Dynamic compaction-reinforced soil slope collaborative construction method based on multi-parameter dynamic control

Abstract

The invention relates to the technical field of dynamic control, and further relates to a dynamic compaction-reinforced soil slope collaborative construction method based on multi-parameter dynamic control, and the method comprises the steps: 1, employing high-precision laser terrain scanning operation, obtaining a slope surface point cloud, and converting the slope surface point cloud into a three-dimensional grid with unified coordinates; registering the three-dimensional grid with an initial displacement baseline in the real-time displacement sequence; 2, variational mode decomposition is carried out on the purification sequence, and the purification sequence is divided into a plurality of intrinsic mode components; step 3, scanning the feature sequence by adopting a fixed-step sliding window, performing point-by-point mapping to obtain four parameters including a tamping energy level, a tamping point distance, a reinforcement layer distance and a reinforcement length, and assembling the four parameters into a terminal construction parameter list; and 4, dynamic compaction operation or reinforcement laying is circularly implemented according to the terminal construction parameter list until all designed compaction points and reinforcement layers are completed. According to the invention, monitoring precision and noise robustness can be ensured.

Description

Technical Field

[0001] This invention belongs to the field of dynamic control technology, specifically relating to a method for coordinated construction of dynamic compaction-reinforced soil slopes based on multi-parameter dynamic control. Background Technology

[0002] In high-speed railway, highway, and mine slope reinforcement projects, dynamic compaction and reinforced soil technology are widely used in parallel: dynamic compaction improves the compaction and bearing capacity of the fill through high-energy impact, while the reinforcement layer uses tensile materials to form a geogrid to resist shear slip and tensile cracking. Engineering practice typically adopts a phased design approach: first, the compaction energy, compaction point spacing, and number of compaction passes are selected based on site survey results; then, the reinforcement layer spacing and reinforcement length are determined based on the slope ratio, soil properties, and potential slip surface depth. To verify the scheme, several test compaction points are often set up on-site, and parameters are fine-tuned using empirical curves of penetration or settlement. Tracking slope deformation largely relies on total station cross-sectional measurements, manual leveling, or caliper readings. These methods have been used for decades and can meet conventional static stability requirements, but they have revealed significant shortcomings in scenarios with large fill heights, complex loads, or tight construction windows.

[0003] First, existing dynamic compaction processes are mostly implemented using a single energy level or a simplified three-stage sequence of "high-energy level—medium-energy level—low-energy level," with the spacing between compaction points and the number of passes predetermined, lacking a mechanism for real-time adjustment based on the dynamic response of the soil. When micro-cracks caused by excessive compaction occur in localized areas of the foundation, or when insufficient compaction settlement occurs due to changes in the moisture content of the fill soil, the construction team often only discovers the problem after the next round of test compaction, with the delay typically measured in hours, missing the optimal window for timely energy reduction or additional compaction. Conversely, conservatively increasing the energy level to avoid risks introduces construction costs and secondary vibration effects.

[0004] Secondly, the design of reinforced soil slopes often follows two-dimensional limit equilibrium or finite element static methods, with layer spacing and length set at the most unfavorable conditions. If, after dynamic compaction, the slope displacement exceeds expectations, the only option is usually to add surface anchor cables or shotcrete. Passive reinforcement is difficult to reach the potential failure surface and results in material waste. More importantly, existing procedures treat dynamic compaction and reinforcement as sequential operations, lacking parameter coupling: dynamic compaction parameters do not reflect the changes in stiffness and stress transfer path after reinforcement, and reinforcement dimensions do not consider the gradient distribution of soil density with depth and time during dynamic compaction, leading to a decrease in synergistic effects. Summary of the Invention

[0005] The main objective of this invention is to provide a collaborative construction method for dynamic compaction and reinforced soil slopes based on multi-parameter dynamic control. This method obtains a unified coordinate data source by registering a 3D point cloud with a displacement baseline. After purifying real-time displacement using median filtering and linear interpolation, modal decomposition is performed to extract multi-scale features. A fixed window-adaptive baseline strategy is then used to divide the slope response into stable, progressive, and abrupt zones, constructing a gradient sequence. Finally, based on multi-level displacement thresholds, the gradient levels are mapped in real-time to four control quantities: compaction energy level, compaction point spacing, reinforcement layer spacing, and reinforcement length. A closed-loop update is performed after each round of construction. This method can achieve real-time linkage adjustment of compaction energy and reinforcement parameters while ensuring monitoring accuracy and noise robustness, significantly reducing material waste and safety risks, and improving the efficiency, economy, and intelligence level of slope reinforcement construction.

[0006] To solve the above problems, the technical solution of the present invention is implemented as follows: A method for coordinated construction of dynamic compaction and reinforced soil slopes based on multi-parameter dynamic control, the method comprising: Step 1: Use high-precision laser terrain scanning to acquire point clouds on the slope surface and convert them into a 3D mesh with unified coordinates; set a series of reference displacement piles at the toe of the slope, start real-time displacement recording, and obtain a real-time displacement sequence including the initial displacement baseline; register the 3D mesh with the initial displacement baseline in the real-time displacement sequence. Step 2: Perform median filtering on the real-time displacement sequence to remove discrete noise and obtain a cleaned sequence; perform variational mode decomposition on the cleaned sequence to split it into several intrinsic mode components; construct feature descriptions for each intrinsic mode component, and combine the feature descriptions of all intrinsic mode components into a feature sequence. Step 3: Use a fixed-step sliding window to scan the feature sequence and mark the feature descriptions in the feature sequence into three categories: quasi-stable region, asymptotic region, and abrupt region. Calculate the displacement increment of the intrinsic modal components corresponding to the feature descriptions of each category compared to the intrinsic modal components of the previous time series to form a gradient sequence, and set multi-level displacement thresholds based on the gradient sequence. Divide the three categories into multiple different gradient levels according to the multi-level displacement thresholds. Through a preset gradient and control quantity comparison table, map point by point to obtain four parameters: compaction energy level, compaction point spacing, reinforcement layer spacing, and reinforcement length, and compile them into a final construction parameter list. Step 4: Perform dynamic compaction or reinforcement laying in a cycle according to the terminal construction parameter list, and collect displacement information and update the terminal construction parameter list in real time after each round of construction until all designed compaction points and reinforcement layers are completed.

[0007] Furthermore, in step 1, the high-precision laser terrain scanning adopts a multi-view encirclement measurement method with no less than three stations, the single-station angular resolution is no greater than 0.01°, and the point cloud density after merging is no less than 2500 points / ㎡; the number of reference displacement piles linearly arranged along the slope toe is no less than 5, and the distance between adjacent reference displacement piles is no less than 2% of the slope length.

[0008] Furthermore, in step 2, the real-time displacement sequence is written into the initial sequence table according to the sampling order; the initial sequence table is slid point by point using a median filter window of odd length, and the width of the median filter window is set to 5 data points; if the difference between the displacement value of a certain point and the median value of the corresponding window is greater than 3 times the median absolute deviation of the same window, then the point is identified as a discrete noise point and temporarily stored in the noise queue; after the noise point is removed, the gap is filled by linear interpolation of the legal data on both sides to obtain a continuous clean sequence.

[0009] Furthermore, in step 2, the number of target intrinsic modal components is set to 8, using the purified sequence as input; iterative decomposition is performed, stopping when the reconstruction error decreases by less than 0.001 for two consecutive rounds or when the number of iterations reaches 500; if the residual energy still accounts for more than 1% of the original energy, the number of target intrinsic modal components is increased by a step size of 2 and decomposed again until the residual energy ratio is no more than 1%; for each intrinsic modal component, at the position corresponding to the original timestamp, four indicators are extracted in sequence: instantaneous displacement amplitude, instantaneous displacement rate of change, center value of the dominant frequency interval, and cumulative energy; the five fields of timestamp, instantaneous displacement amplitude, instantaneous displacement rate of change, center value of the dominant frequency interval, and cumulative energy are written into the feature description in a fixed order; all feature descriptions are written into the feature sequence in ascending order of timestamp.

[0010] Furthermore, in step 3, a window width containing 9 feature descriptions is selected; the fixed step size of the window is set to 3 feature descriptions, and it slides in ascending order of time from the first position in the feature sequence; each slide extracts all feature descriptions within the window to form a temporary window dataset; for each feature description in the temporary window dataset, four indicators are read: instantaneous displacement amplitude, instantaneous displacement rate of change, center value of the dominant frequency interval, and cumulative energy; within the same window, the median, range, interquartile range, and mean of the above four indicators are calculated respectively; the 16 statistics corresponding to the window are written into the window indicator vector in a fixed order; the initial baseline is calculated using the first 10 window indicator vectors obtained from the first traversal, and the arithmetic mean of each indicator is taken; subsequently, whenever 5 new windows are accumulated, the baseline is updated with a weighted average method with a weight of 0.2, so that the old baseline accounts for 0.8 and the new window mean accounts for 0.2.

[0011] Furthermore, in step 3, if all 16 statistics in the current window indicator vector are less than 1.2 times the corresponding statistics of the baseline, then all the corresponding feature descriptions in the current window are marked as stable regions; if at least one statistic falls between 1.2 and 2.0 times the baseline and no statistic exceeds 2.0 times, then the corresponding feature description in the current window is marked as an asymptotic region; if any statistic exceeds 2.0 times the baseline, then the corresponding feature description in the current window is marked as a leap region; if a window with a feature description of a leap region is surrounded by two windows with feature descriptions of stable regions, then the corresponding feature description in that window is marked as a stable region; if two windows with feature descriptions of a leap region are surrounded by two windows with feature descriptions of asymptotic regions, then the marking results of the feature descriptions within that window are retained, but one window is extended on each side, and the feature descriptions corresponding to these two windows are also marked as leap regions.

[0012] Furthermore, in step 3, the feature description corresponding to each timestamp is read sequentially; the instantaneous amplitude difference of displacement is calculated for the intrinsic modal components of adjacent timestamps within the same segment; if the difference is negative, the absolute value is taken, and then written into the gradient temporary storage table in chronological order; when the segment category changes, the current gradient temporary storage table is immediately closed and completely written into the gradient sequence, and then a new gradient temporary storage table is opened for the next segment; each item in the gradient sequence contains the segment category, start timestamp, end timestamp, and cumulative displacement increment.

[0013] Further, in step 3, the minimum, first quartile, median, third quartile, and maximum values ​​of the cumulative displacement increment in the entire gradient sequence are statistically analyzed, totaling five statistics. The minimum and maximum values ​​are used as the lowest and highest levels of the multi-level displacement threshold, respectively. The remaining three statistics are inserted into the intermediate levels in ascending order, forming a multi-level displacement threshold group of no less than five levels. If the difference between two adjacent threshold levels is less than 2 mm, the lower threshold level is retained and adjusted by 0.5 mm increments towards the higher threshold level until the difference between adjacent levels is not less than 2 mm. The cumulative displacement increment of each item in the gradient sequence is compared with the multi-level displacement threshold group one by one. If the cumulative displacement increment is less than the lowest threshold level, it is recorded as the first level gradient. If the cumulative displacement increment is between two adjacent threshold levels, it corresponds to a higher level gradient. If the cumulative displacement increment is equal to or greater than the highest level threshold, it is recorded as the highest level gradient. The results are written into the gradient level table, which maintains a one-to-one correspondence with the timestamp.

[0014] Furthermore, the preset gradient and control quantity comparison table contains no less than 5 gradient levels, each gradient level corresponding to a unique combination of four parameters: compaction energy level, compaction point spacing, reinforcement layer spacing, and reinforcement length. Among them, the compaction energy level is in integer tons-meters, not less than 1 ton-meter and increasing by no less than 2 tons-meters at each level; the compaction point spacing is in meters, ranging from 0.8 meters to 2.0 meters, decreasing by no less than 0.2 meters at each level; the reinforcement layer spacing is in meters, ranging from 0.3 meters to 1.0 meters, decreasing by no less than 0.1 meters at each level; and the reinforcement length is in meters, ranging from 2.0 meters to 6.0 meters, increasing by no less than 0.5 meters at each level.

[0015] This invention discloses a collaborative construction method for dynamic compaction and reinforced soil slopes based on multi-parameter dynamic control, which has the following advantages: First, the three-dimensional point cloud and displacement baseline are registered at one time, allowing displacement and geometric information to be integrated within the same coordinate frame, avoiding common fault-based measurement errors and improving the accuracy of slope deformation identification from the source. Second, the purified sequence can separate high-frequency compaction response and low-frequency settlement trend through variational mode decomposition, and retain key energy components without amplifying noise, providing a multi-scale, holographic data foundation for subsequent sliding window statistics. Third, gradient partitioning based on adaptive baseline and multi-level displacement thresholds replaces the static safety factor, enabling the four parameters—compactment energy level, compaction point spacing, reinforcement layer spacing, and reinforcement length—to be updated in real time with slope behavior, effectively avoiding material waste caused by "over-compaction" and residual settlement caused by "under-compaction." Fourth, window overlap... The dual mechanism of section smoothing correction significantly reduces the risk of misjudgment caused by occasional noise, ensuring both recognition sensitivity and robustness. Fifth, the closed-loop update frequency is linked to the deformation amplitude, which can automatically lengthen the sampling period during long-term stable periods to reduce construction interference, and actively shorten the interval for rapid response during sudden changes, forming a self-regulating construction rhythm. Sixth, multi-view laser scanning and displacement multi-pile deployment provide redundant data channels. Even if a single-point sensor fails, the information flow can be restored in time through bundle reconstruction and inter-pile differential analysis, improving the system's fault tolerance and on-site safety factor. Finally, the continuous accumulation of displacement, energy, and parameter data throughout the construction process provides a reusable large-sample training set for subsequent similar projects, laying the foundation for the transformation of dynamic compaction and reinforcement collaborative design from experience-based to data-driven, and opening the channel for real-time simulation and long-term health monitoring of digital twin slopes. In summary, this method significantly improves the safety, economy, and intelligence of slope reinforcement, and can be widely applied to the rapid remediation and long-term operation and maintenance of high-speed railways, highways, and large foundation pit projects. Attached Figure Description

[0016] Figure 1 A schematic diagram of the method flow for a collaborative construction method of dynamic compaction and reinforced soil slope based on multi-parameter dynamic control, provided in an embodiment of the present invention; Figure 2This is a schematic diagram of a sliding window scanning method provided in an embodiment of the present invention; Figure 3 The instantaneous amplitude change experimental curve is provided for an embodiment of the present invention. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0018] refer to Figure 1 A method for coordinated construction of dynamic compaction and reinforced soil slopes based on multi-parameter dynamic control, the method comprising: Step 1: Use high-precision laser terrain scanning to acquire point clouds on the slope surface and convert them into a 3D mesh with unified coordinates; set a series of reference displacement piles at the toe of the slope, start real-time displacement recording, and obtain a real-time displacement sequence including the initial displacement baseline; register the 3D mesh with the initial displacement baseline in the real-time displacement sequence. Step 1 first involves high-precision laser topographic scanning to achieve full coverage acquisition of the slope surface point cloud. Within the same work cycle, the acquired raw point cloud is converted in real-time into a unified coordinate 3D mesh to ensure continuous data processing and construction control on the same spatial reference. A multi-view encirclement measurement method is employed during scanning. Laser stations are planned and deployed to ensure blind-spot-free scanning of all slope levels, slope transition lines, and slope top edges in a cross-covering manner. After each station completes its scan, point cloud stitching of common feature surfaces between stations is immediately performed. A global matching algorithm is then used to register the merged point cloud, ultimately outputting the preliminary fusion result of the slope surface point cloud. To improve the coupling accuracy between the scan data and subsequent displacement monitoring data, a series of benchmark displacement piles are set up at the slope toe. These piles are laid out linearly along the slope toe, and real-time displacement recording is initiated using precision displacement sensors to generate a real-time displacement sequence including the initial displacement baseline. The initial displacement baseline strictly corresponds to the start time of the laser scan in the time domain and is written into the respective data frame headers via a synchronized time signal to ensure timestamp consistency between different data sources.

[0019] After the point cloud is converted into a 3D mesh with unified coordinates, the method automatically calls the registration step to register the 3D mesh with the initial displacement baseline in the real-time displacement sequence. This registration process uses the physical space coordinates of the reference displacement pile as the anchor point and maps the scanning coordinate system to the construction control coordinate system through rigid body transformation, ensuring that subsequent displacement measurements and point cloud geometric information can be directly superimposed and analyzed within the same coordinate frame. In the specific implementation, the method first identifies the point cloud set corresponding to the reference displacement pile in the 3D mesh and calculates the centroid of the set as an approximation of the pile center. Then, it reads the coordinate information of the initial displacement baseline in the real-time displacement sequence, uses the two as a registration pair, and uses the least squares optimization algorithm to obtain the rotation matrix and translation vector, which are then applied to the entire point cloud set to achieve overall coordinate consistency. After registration is completed, the method immediately outputs the 3D mesh with unified coordinates and the synchronous initial displacement baseline, laying the data foundation for the subsequent median filtering, variational mode decomposition, and feature description construction in step 2. In addition, to ensure the reliability of the implementation on site, the method corrects the internal parameters of the instrument through a self-test function before scanning, and adjusts the point cloud for minor distortion by combining the environmental temperature and humidity compensation model after scanning; the real-time displacement recording step adopts a high-frequency sampling and synchronous multi-path redundant storage strategy, which can still avoid data loss when network jitter or power fluctuation occurs.

[0020] Step 2: Perform median filtering on the real-time displacement sequence to remove discrete noise and obtain a cleaned sequence; perform variational mode decomposition on the cleaned sequence to split it into several intrinsic mode components; construct feature descriptions for each intrinsic mode component, and combine the feature descriptions of all intrinsic mode components into a feature sequence. Step 2 is crucial for converting the real-time displacement sequence into a highly reliable dynamic information stream that can be directly accessed by subsequent intelligent identification and parameter mapping. First, it uses an initial sequence list to organize the real-time displacement sequence temporally. A unified data frame structure is used to timestamp each displacement record, ensuring strict synchronization with the unified coordinate 3D grid output from Step 1. Then, the method calls a median filtering window that slides point-by-point across the initial sequence list. By sorting the displacement values ​​within the window and comparing the median value with the current position, discrete noise points are automatically identified using the statistical results of the absolute deviation. These noise points are temporarily stored in a noise queue to prevent interference with subsequent feature extraction. After noise points are removed, the algorithm immediately uses a linear interpolation function to fill the gaps with valid data on both sides of the noise point, ensuring a continuous and smooth displacement curve, thus obtaining the purified sequence.

[0021] The purified sequence, as the sole input for subsequent time-frequency analysis, is fed into the variational mode decomposition (VM) step. The method first sets the target intrinsic modal component number based on an empirical threshold. Then, an adaptive loop is formed through iterative decomposition and reconstruction error evaluation. If the reduction in reconstruction error after several consecutive iterations is insufficient to indicate information gain, or if the residual energy proportion remains higher than the set threshold, the number of target intrinsic modal components is dynamically increased or decreased until the decomposition result can cover the main energy components in the purified sequence with the fewest possible intrinsic modal components and further compress random noise into the residual term. Each intrinsic modal component corresponds completely to the original purified sequence in the time domain. Therefore, the method can extract four indicators—instantaneous displacement amplitude, instantaneous displacement rate of change, center value of the dominant frequency interval, and cumulative energy—for different modes at the same timestamp, and write them together with the timestamp into the feature description, thus establishing a multi-dimensional index snapshot that spans both modes and time. All feature descriptions are then written into the feature sequence in ascending order of timestamp, forming the basic dataset used in this invention for subsequent multi-level displacement threshold setting and terminal construction parameter mapping.

[0022] Step 3: Use a fixed-step sliding window to scan the feature sequence and mark the feature descriptions in the feature sequence into three categories: quasi-stable region, asymptotic region, and abrupt region. Calculate the displacement increment of the intrinsic modal components corresponding to the feature descriptions of each category compared to the intrinsic modal components of the previous time series to form a gradient sequence, and set multi-level displacement thresholds based on the gradient sequence. Divide the three categories into multiple different gradient levels according to the multi-level displacement thresholds. Through a preset gradient and control quantity comparison table, map point by point to obtain four parameters: compaction energy level, compaction point spacing, reinforcement layer spacing, and reinforcement length, and compile them into a final construction parameter list. Firstly, a fixed-width, overlapping sliding window is introduced to achieve continuous observation without disrupting the original time sequence. The window width is set to 9 feature descriptions, and the step size is set to 3 feature descriptions. This combination ensures that any feature description is continuously included in multiple statistical analyses, thereby mitigating the impact of occasional data fluctuations on the overall judgment. Within each window, four indicators are selected: instantaneous displacement amplitude, instantaneous displacement rate of change, median value of the dominant frequency interval, and cumulative energy. Then, four statistical measures are calculated for each indicator: median, range, interquartile range, and mean. The combination of a total of 16 statistical measures can simultaneously capture information from four dimensions: central trend, fluctuation amplitude, distribution skewness, and energy accumulation, compensating for the inability of a single indicator to fully characterize slope behavior.

[0023] To ensure that the judgment threshold gradually shifts with the overall slope condition rather than being statically fixed from the outset, after the method first traverses the first 10 windows, an initial baseline is constructed using the arithmetic mean of each statistic. Subsequently, the baseline is updated with a weight of 0.2 every 5 new windows, effectively allowing the baseline to remember historical information in an exponential decay manner while rapidly absorbing the latest background. Based on this adaptive baseline, if all 16 statistics in the current window are below 1.2 times the baseline, it indicates that the slope's instantaneous response deviates very little from the long-term background, classifying it as a quasi-stable zone. If at least one statistic falls between 1.2 and 2.0 times the baseline, but none exceed 2.0 times, it indicates that the slope has experienced gradual shift but remains controllable, classifying it as a gradual zone. If any statistic exceeds 2.0 times, it represents a sudden change in the stress-displacement combination, classifying it as a sudden change zone. Since sliding window statistics are essentially a local perspective, they may be misclassified due to extremely short-term disturbances. This method introduces segment smoothing correction: when an isolated abrupt change region is sandwiched between two quasi-stable regions, it is regressed to a quasi-stable region to avoid over-control; when two consecutive abrupt change regions are surrounded by asymptotic regions, the left and right windows are expanded by one to re-label them as abrupt change regions to avoid underestimating the risk of continuous mutations.

[0024] After segment marking is completed, within the same category segment, the difference in instantaneous displacement amplitude of the intrinsic modal components corresponding to adjacent timestamps is accumulated in chronological order to generate a gradient temporary storage table. If the category changes midway, the temporary storage table is closed and written to the gradient sequence. This approach ensures that the gradient sequence naturally carries the segment label and start and end times, allowing for backtracking of the entire displacement behavior process without additional storage. Subsequently, five representative points—the minimum, first quartile, median, third quartile, and maximum—of the cumulative displacement increment of the gradient sequence are statistically analyzed. The minimum and maximum values ​​are directly used as the upper and lower boundaries of the multi-level displacement thresholds, and the other three points are used to fill the intermediate levels in order of magnitude, forming a displacement threshold group of no less than 5 levels. If the difference between adjacent thresholds is less than 2 mm, the higher threshold is increased by 0.5 mm until the spacing requirement is met. This further avoids gradient grading distortion caused by excessively dense thresholds. Each cumulative displacement increment of the gradient sequence is then compared with this threshold group and quantified into a gradient level ranging from level one to the highest level, where higher levels represent greater displacement energy and higher risk. Finally, the preset gradient and control quantity comparison table is invoked, and the gradient level is mapped to four parameters: compaction energy level, compaction point spacing, reinforcement layer spacing, and reinforcement length. These parameters are then written into the terminal construction parameter list according to the timestamp.

[0025] Step 4: Perform dynamic compaction or reinforcement laying in a cycle according to the terminal construction parameter list, and collect displacement information and update the terminal construction parameter list in real time after each round of construction until all designed compaction points and reinforcement layers are completed.

[0026] In practice, before the dynamic compaction operation begins, the construction personnel directly access the latest terminal construction parameter list, set the corresponding gradient level of the compaction energy to the impact energy control unit of the compactor, input the compaction point spacing into the compaction point positioning device, and insert positioning markers on the slope surface according to the reinforcement layer spacing and reinforcement length. The compactor performs compaction point by point according to the trajectory planned in the list. After compaction is completed, the hammer drop distance, number of hammer blows, and actual compaction settlement are automatically recorded so as to be linked to displacement changes in the next displacement sampling. When the area covered by the dynamic compaction is completed, the reinforcement construction team lays, anchors, and tensions the reinforcement according to the reinforcement layer spacing and reinforcement length given in the list at the corresponding positions, and simultaneously records the laying tension force, anchoring depth, and residual displacement after tensioning. The operation sequence of the dynamic compaction and reinforcement processes can be flexibly alternated according to the site space conditions, as long as the compaction and reinforcement construction involved in each round is completed before each round of data collection. The displacement acquisition device is activated immediately after the construction machinery has moved to a safe distance. Aligned with the unified coordinate 3D grid reference points established in step 1, it acquires real-time displacement values ​​of the slope surface and benchmark displacement piles in high-frequency sampling mode. The original samples are timestamped and written into a new real-time displacement sequence, which is then submitted to step 2 for median filtering, variational mode decomposition, and feature description updates. The new feature sequence then proceeds to step 3 to complete sliding window scanning, segment marking, gradient sequence refresh, and gradient level recalculation, thereby generating a list of terminal construction parameters consistent with the latest slope response. Since the four parameters—compacting energy level, compaction point spacing, reinforcement layer spacing, and reinforcement length—are all directly mapped from the gradient level, these parameters are updated immediately whenever the gradient level changes due to variations in slope dynamics. This ensures that subsequent construction activities always closely follow the slope deformation trend rather than relying on prior assumptions. To avoid repeated shutdowns due to excessively high closed-loop frequency, the method defaults to initiating the next data acquisition-calculation-update process only after completing one round of compaction and reinforcement. If the gradient level fluctuation is less than one level in two consecutive updates, the next data acquisition interval is automatically extended to reduce construction interference. If the proportion of the abrupt change zone increases significantly in a short period, the data acquisition interval is shortened to respond quickly to anomalies. The entire cycle only ends when all design compaction points have been compacted, all reinforcement layers have been installed, and the final displacement monitoring result is generally within the quasi-stable zone. This ensures that the slope has reached an acceptable displacement convergence and strength level before shutdown, while retaining a complete data chain for acceptance testing and subsequent operation and maintenance analysis.

[0027] Furthermore, in step 1, to ensure the accuracy of subsequent displacement monitoring and dynamic compaction-reinforcement parameter mapping, it is necessary to fully balance the relationship between measurement coverage, angular resolution, and point cloud density during the initial geometric data acquisition stage. The multi-view encirclement measurement method using no fewer than three stations is adopted because a single station is difficult to achieve full coverage scanning due to terrain obstruction and slope inclination. By setting up laser stations at the slope front, slope shoulder, and slope side, and forming intersecting observation chains, redundant bundles can be constructed in three-dimensional space, significantly reducing the proportion of obscured and shadowed areas, thereby avoiding the amplification of three-dimensional mesh interpolation errors caused by data voids during subsequent registration. The constraint of a single-station angular resolution of no more than 0.01° ensures that each station scans smoothly and delicately in the angular domain, sufficient to distinguish key geometric details such as minute folds, slope toe reversals, and local concavities and convexities. These details often correspond to stress concentration areas during dynamic compaction-reinforcement and are sensitive inputs for the multi-parameter dynamic control model to adjust energy levels and correct layer spacing. After merging multiple point cloud sites, the point cloud density is no less than 2500 points / m², meaning there are at least 2500 spatial coordinate samples per square meter. The spatial sampling interval is approximately 2 cm. This granularity allows the calculated 3D mesh to accurately reproduce the slope curvature radius and local inflections, keeping the point-to-surface registration error during displacement monitoring within millimeters. This further ensures that the instantaneous displacement amplitude in the feature sequence is highly consistent with the actual deformation amplitude. Regarding the deployment of displacement monitoring benchmarks, at least five benchmark displacement piles are linearly deployed along the slope toe. This establishes a sufficiently dense displacement sampling array at the slope toe, the control line where deformation is most significant and runs the entire length, avoiding single-point failures or individual sensor drift masking the overall trend. The spacing between adjacent benchmark displacement piles is no less than 2% of the slope length, ensuring redundancy between piles and preventing the workload of displacement sensor deployment from spiraling out of control due to blind densification. More importantly, it ensures that the displacement sequence spatially meets the Nyquist sampling requirements, preventing the loss of higher-order deformation information during digital filtering and variational mode decomposition. By combining multi-view encirclement measurement with a linear layout strategy of benchmark displacement piles, the method simultaneously establishes a high-precision, highly redundant data base in both spatial geometric capture and temporal displacement measurement dimensions.

[0028] Furthermore, after the real-time displacement sequence enters step 2, it is first written into the initial sequence table according to the sampling order. The significance of this operation is to maintain the original time structure of the displacement data, so that the subsequent filtering and decomposition algorithms can accurately capture the real time distance correlation between adjacent sampling points during the sliding process. Subsequently, a median filter window of odd length and central symmetry is used to slide point by point through the initial sequence table. The width of the median filter window is fixed at 5 data points. The reason for choosing a window width of 5, rather than larger or smaller, is to strike a balance between smoothing capability and detail fidelity: a window that is too small cannot effectively eliminate random jump points, while a window that is too large will smooth out high-frequency displacement characteristics, especially in short-period vibration signals generated immediately after compaction or during the reinforcement tensioning stage, where a large window may weaken useful information. As a nonlinear filtering method based on rank statistics, median filtering has a strong ability to suppress single-point impulse noise, while maintaining good fidelity for waveform edges and abrupt changes. Therefore, it is very suitable for handling the instantaneous measurement point jitter commonly seen in dynamic compaction-reinforcement conditions.

[0029] To further improve sensitivity to outliers without introducing excessive false positives, this invention introduces the robust statistic of median absolute deviation to scale the data within the window: once the median value of the window is determined, the absolute values ​​of the differences between all sample points in the window and the median value are calculated, and the median value is taken as the median absolute deviation. A discrete noise threshold is then set by multiplying the median value by 3. If the difference between the center point of the window and the median value exceeds this threshold, the center point is identified as a discrete noise point. Using the median absolute deviation by 3 instead of a fixed numerical threshold allows the judgment criteria to dynamically change with the window content, maintaining adaptability to signals of different stages and amplitudes, and avoiding false positives or false negatives caused by the slope's static stability period and the high-vibration period during construction sharing the same static threshold. Identified discrete noise points are not immediately deleted but temporarily stored in a noise queue. This facilitates subsequent quality assessment and sensor health diagnosis, while preventing direct removal from causing index confusion. After all sliding is complete, the algorithm marks noise points as empty slots in the initial sequence list, but does not change the relative positions of other data, maintaining the continuous time series characteristics. Next, linear interpolation is used to fill the gaps using valid data from adjacent sides. Linear interpolation is chosen because displacement sampling typically has a sufficiently high frequency, the time interval between adjacent sampling points is very short, and the displacement change satisfies the approximate linear assumption at this scale. Simultaneously, linear interpolation is computationally simple, avoiding overfitting or ringing effects. The resulting continuous cleaned sequence after interpolation not only removes the interference of discrete jump points but also completely inherits the timestamp order of the original sampling, allowing for seamless transfer to subsequent variational mode decomposition steps for multi-scale feature extraction. The high integrity and high fidelity of the cleaned sequence directly determine the decomposition accuracy of subsequent intrinsic mode components, thus affecting the accuracy of sliding window statistics and gradient classification. Ultimately, this relates to whether the dynamic adjustment of the four key control quantities—compactment energy level, compaction point spacing, reinforcement layer spacing, and reinforcement length—can truly reflect the actual deformation state of the slope.

[0030] Furthermore, in step 2, the purified sequence is considered as a complete and continuous original signal. The target intrinsic modal components are initially set to 8 because 8 intrinsic modal components can cover the main frequency bands of the slope vibration signal without significantly increasing the computational burden, allowing the high-frequency impact effect, low-frequency overall settlement, and mid-frequency tension response to be mapped separately. Iterative decomposition is then performed. In each iteration, variational mode decomposition is used to adaptively split the signal's bandpass, while the reconstruction error for that round is calculated in real time and compared with the previous round. If the reduction in reconstruction error for two consecutive rounds is less than 0.001, it indicates that further refinement of the components is unlikely to yield effective information gain, and the decomposition can be considered nearly convergent. If the convergence condition is not met after 500 iterations, the process is forcibly stopped for efficiency reasons to avoid overfitting.

[0031] After stopping, check if the residual energy ratio of the original energy is higher than 1%. If it still exceeds the limit, increase the number of target intrinsic modal components by a step size of 2 and re-decompose. This approach refines the frequency bands step by step to recapture the residual energy until the residual energy ratio is no higher than 1%, ensuring that the main energy of the signal is fully decomposed without missing key shaping components. After decomposition, each intrinsic modal component corresponds strictly one-to-one with the original timestamp. Thus, at each time node, four indicators can be extracted sequentially: instantaneous displacement amplitude, instantaneous displacement rate of change, center value of the dominant frequency interval, and cumulative energy. These four indicators jointly describe the four dimensions of amplitude, rate, frequency, and energy, enabling the method to perceive both instantaneous peaks and capture cumulative effects. Subsequently, the timestamp and the above four indicators are written into the feature description in a fixed field order, forming a five-field structure, maintaining stable field meaning and arrangement. To ensure that subsequent sliding window scanning can be read correctly along the time flow, all feature descriptions must be written into the feature sequence in ascending order of timestamps. This ensures that any step-size-based window movement will not result in time reversal or jumps.

[0032] Furthermore, in step 3, the fixed-step sliding window is set to a width of 9 and a step size of 3 to achieve a dynamic balance between temporal continuity and statistical robustness: a window width of 9 means that the local behavior at each moment is jointly influenced by the four adjacent feature descriptions, thus suppressing single-point anomalies; while a step size of 3 ensures 66% overlap between windows, allowing the sliding sequence to accurately capture rapid evolution without generating redundant calculations. As the window moves in ascending time order, the method synchronously reads four indicators for each feature description in the temporary window dataset: instantaneous displacement amplitude, instantaneous displacement rate of change, median value of the dominant frequency interval, and cumulative energy. Within the same window, it calculates four statistical measures for these four indicators: median, range, interquartile range, and mean, for a total of 16 statistical measures, which are written into the window indicator vector. Here, a dual statistical structure of "central tendency + dispersion" is adopted, which essentially allows each window to have representative values ​​and fluctuation ranges in the four physical dimensions of amplitude, rate, frequency, and energy, thereby avoiding an overly one-sided understanding of the slope state due to relying solely on a single mean or extreme value. The initial baseline is calculated using the first 10 window index vectors during the initial traversal phase. The arithmetic mean of each index is taken. The implicit assumption is that the early operating conditions are often in a relatively stable or predictable range. Therefore, the first 10 windows can provide a reference background for subsequent deviation detection.

[0033] As the window slides forward, the method triggers a baseline update after accumulating five new windows. A weighted average is used to merge the new window mean with a weight of 0.2, while retaining the old baseline with a weight of 0.8. This design is equivalent to a recursive exponential smoothing: the larger weight, incorporating historical information, prevents the baseline from drastically changing with instantaneous fluctuations, while the 0.2 increment ensures the baseline migrates slowly to adapt to the long-term evolution of the slope response. In this way, even if an increase in compaction energy or a change in reinforcement layer spacing leads to an overall increase in the instantaneous displacement amplitude, as long as the change is gradual, the baseline can automatically follow within a few iterations, avoiding misinterpreting normal trends as abrupt jumps. If the slope enters a true abrupt jump phase, the range or interquartile range in the window indicators will quickly exceed the baseline threshold by 1.2 to 2.0 times, rapidly triggering gradient upgrades.

[0034] Furthermore, applying threshold grading to the sliding window index vector is a crucial step in transforming statistically significant abnormal fluctuations into engineering semantics. The method uses the baseline as a dynamic reference, comparing each of the 16 statistics in the current window index vector with the corresponding statistics of the baseline: if all statistics are less than 1.2 times the baseline, it indicates that the instantaneous amplitude of displacement, instantaneous rate of change of displacement, central value of the dominant frequency interval, and cumulative energy within the window have not shown significant amplification in terms of central trend and dispersion, and the slope is in a controlled micro-deformation state. Therefore, the corresponding feature description within the window is marked as a stable zone. When at least one statistic is between 1.2 and 2.0 times the baseline and no statistic exceeds 2.0 times, it means that the slope has exhibited gradual behavior exceeding daily fluctuations but has not yet reached a sudden leap. Therefore, the corresponding feature description within the window is marked as a gradual zone. Once any statistic exceeds 2.0 times the baseline, it indicates a rapid amplification of the displacement-energy-frequency combination within a short period, posing a risk of sliding or local instability. Therefore, the corresponding feature description within the window is marked as a leap zone. While the aforementioned thresholds can quickly classify local behaviors, the sliding window is still essentially a local perspective and is susceptible to accidental high-amplitude or transient measurement errors. Therefore, this method introduces a spatial smoothing correction strategy: if a sudden jump zone window is sandwiched between two stable zone windows, the corresponding feature description within that window is regressed to a stable zone, considered a brief pulse rather than a sustained danger. If two adjacent sudden jump zone windows are surrounded by two progressive zone windows, the labels of these two windows are retained, and one window is extended to each side. The feature descriptions corresponding to the extended windows are also labeled as sudden jump zones, thus fully capturing the period of potentially continuous accelerated deformation. By combining threshold classification with smoothing correction, the method achieves time-varying identification of slope conditions: the compaction energy level corresponding to stable zones remains or is slightly adjusted; the spacing between compaction points and the reinforcement layer spacing are moderately tightened in progressive zones; and the compaction energy level is immediately increased and the reinforcement length is increased in sudden jump zones. Simultaneously, smoothing correction avoids excessive construction pauses due to single-point outliers and prevents underestimation of true continuous deformation. The final output segment markers not only provide real-time safety warnings, but also serve as the dividing line for gradient sequence construction. This ensures that subsequent multi-level displacement thresholds, gradient level divisions, and mappings of compaction energy levels, compaction point spacing, reinforcement layer spacing, and reinforcement length have clear, coherent, and physically based spatiotemporal boundaries, thereby ensuring that dynamic compaction-reinforcement construction can dynamically respond to the actual deformation evolution of the slope while maintaining efficiency.

[0035] Furthermore, in step 3, the method first reads the feature descriptions one by one in ascending chronological order, and automatically assigns the feature descriptions to the current segment based on the previous division results of the stable region, gradual region, and abrupt region. Within the same segment, the method retrieves the intrinsic modal components corresponding to adjacent timestamps in real time and calculates the difference in their instantaneous displacement amplitude. If the difference is negative, the absolute value is immediately taken to ensure that the gradient metric represents a scalar increment rather than a directional change, thereby avoiding a reduction in the accuracy of the actual cumulative displacement assessment due to sign cancellation in the subsequent accumulation process. After each difference calculation is completed, the method immediately writes the difference into the gradient temporary storage table in chronological order and continuously accumulates it to form the real-time displacement increment curve of the segment, so that the temporary storage table retains both the original time series and records the fine-grained deformation trajectory of the continuous slope response. When the method detects a change in segment category—for example, from a gradual transition to a sudden transition—it immediately executes the gradient temporary table closing operation. The segment category, start timestamp, end timestamp, and cumulative displacement increment recorded in the current temporary table are completely encapsulated into a single record and written to the gradient sequence. At the same time, the internal pointer is reset and a new gradient temporary table is opened for the next segment. Through this "open-close-continue" mechanism, the gradient sequence is kept strictly aligned with the segment classification in space and ensures seamless connection in time, preventing data overlap or omission.

[0036] Since each gradient sequence record clearly indicates the segment category, subsequent multi-level displacement threshold construction and gradient level classification can be performed by separately calculating the minimum, quartile, and maximum values ​​according to the category, ensuring both the sensitivity and reliability of the threshold system. The storage of start and end timestamps provides precise positioning for construction scheduling, facilitating the tracing of the trigger points on the time axis for adjustments to specific compaction energy levels, compaction point spacing, or reinforcement layer spacing. The cumulative displacement increment field directly reflects the true scale of deformation energy within each response stage and is the core bridge for converting static classification results into dynamic control quantities. Through this field, the method can quantify gradual or abrupt trends into specific gradient levels in real time, thereby driving the increase of compaction energy level or reinforcement length, achieving precise coordination between slope dynamic compaction and reinforcement construction within the same closed loop. Overall, the hierarchical design of the gradient temporary table and gradient sequence takes into account real-time performance, completeness, and traceability. This allows the method to continuously capture the details of slope deformation evolution during continuous construction and promptly map these details into specific and executable construction parameters. This ensures that the adjustment of compaction energy, compaction point spacing, reinforcement layer spacing, and reinforcement length is both fast and accurate, ultimately achieving the engineering goal of improving construction safety and slope stability.

[0037] Furthermore, in step 3, the gradient sequence has accumulated displacement increments for each response segment. To truly transform these raw increments into control signals that can drive the terminal construction parameter list, it is necessary to first construct a multi-level displacement threshold set that can both reflect the overall deformation amplitude distribution of the slope and has sufficient resolution. The method first performs full-sample statistics on the cumulative displacement increments of the entire gradient sequence, extracting five representative statistics: the minimum, the first quartile, the median, the third quartile, and the maximum. The reason for choosing the minimum and quartile system instead of the mean or standard deviation is twofold: firstly, gradient sequences often exhibit skewed distributions, and extreme values ​​may drag down the mean, causing distortion; secondly, the quartile system is naturally suitable for describing the internal structure of skewed data while maintaining sensitivity to tail risks. Subsequently, the minimum and maximum values ​​are used as the lowest and highest levels of the multi-level displacement thresholds, respectively. The remaining three statistics are then used to fill the intermediate levels in ascending order, thus forming an initial multi-level displacement threshold set of no less than five levels.

[0038] At this point, although the threshold set is complete, the difference between adjacent threshold levels may be too small. If it is less than 2 mm, the gradient levels will be so dense that they can be overwhelmed by measurement errors. To avoid this, the method performs a secondary correction on the threshold set: if the difference between adjacent threshold levels is detected to be less than 2 mm, the lower threshold level is retained, and the higher threshold level is increased in steps of 0.5 mm until the adjacent difference is not less than 2 mm; if the increase touches the next or highest threshold level, the touched threshold level is simultaneously shifted upwards by the same increment to maintain the overall hierarchical structure. This low-to-high incremental correction method can prevent the threshold set from being pushed up as a whole, resulting in the loss of lower-level thresholds, while ensuring that the final threshold difference is not less than 2 mm, providing a clear boundary for subsequent gradient determination.

[0039] After obtaining the final threshold set, the method iterates through the gradient sequence in chronological order, comparing each cumulative displacement increment with the multi-level displacement threshold set one by one: if the cumulative displacement increment is less than the lowest threshold, it is recorded as the first-level gradient; if it is between two adjacent thresholds, it directly corresponds to a higher-level gradient; if the increment is equal to or greater than the highest-level threshold, it is recorded as the highest-level gradient. The "rounding up" judgment principle is used here for safety redundancy considerations—when the displacement increment is between two thresholds, a higher-level response ensures that control quantities such as compaction energy level and reinforcement length keep up with the slope deformation trend in a timely manner, without delaying risk suppression due to conservative estimation. After all gradient judgments are completed, the method writes the judgment results into the gradient level table, maintaining a one-to-one correspondence with the timestamp, so that subsequent mapping of compaction energy level, compaction point spacing, reinforcement layer spacing, and reinforcement length can quickly locate the deformation level at that time using the time axis as an index. In this process, the minimum-maximum value establishes the upper and lower boundaries of the displacement threshold range, while quartile statistics provide an objective basis for internal stratification; the secondary correction ensures that the threshold spacing is greater than 2 mm but not excessively sparse, leaving sufficient adjustment space for high-precision loading and reinforcement control; the "rounding up" mapping ensures that the actual displacement increment always drives a construction response no less than required, ensuring a safety margin.

[0040] Furthermore, the preset gradient and control quantity comparison table is responsible for seamlessly mapping displacement gradient levels to on-site operable parameters. Its core concept is to simultaneously regulate compaction and reinforcement using an engineering adjustment strategy that corresponds to increasing energy, decreasing spacing, and increasing length in response to increasing risk. The comparison table first sets no fewer than 5 gradient levels, and pre-sets a unique combination of four parameters for each gradient level: compaction energy level, compaction point spacing, reinforcement layer spacing, and reinforcement length. This ensures that once the gradient level is determined, on-site personnel can directly call the corresponding combination without further judgment. The compaction energy level is measured in integer ton-meters, with a lower limit of 1 ton-meter. Each subsequent level must increase by at least 2 ton-meters. This rule ensures that the increase in compaction energy level has a significant range of adjustment while avoiding excessive aggression. The spacing between compaction points, measured in meters, is allowed to vary between 0.8 meters and 2.0 meters. It is required that the spacing decreases as the gradient increases, with a decrease of at least 0.2 meters per level. This allows for a denser compaction point layout corresponding to high displacement gradients to further strengthen the foundation. The reinforcement layer spacing also adopts a decreasing logic, allowing it to tighten as the gradient increases within the range of 0.3 meters to 1.0 meters. Each level must reduce the spacing by at least 0.1 meters to increase the reinforcement density in high-risk sections. The reinforcement length is proportional to the risk, increasing as the gradient increases within the range of 2.0 meters to 6.0 meters. Each level must increase the length by at least 0.5 meters, thereby simultaneously advancing the reinforcement consolidation depth below the potential failure surface. Through this combination rule, the reference table can generate specific sequences based on engineer settings or historical data calibration. For example, the first-level gradient corresponds to a compaction energy level of 1 ton-meter, a compaction point spacing of 2.0 meters, a reinforcement layer spacing of 1.0 meter, and a reinforcement length of 2.0 meters. The fifth-level gradient corresponds to a compaction energy level of 9 ton-meter, a compaction point spacing of 0.8 meters, a reinforcement layer spacing of 0.3 meters, and a reinforcement length of 6.0 meters. The intermediate levels are smoothly interpolated according to an increasing and decreasing rule. Since each row of parameters in the table is unique, when the gradient level table outputs a change in level on the time axis, the field control unit can immediately read the corresponding row, set the new compaction energy level to the compactor energy control, update the compaction point positioning grid, adjust the reinforcement roll laying pitch, and change the anchor bolt or reinforcement strip burial depth, forming a closed-loop scheduling mode with the shortest response link and the least manual judgment. The reference table itself can be determined in the early stages of the project through sensitivity analysis of past slope deformation-energy level-mass data, or it can be iteratively optimized during construction based on real-time feedback. It only needs to adhere to four hard constraints: energy level not less than 1 ton-meter with an increment of not less than 2 ton-meters; compaction point spacing not more than 2.0 meters with a decrease of not less than 0.2 meters; reinforcement layer spacing not more than 1.0 meter with a decrease of not less than 0.1 meters; and reinforcement length not less than 2.0 meters with an increment of not less than 0.5 meters. This will maintain the table's engineering feasibility and safety margin. Ultimately, the preset gradient and control quantity reference table serves as a bridge linking displacement gradient levels and compaction-reinforcement parameters. This allows the method to dynamically monitor displacement evolution while instantly translating risk perception into targeted energy application and reinforcement placement, fully leveraging the comprehensive advantages of multi-parameter dynamic control in improving slope stability and construction efficiency.

[0041] This example selects a long segment ,high Slope The fill slope. Before construction begins, markings are laid out along the slope toe. The reference displacement piles have a spacing of approximately [missing information]. .

[0042] The first step, high-precision laser terrain scanning, employs... Multiple stations are deployed in a multi-view encirclement, with a single station angular resolution. After merging, the point cloud density reaches After the scan is complete, rigid body transformation is used. Point cloud coordinates Mapped to the construction coordinate system; where for Rotation matrix, The translation vectors are obtained from the centroid of the least-squares matching reference displacement pile. Simultaneously, the displacement sensor uses... The frequency records the three-dimensional displacement of each reference displacement pile to form a real-time displacement sequence. .

[0043] The second step is to write the longitudinal displacement of each pile into the initial sequence table according to the sampling order. Apply width independently to each pile. Median filtering. Window center point. With window midpoint value If the difference satisfies If the noise is not clearly defined, it is classified as discrete noise and temporarily stored; after noise removal, linear interpolation is used. Fill in the gaps to obtain the purification sequence. .by Perform variational mode decomposition on the input. Initially set the target number of modes. The reconstruction error is calculated after each update. ,like And residual energy ratio Then stop; otherwise, when the iteration reaches... If the condition is still not met, then use the step size. Increase Decompose it again. Finally, you will get... intrinsic modal components At each timestamp The above is the extraction of all components. ; and together with the timestamp in a fixed order, they form a feature description. All feature descriptions are written to the feature sequence in ascending chronological order. .

[0044] The third step, The upper window width Step length Calculate the median value for the four indicators within each window. Range Interquartile range mean ,common A vector of statistics forms a window index. Before the first traversal The mean of each window is used as the initial baseline. Each cumulative New window updates baseline ,in For near Average of the indicators within the current window. If all statistics in the current window satisfy... If so, the window is marked as a stable region; if it exists And without excess Those are marked as asymptotic regions; if any statistic exceeds This is then marked as a jump zone. Isolated jump zones and clamped jump zones are then corrected according to smoothing rules. Afterwards, the instantaneous displacement amplitude difference is calculated for adjacent timestamps within the same segment. This is accumulated as the segment displacement increment. When a segment changes, the gradient temporary table is closed, and a gradient sequence entry is written. .

[0045] Find the minimum value of the cumulative displacement increment for the entire gradient sequence. First quartile Median Third and fourth quartile values Maximum value Based on this, an initial threshold set is constructed. Check adjacent differences: Therefore, Adjust to After further upward shifting, all differences are ≥2.0 mm, resulting in the final threshold group. The gradient level table is obtained by comparing the increments of gradient sequence entries with the set and rounding up. For example, the cumulative increment of a certain jump zone... lie in – Between these values, it is determined to be a fourth-level gradient.

[0046] The fourth step is to set the gradient and control quantity in the following table: The first-level gradient corresponds to the impact energy level. Spacing between ramming points Reinforcement layer spacing , length of reinforcement The second-order gradient corresponds to the impact energy level. Spacing between ramming points Reinforcement layer spacing , length of reinforcement The third-order gradient corresponds to the impact energy level. Spacing between ramming points Reinforcement layer spacing , length of reinforcement The fourth gradient corresponds to the impact energy level. Spacing between ramming points Reinforcement layer spacing , length of reinforcement The fifth gradient corresponds to the impact energy level. Spacing between ramming points Reinforcement layer spacing , length of reinforcement Taking the previously mentioned fourth-level jump zone as an example, the control unit immediately sets the impact energy of the rammer to... Automatically shrink the compaction grid to The spacing was determined, and the reinforcement team was instructed to adjust the reinforcement layer spacing of that section to [specified value]. Length increased to After this round of compaction and reinforcement is completed, the method restarts displacement acquisition and enters the next cycle. If the monitoring results of two consecutive rounds both fall within the first-level gradient and the displacement curve tends to be stable, the closed-loop determination indicates slope convergence, the dynamic compaction is stopped, and the reinforcement construction is terminated.

[0047] in, These are the slope length and height, respectively. These are the number and spacing of the reference displacement piles, respectively; These are the number of scanning stations, angular resolution, and point cloud density, respectively. The sampling frequency; These are point cloud registration, rotation, and translation, respectively. The first pile at time The displacement; These are the instantaneous amplitude of displacement, rate of change, center of dominant frequency, and cumulative energy, respectively. For window index vectors; The difference between adjacent amplitudes; These are the quintiles of cumulative displacement increment statistics; the threshold group elements are multi-level displacement thresholds; and the impact energy level units are... .

[0048] refer to Figure 2The sliding window scanning technique in this invention is the core component for identifying feature sequence segments. First, the feature sequence obtained after variational mode decomposition is arranged in ascending order of timestamps, forming a continuous feature description data stream. Each feature description includes key information such as timestamp, instantaneous displacement amplitude, instantaneous displacement rate of change, center value of the dominant frequency interval, and cumulative energy. The sliding window scanning process employs a fixed window width and a fixed step size. Specifically, the window width is set to include nine feature descriptions. This setting ensures sufficient statistical information within a limited data range while avoiding a decrease in temporal resolution due to an excessively large window. The window slides with a fixed step size of three feature descriptions. This step size ensures data coverage integrity while achieving moderate overlapping scanning, which is beneficial for capturing continuous features of change. During the sliding process, the first sliding window covers the first to ninth feature descriptions in the feature sequence, and these nine feature descriptions are processed and analyzed uniformly. Subsequently, moving forward three positions according to the fixed step size, the second sliding window covers the fourth to twelfth feature descriptions. Continuing with the same step size pattern, the third sliding window covers the seventh to fifteenth feature descriptions. This orderly sliding scanning method ensures comprehensive coverage of the entire feature sequence, while effectively identifying transitional regions of feature changes through overlapping areas between windows.

[0049] refer to Figure 3 , Figure 3 This is an experimental curve showing the instantaneous displacement amplitude variation of the present invention. The figure illustrates the variation of the instantaneous displacement amplitude over time, obtained through variational mode decomposition technology, and the corresponding segmentation results during the collaborative construction of a dynamic compaction-reinforced soil slope. Figure 3 As shown, the horizontal axis represents the time axis, with units of minutes, ranging from 0 to 100 minutes, reflecting the entire construction monitoring cycle. The vertical axis represents the instantaneous displacement amplitude, with units of millimeters, ranging from 0 to 60 millimeters, characterizing the instantaneous change in slope displacement. The thick solid line in the figure represents the original displacement sequence, which is the instantaneous displacement amplitude extracted after variational mode decomposition processing of the purified sequence obtained through real-time monitoring of the benchmark displacement pile. The curve shape shows that the instantaneous displacement amplitude exhibits obvious stage-wise changes throughout the monitoring period. In the 0-30 minute period, the instantaneous displacement amplitude slowly increases from an initial 2 mm to 15 mm, with a relatively gentle growth rate; in the 30-60 minute period, the instantaneous displacement amplitude accelerates from 15 mm to 45 mm, with a significantly faster growth trend; in the 60-100 minute period, the instantaneous displacement amplitude jumps sharply, rapidly rising from 45 mm to 55 mm before stabilizing.

[0050] The thin solid line and dashed line represent intrinsic mode component 1 and intrinsic mode component 2, respectively, which are separated from the original displacement sequence using variational mode decomposition technology. Intrinsic mode component 1 exhibits low-frequency periodic fluctuation characteristics, reflecting the long-term deformation trend of the slope; intrinsic mode component 2 exhibits mid-frequency oscillation characteristics, reflecting the medium-term dynamic response of the slope. These intrinsic mode components provide an important basis for subsequent feature description extraction and segment marking. The two vertical dotted lines in the figure mark the segment boundaries, dividing the entire monitoring time into three different technical segments: the quasi-stable zone (0-30 minutes), the gradual zone (30-60 minutes), and the abrupt transition zone (60-100 minutes). This segment division is determined based on the fixed-step sliding window scanning technology and the results of window index vector analysis, reflecting the technical feature of automatic segment identification in the multi-parameter dynamic control method of this invention. The experimental curves demonstrate that the variational mode decomposition technology of this invention can effectively extract key characteristic parameters such as instantaneous displacement amplitude, providing a reliable data foundation for subsequent gradient level classification and dynamic adjustment of construction parameters, and realizing intelligent monitoring and control of the slope construction process.

[0051] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for coordinated construction of dynamic compaction and reinforced soil slopes based on multi-parameter dynamic control, characterized in that, The method includes: Step 1: Use high-precision laser terrain scanning to acquire point clouds on the slope surface and convert them into a 3D mesh with unified coordinates; set a series of reference displacement piles at the toe of the slope, start real-time displacement recording, and obtain a real-time displacement sequence including the initial displacement baseline; register the 3D mesh with the initial displacement baseline in the real-time displacement sequence. Step 2: Perform median filtering on the real-time displacement sequence to remove discrete noise and obtain a cleaned sequence; perform variational mode decomposition on the cleaned sequence to split it into several intrinsic mode components; construct feature descriptions for each intrinsic mode component, and combine the feature descriptions of all intrinsic mode components into a feature sequence. Step 3: Use a fixed-step sliding window to scan the feature sequence and mark the feature descriptions in the feature sequence into three categories: quasi-stable region, asymptotic region, and abrupt region. Calculate the displacement increment of the intrinsic modal components corresponding to the feature descriptions of each category compared to the intrinsic modal components of the previous time series to form a gradient sequence, and set multi-level displacement thresholds based on the gradient sequence. Divide the three categories into multiple different gradient levels according to the multi-level displacement thresholds. Through a preset gradient and control quantity comparison table, map point by point to obtain four parameters: compaction energy level, compaction point spacing, reinforcement layer spacing, and reinforcement length, and compile them into a final construction parameter list. Step 4: Perform dynamic compaction or reinforcement laying in a cycle according to the terminal construction parameter list, and collect displacement information and update the terminal construction parameter list in real time after each round of construction until all designed compaction points and reinforcement layers are completed.

2. The method for coordinated construction of dynamic compaction-reinforced soil slopes based on multi-parameter dynamic control as described in claim 1, characterized in that, In step 1, the high-precision laser terrain scanning adopts a multi-view encirclement measurement method with no less than three stations, the angular resolution of a single station is no greater than 0.01°, and the point cloud density after merging is no less than 2500 points / ㎡; the number of reference displacement piles linearly arranged along the slope toe is no less than 5, and the distance between adjacent reference displacement piles is no less than 2% of the slope length.

3. The method for coordinated construction of dynamic compaction-reinforced soil slopes based on multi-parameter dynamic control as described in claim 2, characterized in that, In step 2, the real-time displacement sequence is written into the initial sequence table in the order of sampling; the initial sequence table is slid point by point using a median filter window of odd length, with the width of the median filter window set to 5 data points; if the difference between the displacement value of a certain point and the median value of the corresponding window is greater than 3 times the median absolute deviation of the same window, then the point is identified as a discrete noise point and temporarily stored in the noise queue; after the noise point is removed, the gap is filled by linear interpolation of the legal data on both sides to obtain a continuous clean sequence.

4. The method for coordinated construction of dynamic compaction-reinforced soil slopes based on multi-parameter dynamic control as described in claim 2, characterized in that, In step 2, the purified sequence is used as input, and the number of target intrinsic modal components is set to 8. Iterative decomposition is performed, and the process stops when the reconstruction error decreases by less than 0.001 for two consecutive rounds or when the number of iterations reaches 500. If the residual energy still accounts for more than 1% of the original energy, the number of target intrinsic modal components is increased by a step size of 2 and the decomposition is repeated until the residual energy ratio is no more than 1%. For each intrinsic modal component, at the position corresponding to the original timestamp, four indicators are extracted in sequence: instantaneous displacement amplitude, instantaneous displacement rate of change, center value of the dominant frequency interval, and cumulative energy. The five fields of timestamp, instantaneous displacement amplitude, instantaneous displacement rate of change, center value of the dominant frequency interval, and cumulative energy are written into the feature description in a fixed order. All feature descriptions are written into the feature sequence in ascending order of timestamp.

5. The method for coordinated construction of dynamic compaction-reinforced soil slopes based on multi-parameter dynamic control as described in claim 4, characterized in that, In step 3, a window width containing 9 feature descriptions is selected; the fixed step size of the window is set to 3 feature descriptions, and it slides in ascending order of time starting from the first position in the feature sequence; each slide extracts all feature descriptions within the window to form a temporary window dataset; for each feature description in the temporary window dataset, four indicators are read: instantaneous displacement amplitude, instantaneous displacement rate of change, center value of the dominant frequency interval, and cumulative energy; the median, range, interquartile range, and mean of the above four indicators are calculated respectively within the same window; the 16 statistics corresponding to the window are written into the window indicator vector in a fixed order; the initial baseline is calculated using the first 10 window indicator vectors obtained from the first traversal, and the arithmetic mean of each indicator is taken; Subsequently, whenever 5 new windows have been traversed cumulatively, a weighted average method is used to update the baseline with a weight of 0.2, so that the old baseline accounts for 0.8 and the new window mean accounts for 0.

2.

6. The method for coordinated construction of dynamic compaction-reinforced soil slopes based on multi-parameter dynamic control as described in claim 5, characterized in that, In step 3, if all 16 statistics in the current window's indicator vector are less than 1.2 times the corresponding baseline statistics, then all feature descriptions in the current window are marked as stable regions; if at least one statistic falls between 1.2 and 2.0 times the baseline and no statistic exceeds 2.0 times, then the corresponding feature description in the current window is marked as an asymptotic region; if any statistic exceeds 2.0 times the baseline, then the corresponding feature description in the current window is marked as a leap region; if a window with a leap region feature description is surrounded by two windows with stable feature descriptions, then the corresponding feature description in that window is marked as a stable region. If two windows with feature descriptions of abrupt transition regions are surrounded by two windows with feature descriptions of progressive transition regions, the labeling results of the feature descriptions within those windows are retained, but one window is expanded on each side, and the feature descriptions corresponding to these two transition windows are also labeled as abrupt transition regions.

7. The method for coordinated construction of dynamic compaction-reinforced soil slopes based on multi-parameter dynamic control as described in claim 6, characterized in that, In step 3, the feature description corresponding to each timestamp is read sequentially; the instantaneous amplitude difference of displacement is calculated for the intrinsic modal components of adjacent timestamps within the same segment. If the difference is negative, the absolute value is taken, and then written into the gradient temporary storage table in chronological order; when the segment category changes, the current gradient temporary storage table is immediately closed and completely written into the gradient sequence, and then a new gradient temporary storage table is opened for the next segment; each item in the gradient sequence contains the segment category, start timestamp, end timestamp, and cumulative displacement increment.

8. The method for coordinated construction of dynamic compaction-reinforced soil slopes based on multi-parameter dynamic control as described in claim 7, characterized in that, In step 3, the minimum, first quartile, median, third quartile, and maximum values ​​of the cumulative displacement increment in the entire gradient sequence are statistically analyzed, totaling five statistics. The minimum and maximum values ​​are used as the lowest and highest levels of the multi-level displacement threshold, respectively. The remaining three statistics are inserted into the intermediate levels in ascending order, forming a multi-level displacement threshold group of no less than five levels. If the difference between two adjacent threshold levels is less than 2 mm, the lower threshold level is retained and adjusted by incrementing by 0.5 mm towards the higher threshold level until the difference between adjacent threshold levels is not less than 2 mm. The cumulative displacement increment of each item in the gradient sequence is compared with the multi-level displacement threshold group one by one. If the cumulative displacement increment is less than the lowest threshold level, it is recorded as the first level gradient. If the cumulative displacement increment is between two adjacent threshold levels, it corresponds to a higher level gradient. If the cumulative displacement increment is equal to or greater than the highest level threshold, it is recorded as the highest level gradient. The results are written into the gradient level table, which is in one-to-one correspondence with the timestamp.

9. The method for coordinated construction of dynamic compaction-reinforced soil slopes based on multi-parameter dynamic control as described in claim 8, characterized in that, The preset gradient and control quantity comparison table contains no less than 5 gradient levels. Each gradient level corresponds to a unique combination of four parameters: compaction energy level, compaction point spacing, reinforcement layer spacing, and reinforcement length. Among them, the compaction energy level is expressed in integer tons-meters, not less than 1 ton-meter and increasing by no less than 2 tons-meters at each level; the compaction point spacing is expressed in meters, ranging from 0.8 meters to 2.0 meters, decreasing by no less than 0.2 meters at each level; the reinforcement layer spacing is expressed in meters, ranging from 0.3 meters to 1.0 meters, decreasing by no less than 0.1 meters at each level; and the reinforcement length is expressed in meters, ranging from 2.0 meters to 6.0 meters, increasing by no less than 0.5 meters at each level.

Images