Rock mass slope deformation monitoring method based on multi-modal data

Abstract

The application discloses a rock mass slope deformation monitoring method based on multi-modal data, relates to the technical field of geotechnical engineering monitoring, and obtains a plurality of rock block subareas by structurally decomposing a to-be-monitored slope region; multi-modal data collection is carried out based on each rock block subarea to obtain a multi-modal data set containing high-precision three-dimensional point cloud data, inclination angle data, wave speed data and prism original coordinate data; key deformation characteristic parameters and a multi-modal deformation comprehensive index of each rock block subarea are further obtained according to the multi-modal data set; a warning evaluation model is constructed, the key deformation characteristic parameters and the multi-modal deformation comprehensive index are taken as inputs, a warning level is output, and corresponding response measures are executed based on the warning level; the application realizes fine monitoring and intelligent warning of rock mass slope deformation through multi-modal data fusion, and improves the monitoring accuracy and response efficiency.

Description

Technical Field

[0001] This invention relates to the field of geotechnical engineering monitoring technology, specifically a method for monitoring rock slope deformation based on multimodal data. Background Technology

[0002] For monitoring rock slopes, traditional monitoring methods mainly rely on single sensor technologies, such as crack gauges or ground radar, which have obvious limitations. First, the monitoring range of a single sensor is limited, making it difficult to achieve full coverage of the slope. Especially under complex terrain conditions, monitoring blind spots are common, and the gradual evolution of disasters cannot be captured. Second, single-point monitoring data lacks spatial correlation and cannot effectively characterize the spatiotemporal distribution characteristics of the overall deformation field of the slope, thus limiting the accuracy of safety assessment.

[0003] Therefore, there is an urgent need for a technical solution that can effectively integrate multimodal data and improve monitoring accuracy and early warning efficiency. Here, we provide a rock slope deformation monitoring method based on multimodal data. Summary of the Invention

[0004] The purpose of this invention is to provide a method for monitoring rock slope deformation based on multimodal data.

[0005] The objective of this invention can be achieved through the following technical solution: a rock mass slope deformation monitoring method based on multimodal data, comprising the following steps:

[0006] Step S1: Perform structural decomposition of the slope area to be monitored to obtain several rock block sub-regions;

[0007] Step S2: Collect data based on each rock block sub-region to obtain a multimodal dataset;

[0008] Step S3: Based on the multimodal dataset, obtain the key deformation characteristic parameters and the multimodal deformation comprehensive index for each rock block sub-region;

[0009] Step S4: Construct an early warning assessment model, using key deformation characteristic parameters and multimodal deformation comprehensive index as inputs, outputting the early warning level, and implementing corresponding response measures based on the early warning level output by the early warning assessment model.

[0010] Furthermore, the process of structurally decomposing the slope area to be monitored into several rock block sub-regions includes:

[0011] Acquire overall image data and 3D point cloud data of the slope area to be monitored;

[0012] The overall image data is used to identify the boundary range, slope distribution, slope orientation, and slope inclination of the slope area to be monitored.

[0013] Based on the overall image data and 3D point cloud data, square grid units are divided at equal intervals along the slope direction and slope inclination within the boundary area of ​​the slope to be monitored.

[0014] For irregular areas at the boundary of the slope to be monitored, where it is impossible to divide into complete square grid units, incomplete grid units of varying sizes are generated based on the overall geometry to ensure complete coverage of the slope to be monitored.

[0015] In each grid cell, based on the distribution of structural planes, the grid cell is divided into several small blocks by the structural planes, which are called rock block sub-regions. If no structural plane passes through the grid cell, the entire grid cell is regarded as a rock block sub-region.

[0016] Furthermore, the process of defining the structure of the rock block sub-regions includes:

[0017] In each rock block sub-region, two adjacent slopes that are perpendicular to each other are defined. Either slope is called the first slope, and the other slope that is adjacent to and perpendicular to the first slope is called the second slope. The first slope and the second slope share a corner edge.

[0018] Furthermore, the process of acquiring multimodal datasets based on data collection from each rock sub-region includes:

[0019] A laser scanner is deployed on the first slope to acquire high-precision three-dimensional point cloud data, and a high-precision tilt sensor array is deployed on the second slope to record the tilt angle data output by the sensor array.

[0020] Geometric distance and wave velocity measurements were performed in the corner regions of each rock block sub-region, and the actual distance from the geometric center of the transmitter to the corner edge, the actual distance from the geometric center of the receiver to the corner edge, and the angle between the line connecting the transmitter and receiver and the slope normal were obtained.

[0021] Each rock sub-region was measured under three different combinations of horizontal and vertical distances to obtain the corresponding wave velocity data.

[0022] Prisms are deployed at the four corners of each rock block sub-region, and the three-dimensional coordinates of each prism are measured according to a preset cycle to obtain the original coordinate data of the prisms.

[0023] A multimodal dataset is obtained by integrating high-precision 3D point cloud data, tilt angle data, wave velocity data, and the original coordinate data of the prism.

[0024] Furthermore, the process of obtaining the key deformation feature parameters for each rock sub-region based on the multimodal dataset includes:

[0025] The key deformation characteristic parameters include the surface irregularity index, the rate of change of angle, the representative value of internal wave velocity, and the absolute deformation vector.

[0026] Based on the high-precision three-dimensional point cloud data of each rock block sub-region, the best fitting plane of the corresponding rock block sub-region is generated. After obtaining the best fitting plane corresponding to each rock block sub-region, the surface irregularity index is obtained.

[0027] Based on the tilt angle data, the change in tilt angle of the corresponding rock block sub-region at the measurement time relative to the initial reference time is obtained;

[0028] The rate of angle change and acceleration of the corresponding rock block sub-region are obtained based on the change in tilt angle.

[0029] For wave velocity data, the actual propagation distance of ultrasonic waves inside the rock mass is obtained by calculating the horizontal and vertical distances from the measurement center of the measuring device to the corner of the grid unit, the actual distance from the geometric center of the transmitting end to the corner edge, and the actual distance from the geometric center of the receiving end to the corner edge. Based on the actual propagation distance, the rock wave velocity data of this measuring point is obtained.

[0030] Each rock block sub-region was measured under three different combinations of horizontal and vertical distances. Wave velocity data was collected once for each combination, resulting in three independent rock mass wave velocity data, and thus obtaining the representative value of the internal wave velocity.

[0031] Based on the rock mass wave velocity data and the corresponding representative internal wave velocity values, the wave velocity measurement variation coefficient is obtained;

[0032] If the wave velocity measurement coefficient of variation is greater than the preset wave velocity measurement coefficient of variation threshold, it indicates that there are abnormal fluctuations in the measured wave velocity data, and the wave velocity data of the corresponding rock block sub-region needs to be remeasured.

[0033] Based on the original coordinate data of the prisms, the coordinates of the center points of the four prisms in the rock block sub-region are obtained, and then the absolute deformation vector at the measurement time relative to the initial time is obtained.

[0034] Furthermore, the process of obtaining the multimodal deformation composite index includes:

[0035] The deformation stage indicator factor is obtained based on the angle change rate. The fusion weight is dynamically adjusted based on the deformation stage indicator factor. The multimodal deformation comprehensive index is then obtained based on the fusion weight.

[0036] Furthermore, the process of constructing an early warning assessment model includes:

[0037] A warning assessment model based on a multilayer perceptron is constructed. The input layer of the warning assessment model contains five nodes, which correspond to five input features, namely, the surface irregularity index of the rock block sub-region at the measurement time, the angle change rate, the internal wave velocity representative value, the absolute deformation vector, and the multimodal deformation comprehensive index.

[0038] The output layer contains three nodes, which output the probabilities of blue, yellow and red alerts respectively, and the level with the highest probability is taken as the final alert result;

[0039] Several representative time periods were selected from the historical multimodal dataset. Within each time period, the above input features were extracted by rock block sub-regions. Geotechnical engineering experts then labeled the corresponding warning levels based on the actual slope deformation and failure that occurred in that time period.

[0040] Each set of input features and its corresponding warning level label constitutes a training sample. All samples are randomly divided into training set and test set according to a preset ratio. The training set is used to supervise the training of the warning assessment model. After each training cycle, the prediction accuracy of the warning assessment model is calculated on the test set.

[0041] When the accuracy of the test set reaches the preset threshold and does not improve for three consecutive cycles, training is stopped, and the current model parameters are saved as the final early warning evaluation model.

[0042] Furthermore, the process of using key deformation characteristic parameters and multimodal deformation comprehensive index as inputs, outputting early warning levels, and executing corresponding response measures based on the early warning level output by the early warning assessment model includes:

[0043] In actual monitoring, for each rock sub-region, the early warning assessment model automatically outputs the corresponding early warning level based on the key deformation characteristic parameters and multimodal deformation comprehensive index of each rock sub-region. At the same time, it determines the dominant deformation mode based on the tilt angle and executes corresponding response measures according to the output early warning level.

[0044] Compared with the prior art, the beneficial effects of the present invention are: by structurally decomposing the slope area to be monitored, the present invention obtains several rock block sub-regions, decomposes the complex slope into regular rock block sub-regions, makes the monitoring unit more refined and standardized, facilitates subsequent regional directional data collection, improves the spatial resolution and targeting of monitoring, and avoids the information ambiguity caused by overall monitoring.

[0045] Multimodal data collection was conducted on each rock block sub-region to obtain a multimodal dataset containing high-precision three-dimensional point cloud data, tilt angle data, wave velocity data, and original coordinate data of the prism. This dataset integrates multi-dimensional data such as surface morphology, tilt deformation, internal wave velocity, and spatial reference, making up for the incomplete information of a single data source and providing a rich and complementary information foundation for a comprehensive assessment of slope deformation.

[0046] Based on the multimodal dataset, key deformation characteristic parameters and multimodal deformation comprehensive index of each rock block sub-region are obtained. The comprehensive index can accurately reflect the overall deformation state of the rock block, improve the reliability and anti-interference ability of the monitoring results, and provide a scientific basis for early warning decision-making.

[0047] An early warning assessment model was constructed, which takes key deformation characteristic parameters and multimodal deformation comprehensive index as inputs, outputs early warning level, and executes corresponding response measures based on the early warning level. This realizes the automation and standardization of early warning, avoids the subjectivity and delay of human judgment, and the hierarchical response mechanism ensures the timeliness and appropriateness of emergency measures, effectively reducing the risk and loss of disasters.

[0048] This invention achieves refined monitoring and intelligent early warning of rock slope deformation through multimodal data fusion, thereby improving the accuracy and response efficiency of monitoring. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0050] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0051] like Figure 1 As shown, the rock mass slope deformation monitoring method based on multimodal data includes the following steps:

[0052] Step S1: Perform structural decomposition of the slope area to be monitored to obtain several rock block sub-regions;

[0053] Step S2: Collect data based on each rock block sub-region to obtain a multimodal dataset;

[0054] Step S3: Based on the multimodal dataset, obtain the key deformation characteristic parameters and the multimodal deformation comprehensive index for each rock block sub-region;

[0055] Step S4: Construct an early warning assessment model, using key deformation characteristic parameters and multimodal deformation comprehensive index as inputs, outputting the early warning level, and implementing corresponding response measures based on the early warning level output by the early warning assessment model.

[0056] It should be further explained that, in the specific implementation process, the process of structurally decomposing the slope area to be monitored into several rock block sub-regions includes:

[0057] Using drones equipped with high-resolution cameras and lidar, aerial scanning of the entire slope area to be monitored was conducted to acquire overall image data and 3D point cloud data of the slope area.

[0058] The overall image data is used to identify the boundary range, slope distribution, slope direction and slope inclination of the slope area to be monitored, and the overall geometric shape of the slope area to be monitored is obtained through the three-dimensional point cloud data.

[0059] Based on the overall image data and 3D point cloud data, square grid units are divided at equal intervals along the slope direction and slope inclination within the boundary area of ​​the slope to be monitored.

[0060] For irregular areas at the boundary of the slope to be monitored, where it is impossible to divide into complete square grid units, incomplete grid units of varying sizes are generated based on the overall geometry to ensure complete coverage of the slope to be monitored.

[0061] In each grid cell, based on the distribution of structural planes, the grid cell is divided into several small blocks by the structural planes, which are called rock block sub-regions. If no structural plane passes through the grid cell, the entire grid cell is regarded as a rock block sub-region.

[0062] The structural surfaces include major cracks extending more than three meters on the slope surface, rock strata boundaries, and fault planes.

[0063] It should be further explained that, in the specific implementation process, the process of collecting data based on each rock block sub-region to obtain a multimodal dataset includes:

[0064] In each rock block sub-region, two adjacent slopes that are perpendicular to each other are defined. Either slope is called the first slope, and the other slope that is adjacent to and perpendicular to the first slope is called the second slope. The first slope and the second slope share a corner edge line.

[0065] A laser scanner survey area was set up on the first slope. A stand-up 3D laser scanner was used to scan the surface contour of the rock block sub-area. The scanning range completely covered the boundary of the rock block sub-area and extended outward by a preset distance to obtain high-precision 3D point cloud data.

[0066] A high-precision tilt sensor array was deployed on the second slope. The sensor array base was firmly adhered to the rock surface with quick-drying epoxy resin and waterproofed. The data acquisition frequency was set to once every 30 minutes, recording the tilt angle data output by the sensor array. The tilt angle data included the X-axis tilt angle and the Y-axis tilt angle, denoted as [reference needed]. Simultaneously record the measurement time, denoted as . ;

[0067] Geometric distance and wave velocity measurements were performed in the corner region of each rock block sub-region. The corner region refers to the local spatial range near the corner edge shared by the first slope and the second slope, which belongs to both slopes.

[0068] During measurement, the transmitter of the measuring device is placed close to the first slope, and the measuring center of the device is located at different horizontal distances starting from the corner region of the grid cell. These horizontal distances are measured at three fixed points, denoted as . ;

[0069] Simultaneously, the receiving end is placed close to the second slope, ensuring its measurement center is located at the vertical distance calculated from the same corner region. Place, Take and The corresponding values ​​of the three measuring points;

[0070] The actual distance from the geometric center of the transmitter to the corner edge is read by a measuring device and recorded as follows: And the actual distance from the geometric center of the receiving end to the corner edge, denoted as The angle between the line connecting the transmitter and receiver and the normal to the slope is denoted as . ;

[0071] Each rock sub-region was measured under three different combinations of horizontal and vertical distances to obtain the corresponding wave velocity data.

[0072] Prisms are deployed at the four corners of each rock sub-region. The three-dimensional coordinates of each prism are measured at a preset interval to obtain the original coordinate data of the prisms, which is denoted as . , where j = 1, 2, 3, 4;

[0073] The prism refers to the prism reflector of the engineering surveying device, which serves as a fixed three-dimensional coordinate reference point to track the spatial position changes of the rock block sub-region.

[0074] A multimodal dataset is obtained by integrating high-precision 3D point cloud data, tilt angle data, wave velocity data, and the original coordinate data of the prism.

[0075] It should be further explained that, in the specific implementation process, the process of obtaining the key deformation characteristic parameters of each rock sub-region based on the multimodal dataset includes:

[0076] The key deformation characteristic parameters include the surface irregularity index, the rate of change of angle, the representative value of internal wave velocity, and the absolute deformation vector.

[0077] For each rock block sub-region's high-precision 3D point cloud data, three points are randomly selected from it. These three points uniquely determine a candidate plane. The vertical distance from all points in the high-precision 3D point cloud data to the candidate plane is calculated. The number of interior points is counted. This process is repeated multiple times. The candidate plane with the most interior points is selected as the best fitting plane for the corresponding rock block sub-region.

[0078] The interior point is a point whose vertical distance is less than a preset vertical distance threshold;

[0079] After obtaining the best-fit plane for each rock sub-region, the average vertical distance from all points in the high-precision 3D point cloud data to the best-fit plane is calculated to obtain the surface irregularity index, denoted as . ,Right now:

[0080] ;

[0081] in, Indicates the number of points. This represents the vertical distance from the i-th point to the best-fit plane;

[0082] Calculate the standard deviation of the surface irregularity index, which indicates the roughness of the best-fit plane, denoted as . ,Right now:

[0083] ;

[0084] For the dip angle data, the change in dip angle of the rock block sub-region at the measurement time relative to the initial reference time is denoted as... ,Right now:

[0085] ;

[0086] in, This indicates the X-axis tilt angle at the moment of measurement. This represents the X-axis tilt angle at the initial reference moment. This indicates the Y-axis tilt angle at the moment of measurement. This represents the Y-axis tilt angle at the initial reference time;

[0087] The rate of change of angle and acceleration of the corresponding rock block sub-region are obtained from the change of tilt angle, and are denoted as follows: , ,Right now:

[0088] , ;

[0089] in, The sampling interval;

[0090] For wave velocity data, the actual propagation distance of the ultrasonic wave within the rock mass is obtained by calculating the horizontal and vertical distances from the measurement center of the measuring device to the corner region of the grid cell, the actual distance from the geometric center of the transmitting end to the corner edge, and the actual distance from the geometric center of the receiving end to the corner edge. This distance is denoted as... ,Right now:

[0091] ;

[0092] in, This is the actual distance from the geometric center of the transmitter to the corner ridge line. This is the actual distance from the geometric center of the receiving end to the corner ridge line. express and The angle between the two direction lines;

[0093] The measuring device includes a transmitter and a receiver, with the measuring center located in the corner region of the grid cell;

[0094] Based on the actual propagation distance, the rock wave velocity data at this measuring point is obtained and denoted as . ,Right now:

[0095] ;

[0096] Each rock sub-region was measured under three different combinations of horizontal and vertical distances. Wave velocity data was collected once for each combination, resulting in three independent sets of rock mass wave velocity data, denoted as follows: , , ;

[0097] The arithmetic mean of the three sets of rock mass wave velocity data is taken as the representative value of the internal wave velocity of the corresponding rock block sub-region, denoted as . ,Right now:

[0098] ;

[0099] Then, based on the rock mass wave velocity data and the corresponding representative internal wave velocity values, the wave velocity measurement variation coefficient is obtained, denoted as . ,Right now:

[0100] ;

[0101] If the wave velocity measurement coefficient of variation is greater than the preset wave velocity measurement coefficient of variation threshold, it indicates that there are abnormal fluctuations in the measured wave velocity data, and the wave velocity data of the corresponding rock block sub-region needs to be remeasured.

[0102] For the original coordinate data of the prism, calculate the coordinates of the center point of the prism corresponding to each rock block sub-region, and let the first... The three-dimensional coordinates of the prism at the measurement time are: Then the coordinates of the center points of the four prisms in the sub-region of the rock block are:

[0103] , , ;

[0104] Based on the coordinates of the center points of the four prisms in the sub-region of the rock block, the absolute deformation vector at the measurement time relative to the initial time is obtained, denoted as . ,Right now:

[0105] .

[0106] It should be further explained that, in the specific implementation process, the process of obtaining the multimodal deformation composite index includes:

[0107] The deformation stage indicator factor is obtained based on the rate of angle change, denoted as... ,Right now:

[0108] ;

[0109] in, To preset the first speed threshold, To preset the second speed threshold, ;

[0110] For each rock block sub-region, the fusion weight is dynamically adjusted based on the deformation stage indicator factor;

[0111] The fusion weights include surface irregularity index weights, angular variation characteristic weights, and internal wave velocity representative value weights, denoted as follows: , , ,Right now:

[0112] ;

[0113] ;

[0114] ;

[0115] in, , , , , , The preset weighting coefficients satisfy... , ,and ;

[0116] The multimodal deformation composite index is obtained based on the fusion weights, denoted as... ,Right now:

[0117] ;

[0118] in, The initial surface irregularity index, The threshold for the change in tilt angle. This represents the initial internal wave velocity.

[0119] It should be further explained that, in the specific implementation process, the process of constructing the early warning assessment model includes:

[0120] A warning assessment model based on a multilayer perceptron is constructed. The input layer of the warning assessment model contains five nodes, which correspond to five input features, namely, the surface irregularity index of the rock block sub-region at the measurement time, the angle change rate, the internal wave velocity representative value, the absolute deformation vector, and the multimodal deformation comprehensive index.

[0121] The output layer contains three nodes, which output the probabilities of blue, yellow and red alerts respectively, and the level with the highest probability is taken as the final alert result;

[0122] Several representative time periods were selected from the historical multimodal dataset. Within each time period, the above input features were extracted by rock block sub-regions. Geotechnical engineering experts then labeled the corresponding warning levels based on the actual slope deformation and failure that occurred in that time period.

[0123] Each set of input features and its corresponding warning level label constitutes a training sample. All samples are randomly divided into training set and test set according to a preset ratio. The training set is used to supervise the training of the warning assessment model. After each training cycle, the prediction accuracy of the warning assessment model is calculated on the test set.

[0124] When the accuracy of the test set reaches the preset threshold and does not improve for three consecutive cycles, training is stopped, and the current model parameters are saved as the final early warning evaluation model.

[0125] It should be further explained that, in the specific implementation process, the process of using key deformation characteristic parameters and multimodal deformation comprehensive index as inputs, outputting early warning levels, and executing corresponding response measures based on the early warning levels output by the early warning assessment model includes:

[0126] In the actual monitoring process, for each rock block sub-region, the early warning assessment model automatically outputs the corresponding early warning level based on the key deformation characteristic parameters of each rock block sub-region and the multimodal deformation comprehensive index;

[0127] Simultaneously, the dominant deformation mode is determined based on the ratio of the changes in the X-axis tilt angle to the Y-axis tilt angle. If the deformation is tensile, it is determined to be tensile deformation; otherwise, it is shear deformation.

[0128] Implement corresponding response measures based on the output warning level:

[0129] During a blue alert, increased monitoring frequency and on-site inspections will be arranged.

[0130] When a yellow alert is issued, the management department should be notified, emergency supplies should be prepared, and manual displacement gauges should be deployed.

[0131] When a red alert is issued, immediately evacuate personnel, seal off the area below the slope, and activate the emergency response plan;

[0132] When issuing an early warning, the spatial location, warning level, and dominant deformation mode of the corresponding rock block sub-region are provided simultaneously.

[0133] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications or equivalent substitutions made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for monitoring rock mass slope deformation based on multimodal data, characterized in that, Includes the following steps: Step S1: Perform structural decomposition of the slope area to be monitored to obtain several rock block sub-regions; Step S2: Collect data based on each rock block sub-region to obtain a multimodal dataset; Step S3: Based on the multimodal dataset, obtain the key deformation characteristic parameters and the multimodal deformation comprehensive index for each rock block sub-region; Step S4: Construct an early warning assessment model, using key deformation characteristic parameters and multimodal deformation comprehensive index as inputs, outputting the early warning level, and implementing corresponding response measures based on the early warning level output by the early warning assessment model; The process of obtaining the key deformation feature parameters of each rock sub-region based on the multimodal dataset includes: The key deformation characteristic parameters include the surface irregularity index, the rate of change of angle, the representative value of internal wave velocity, and the absolute deformation vector. Based on the high-precision three-dimensional point cloud data of each rock block sub-region, the best fitting plane of the corresponding rock block sub-region is generated. After obtaining the best fitting plane corresponding to each rock block sub-region, the surface irregularity index is obtained. Based on the tilt angle data, the change in tilt angle of the corresponding rock block sub-region at the measurement time relative to the initial reference time is obtained; The rate of angle change and acceleration of the corresponding rock block sub-region are obtained based on the change in tilt angle. For wave velocity data, the actual propagation distance of ultrasonic waves inside the rock mass is obtained by calculating the horizontal and vertical distances from the measurement center of the measuring device to the corner of the grid unit, the actual distance from the geometric center of the transmitting end to the corner edge, and the actual distance from the geometric center of the receiving end to the corner edge. Based on the actual propagation distance, the rock wave velocity data of this measuring point is obtained. Each rock block sub-region was measured under three different combinations of horizontal and vertical distances. Wave velocity data was collected once for each combination, resulting in three independent rock mass wave velocity data, and thus obtaining the representative value of the internal wave velocity. Based on the rock mass wave velocity data and the corresponding representative internal wave velocity values, the wave velocity measurement variation coefficient is obtained; If the wave velocity measurement coefficient of variation is greater than the preset wave velocity measurement coefficient of variation threshold, it indicates that there are abnormal fluctuations in the measured wave velocity data, and the wave velocity data of the corresponding rock block sub-region needs to be remeasured. Based on the original coordinate data of the prisms, the coordinates of the center points of the four prisms in the rock block sub-region are obtained, and then the absolute deformation vector at the measurement time relative to the initial time is obtained. The process of obtaining the multimodal deformation composite index includes: The deformation stage indicator factor is obtained based on the angle change rate. The fusion weight is dynamically adjusted based on the deformation stage indicator factor. The multimodal deformation comprehensive index is then obtained based on the fusion weight.

2. The rock mass slope deformation monitoring method based on multimodal data according to claim 1, characterized in that, The process of structurally decomposing the monitored slope area to obtain several rock block sub-regions includes: Acquire overall image data and 3D point cloud data of the slope area to be monitored; The overall image data is used to identify the boundary range, slope distribution, slope orientation, and slope inclination of the slope area to be monitored. Based on the overall image data and 3D point cloud data, square grid units are divided at equal intervals along the slope direction and slope inclination within the boundary area of ​​the slope to be monitored. For irregular areas at the boundary of the slope to be monitored, where it is impossible to divide into complete square grid units, incomplete grid units of varying sizes are generated based on the overall geometry to ensure complete coverage of the slope to be monitored. In each grid cell, based on the distribution of structural planes, the grid cell is divided into several small blocks by the structural planes, which are called rock block sub-regions. If no structural plane passes through the grid cell, the entire grid cell is regarded as a rock block sub-region.

3. The rock mass slope deformation monitoring method based on multimodal data according to claim 2, characterized in that, The process of defining the structure of a rock block subregion includes: In each rock block sub-region, two adjacent slopes that are perpendicular to each other are defined. Either slope is called the first slope, and the other slope that is adjacent to and perpendicular to the first slope is called the second slope. The first slope and the second slope share a corner edge.

4. The rock mass slope deformation monitoring method based on multimodal data according to claim 3, characterized in that, The process of acquiring multimodal datasets based on data collection from various rock sub-regions includes: A laser scanner is deployed on the first slope to acquire high-precision three-dimensional point cloud data, and a high-precision tilt sensor array is deployed on the second slope to record the tilt angle data output by the sensor array. Geometric distance and wave velocity measurements were performed in the corner regions of each rock block sub-region, and the actual distance from the geometric center of the transmitter to the corner edge, the actual distance from the geometric center of the receiver to the corner edge, and the angle between the line connecting the transmitter and receiver and the slope normal were obtained. Each rock sub-region was measured under three different combinations of horizontal and vertical distances to obtain the corresponding wave velocity data. Prisms are deployed at the four corners of each rock block sub-region, and the three-dimensional coordinates of each prism are measured according to a preset cycle to obtain the original coordinate data of the prisms. A multimodal dataset is obtained by integrating high-precision 3D point cloud data, tilt angle data, wave velocity data, and the original coordinate data of the prism.

5. The rock mass slope deformation monitoring method based on multimodal data according to claim 1, characterized in that, The process of building an early warning assessment model includes: A warning assessment model based on a multilayer perceptron is constructed. The input layer of the warning assessment model contains five nodes, which correspond to five input features, namely, the surface irregularity index of the rock block sub-region at the measurement time, the angle change rate, the internal wave velocity representative value, the absolute deformation vector, and the multimodal deformation comprehensive index. The output layer contains three nodes, which output the probabilities of blue, yellow and red alerts respectively, and the level with the highest probability is taken as the final alert result; Several representative time periods were selected from the historical multimodal dataset. Within each time period, the above input features were extracted by rock block sub-regions. Geotechnical engineering experts then labeled the corresponding warning levels based on the actual slope deformation and failure that occurred in that time period. Each set of input features and its corresponding warning level label constitutes a training sample. All samples are randomly divided into training set and test set according to a preset ratio. The training set is used to supervise the training of the warning assessment model. After each training cycle, the prediction accuracy of the warning assessment model is calculated on the test set. When the accuracy of the test set reaches the preset threshold and does not improve for three consecutive cycles, training is stopped, and the current model parameters are saved as the final early warning evaluation model.

6. The rock mass slope deformation monitoring method based on multimodal data according to claim 5, characterized in that, The process of taking key deformation characteristic parameters and multimodal deformation comprehensive index as inputs, outputting early warning level, and executing corresponding response measures based on the early warning level output by the early warning assessment model includes: In actual monitoring, for each rock sub-region, the early warning assessment model automatically outputs the corresponding early warning level based on the key deformation characteristic parameters and multimodal deformation comprehensive index of each rock sub-region. At the same time, it determines the dominant deformation mode based on the tilt angle and executes corresponding response measures according to the output early warning level.

Images