for the lack of monitoring data in this area and improve the accuracy of deformation monitoring in this area.

[0126] 3) Remove the M measuring point among the adjacent M4 and M5 measuring points in the P1 area where the importance score is high (higher than the threshold of 0.8) but the monitoring point density exceeds the density threshold (0.07 points / m²).

[0127] After completing the dynamic optimization adjustment of the monitoring points, collect monitoring data again and construct a new matrix that combines partition features and spatio-temporal features. It is still divided into a training set, a validation set, and a test set according to the ratio of 8:1:1 and input into the EXDBSCAN-ATTCNN-LSTM model for training with the same parameters. The training results show that the accuracy of the optimized model is improved to 91%, indicating that the optimization strategy is effective and verifying the effectiveness of the method proposed in the present invention.

[0128] By combining the deformation characteristics of high slopes and deep learning techniques, the present invention dynamically optimizes the layout scheme of monitoring points, can maximize the monitoring effect with a limited number of monitoring points, and realizes the real-time monitoring and early warning of the deformation and stability of high slopes. This method has high scientificity, effectiveness, and practicality, and can significantly improve the efficiency and accuracy of high slope monitoring. The specific effects are as follows:

[0129] (1) Improve monitoring accuracy: By using the proposed EXDBSCAN algorithm to divide the deformation zones of monitoring points, the slope can be divided into regions according to stability characteristics, making subsequent analysis more targeted. ATTCNN is responsible for extracting the spatial features of monitoring data, accurately capturing the spatial relationships between monitoring points, and identifying the correlations of deformations at different positions on the slope; LSTM focuses on time series analysis and effectively mines the changing trends of monitoring data over time. The combination of the three processes monitoring data comprehensively, making the model's judgment of the slope state more accurate and significantly improving the prediction accuracy. The embodiments of the present invention combine deep learning with dynamic optimization algorithms, providing new ideas and methods for the layout of high slope monitoring points. This innovative method can give full play to the advantages of the two technologies, realize the intelligent and dynamic optimization layout of high slope monitoring points, improve the accuracy and efficiency of monitoring, and provide more reliable technical support for the safety monitoring of high slopes.

[0130] (2) Optimize the layout of monitoring points: In the prior art, the layout of monitoring points is mostly static and difficult to adjust in real time according to the slope state. The method of the present invention reasonably adjusts the number and position of monitoring points based on the importance scores of monitoring points output by the model. In areas with low importance scores and unstable slopes, the monitoring points are encrypted; in areas with high scores but dense monitoring points, the monitoring points are reduced or adjusted to ensure a more reasonable distribution of monitoring points, realize the optimized layout of monitoring points, improve the monitoring efficiency and accuracy, and avoid wasting resources at the same time, obtaining the most critical monitoring information with the fewest monitoring points.

[0131] (3) Enhance the adaptability of the model: The EXDBSCAN-ATTCNN-LSTM coupled model can comprehensively consider various factors such as the spatial distribution of monitoring data, time changes, and the geological characteristics of slopes. Whether facing complex terrain and landforms or different monitoring periods and data types, it can be dynamically adjusted according to the actual situation and has stronger adaptability.

[0132] (4) Early risk warning: The constructed model can analyze the slope state more accurately and predict the changing trends, can discover potential disaster risks such as landslides and collapses in advance, gain more time for taking protective measures, and effectively ensure the safety of people and property around the slope.

[0133] These are only several embodiments of the present invention and do not impose any form of limitation on the present invention. Although the present invention is disclosed as above with preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art, without departing from the scope of the technical solution of the present invention, makes some changes or modifications using the disclosed technical content, which are all equivalent to equivalent embodiments and fall within the scope of the technical solution.

Claims

1. A dynamic optimization layout method for high slope monitoring points, characterized in that Including: Obtain multi-source monitoring data of each monitoring point among multiple monitoring points in the high-slope monitoring area; Determine the neighborhood radius and characteristic density of each monitoring point according to the multi-source monitoring data, and partition the monitoring area according to the neighborhood radius and characteristic density; Obtain the partition characteristics of each partition, and input the partition characteristics and the multi-source monitoring data into a deep learning model based on the attention mechanism to obtain the importance score and stability state of each partition; Determine the optimization strategy for arranging monitoring points in each partition according to the importance score, stability state and characteristic density; If the multi-source monitoring data includes displacement spatio-temporal data and geological data, then determining the neighborhood radius and characteristic density of each monitoring point according to the multi-source monitoring data specifically includes: Determine the deformation feature weight according to the standard deviation and mean of the displacement spatio-temporal data; determine the geological feature weight according to the standard deviation and mean of the geological data; Determine the neighborhood radius of each monitoring point and the deformation feature density and geological feature density within the neighborhood radius; Determine the characteristic density of each monitoring point according to the deformation feature weight, deformation feature density, geological feature weight and geological feature density; Partition the monitoring area according to the neighborhood radius and characteristic density, specifically including: Determine whether each monitoring point is a core point according to the neighborhood radius and characteristic density, and determine the adaptive neighborhood radius of the core point; All monitoring points within each core point and its adaptive neighborhood radius form the normal partition of the monitoring area; the monitoring points that do not belong to the normal partition are regarded as noise points, and the neighborhood radius of the noise points is used as the abnormal partition; Inputting the partition characteristics and the multi-source monitoring data into a deep learning model based on the attention mechanism specifically includes: Construct a fusion matrix according to the partition characteristics and the multi-source monitoring data; Input the fusion matrix into a convolutional neural network layer for spatial feature extraction to obtain a feature map; Input the feature map into a CNN-LSTM model based on the attention mechanism; Determine the optimization strategy for arranging monitoring points on each partition according to the importance score, stability state and characteristic density, specifically including: Determine the minimum score threshold and the maximum score threshold according to the mean and standard deviation of the importance scores of all partitions; When the importance score of a partition is less than the minimum score threshold and is unstable, the optimization strategy for this partition is to increase the number of monitoring points; When the importance score of a partition is greater than the maximum score threshold and the characteristic density of this partition is greater than the partition density threshold, the optimization strategy for this partition is to reduce the number of monitoring points; When the partition is an abnormal partition, the optimization strategy for this partition is to increase the number of monitoring points.

2. The dynamic optimization layout method of high slope monitoring points according to claim 1, characterized in that Determine the neighborhood radius of each monitoring point, specifically including: Sort the distances between each monitoring point and other monitoring points from small to large, and record the first distance as the nearest neighbor distance of the corresponding monitoring point; Determine the average value of the nearest neighbor distances corresponding to all monitoring points, and determine the standard deviation of the distances between monitoring points according to the average value; Determine the neighborhood parameter according to the standard deviation of the average value and the distance between monitoring points k , and use the k th distance between each monitoring point and other monitoring points as the neighborhood radius of this monitoring point 3. The dynamic optimization layout method for high slope monitoring points according to claim 1 is characterized in that, Determine whether each monitoring point is a core point according to the neighborhood radius and characteristic density, specifically including: Determine the density threshold and the minimum number of points threshold; Determine whether the number of monitoring points within the neighborhood radius of each monitoring point is greater than or equal to the minimum point number threshold. If so, when the feature density of each monitoring point is greater than the density threshold, the corresponding monitoring point is considered a core point.

4. The dynamic optimization layout method for high slope monitoring points according to claim 3, characterized in that Determine the minimum point number threshold, specifically including: Determine the expected neighborhood radius according to the average value of the distances between all monitoring points; Obtain the number of monitoring points within the expected neighborhood radius corresponding to each monitoring point, and record the median of the numbers corresponding to all monitoring points as the minimum point number threshold.

5. The dynamic optimization layout method of high slope monitoring points according to claim 4, characterized in that, Determine the density threshold, specifically including: Determine the feature density of each monitoring point within the corresponding expected neighborhood radius, as well as the mean and standard deviation of all feature densities; Determine the density threshold according to the mean and standard deviation of all feature densities.

6. The dynamic optimization layout method for high slope monitoring points according to claim 1, characterized in that Determine the adaptive neighborhood radius of the core point, specifically including: Record the quotient of the average value of the feature densities of all monitoring points and the feature density of the core point as the first value; Determine the adaptive neighborhood radius of the core point according to the product of the neighborhood radius and the first value.

Images