A method and system for generating slope prevention and control measures based on multi-source monitoring data

Abstract

The present application relates to the technical field of slope intelligent monitoring, and particularly relates to a slope prevention and treatment measure generation method and system based on multi-source monitoring data, which comprises: acquiring slope multi-source monitoring data, extracting multi-dimensional monitoring features from the slope multi-source monitoring data; based on the monitoring features, acquiring future time sequence features by using a time sequence prediction model; based on the monitoring features and the future time sequence features, judging the stability grade of the slope and generating corresponding prevention and treatment measures by using a multi-modal fusion model. The method extracts multi-dimensional monitoring features based on multi-source monitoring data, realizes forward-looking prediction of slope stability relying on a time sequence prediction model, synchronously determines the stability grade by means of a multi-modal fusion model and generates measures, forms an intelligent decision-making closed loop, improves the adaptability and accuracy of the scheme, reduces the risk of instability, and has both engineering safety and economy.

Description

Technical Field

[0001] This invention relates to the field of intelligent slope monitoring technology, and in particular to a method and system for generating slope prevention and control measures based on multi-source monitoring data. Background Technology

[0002] Slope stability monitoring and prevention are core components in ensuring the safe operation of major infrastructure projects such as highways, railways, and mines. In recent years, with the rapid development and widespread adoption of sensor networks (such as GNSS, inclinometers, and fiber optic sensors) and remote sensing technologies (such as InSAR and LiDAR), multi-source integrated monitoring systems that combine deformation, stress, groundwater, and environmental information have become the mainstream technology for slope engineering safety monitoring. This system can comprehensively perceive the slope's condition, providing an unprecedented data foundation for stability analysis.

[0003] However, in the crucial step of transforming multi-source monitoring data into effective engineering decisions and precise prevention and control measures, existing technical solutions still have significant shortcomings, mainly in the following two aspects:

[0004] On the one hand, the generation of existing prevention and control measures relies heavily on the personal experience of professionals. Currently, geological survey reports, monitoring data, and other information obtained on-site usually require engineers to conduct comprehensive analysis based on their own experience before proposing corresponding support, drainage, or reinforcement schemes. This decision-making model based on human experience has inherent limitations: First, it has a strong response lag, making it difficult to adapt to the rapid evolution of slope risks in real time and dynamically. Actions are often taken only after the danger intensifies or a disaster occurs, leading to delays in prevention and control. Second, it is highly subjective and difficult to standardize. The experience and judgment of different experts may differ, making it difficult to guarantee the scientific nature, consistency, and optimality of prevention and control schemes, especially when facing complex and ever-changing slope conditions.

[0005] On the other hand, although multi-source monitoring systems can collect massive amounts of data, the heterogeneity of the data itself and the limitations of the analysis models hinder the effective transformation of data value into engineering decisions. Specifically: (1) Difficulty in data fusion: Monitoring data from different devices, different frequencies, and different physical meanings (such as millimeter-level displacement and megapascal-level stress) are not uniform in time, space, and dimensions, forming "data silos" that are difficult to deeply integrate and jointly analyze. (2) Disconnection between model application: Existing data analysis models, whether traditional statistical methods or some machine learning algorithms, mostly focus on the evaluation of historical states or the prediction of single indicators, failing to form a closed-loop linkage with specific engineering prevention and control measures. That is, the model can output "unstable" conclusions or predict deformation trends, but it cannot automatically generate, optimize, and recommend a set of economical, reasonable, and executable specific engineering measures (such as anchor bolt parameters, drainage ditch dimensions, etc.), resulting in a decision gap between "monitoring" and "treatment".

[0006] In summary, the core technological contradiction currently facing the field of slope safety lies in the mismatch between the increasingly abundant multi-source monitoring data and the lagging, experience-dependent prevention and control decision-making capabilities. How to achieve the automatic, accurate, and rapid generation of intelligent, adaptive prevention and control solutions from multi-source heterogeneous data has become a critical technical problem that urgently needs to be solved. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of existing technologies, such as reliance on human experience for slope prevention and control decisions, difficulty in converting multi-source monitoring data into engineering decisions, and the existence of decision-making gaps between monitoring and treatment. This invention provides a method and system for generating slope prevention and control measures based on multi-source monitoring data.

[0008] In a first aspect, the present invention provides a method for generating slope prevention and control measures based on multi-source monitoring data, comprising the following steps: Acquire multi-source monitoring data of slopes, and extract multi-dimensional monitoring features from the multi-source monitoring data of slopes; Based on the monitoring characteristics, future time-series characteristics are obtained using a time-series prediction model; Based on the monitoring characteristics and future time series characteristics, a multimodal fusion model is used to determine the stability level of the slope and generate corresponding prevention and control measures.

[0009] Preferably, the multi-source monitoring data of the slope includes: surface displacement data, deep displacement data, stress and strain data, and environmental data.

[0010] Preferably, the method further includes preprocessing the multi-source monitoring data of the slope, including: data cleaning, missing value imputation, and standardization.

[0011] Preferably, the monitoring features include: temporal dynamic features, spatial correlation features, and environmental coupling features.

[0012] Furthermore, the time-series prediction model is a long short-term memory network-attention hybrid model, used to predict future displacement trends and instability probabilities based on the monitoring features.

[0013] Preferably, the multimodal fusion model adopts a hierarchical architecture, including data layer weighted fusion, feature layer Transformer fusion, and decision layer classifier fusion.

[0014] Preferably, the stability level of the slope includes stable, basically stable, slightly unstable, and unstable.

[0015] Furthermore, the generation of corresponding prevention and control measures using a multimodal fusion model specifically includes: Calculate the corresponding importance score based on the Gini index of each monitored feature; Based on the monitoring feature with the highest importance score, a corresponding prevention and control measure library is matched; The objective function is constructed with the goals of maximizing the safety factor increment, minimizing the total cost, and minimizing the construction period. The Pareto optimal solution set is obtained by using a non-dominated sorting genetic algorithm. Finally, the comprehensive optimal solution is selected from the prevention and control measure library as the prevention and control measure corresponding to the stability level by using the fuzzy comprehensive evaluation method.

[0016] Preferably, the method further includes optimizing prevention and control measures and multimodal fusion models through a scheme dynamic modulation mechanism, wherein the scheme dynamic modulation mechanism includes: If the slope stability level does not meet the preset requirements after implementing prevention and control measures, optimize the prevention and control measures so that the slope stability level meets the preset requirements. Based on the optimized prevention and control measures, the multimodal fusion model is optimized.

[0017] In a second aspect, the present invention provides a slope prevention and control measure generation system based on multi-source monitoring data, comprising: The multi-source monitoring module includes displacement sensors, stress-strain sensors, deformation monitoring sensors, and environmental factor sensors, which are used to collect multi-source monitoring data of slopes; The data processing module is used to process the slope multi-source monitoring data using the slope prevention and control measures generation method based on multi-source monitoring data described in the first aspect, output the slope stability level, and generate corresponding prevention and control measures.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The first aspect of this invention provides a method for generating slope prevention and control measures based on multi-source monitoring data. By extracting multi-dimensional monitoring features from multi-source slope monitoring data, it overcomes the information limitations of single monitoring data. The constructed monitoring features can more comprehensively and accurately depict the evolution of slope state, laying a solid data foundation for subsequent stability judgment. Relying on a time-series prediction model to obtain future time-series features, it realizes the transformation of slope stability from passive response to active prediction, which can capture potential instability trends in advance and significantly improve the timeliness and foresight of prevention and control measures. By using a multi-modal fusion model to simultaneously complete the determination of slope stability level and the generation of prevention and control measures, it organically connects state analysis and decision generation, reduces intermediate links of manual intervention, and improves the efficiency and accuracy of scheme generation. The whole process constructs an intelligent decision-making closed loop from multi-source monitoring and feature fusion to forward prediction and judgment generation, which greatly improves the adaptability of prevention and control schemes, effectively reduces the risk of slope instability, and combines engineering safety and economy. 2. A second aspect of this invention provides a slope prevention and control measure generation system based on multi-source monitoring data. By integrating multiple types of sensors to form a multi-source monitoring module, it can comprehensively collect state data such as slope displacement, stress and strain, deformation, and environment, providing complete and reliable data support for subsequent monitoring feature fusion and model analysis. At the same time, the data processing module is deeply adapted to the method described in the first aspect, and can efficiently undertake the entire intelligent decision-making chain from data acquisition to measure generation, reducing manual intervention, ensuring the timeliness and accuracy of prevention and control measure generation, effectively reducing the risk of slope instability, and combining engineering safety and economy. Attached Figure Description

[0019] Figure 1 This is a flowchart of a slope prevention and control measure generation method based on multi-source monitoring data in Embodiment 1. Figure 2 A detailed flowchart is generated for a slope prevention and control measure based on multi-source monitoring data in Embodiment 1. Figure 3 This is a hardware deployment architecture diagram of the multi-source monitoring module in Embodiment 2. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to specific embodiments. However, this should not be construed as limiting the scope of the present invention to the following embodiments; all technologies implemented based on the content of the present invention fall within the scope of the present invention.

[0021] Unless otherwise specified, the terms "upper," "lower," "left," "right," "center," "inner," and "outer," etc., used in the description of specific embodiments of the present invention to indicate orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings, or the orientation or positional relationship in which the product / equipment / device is usually placed during use. These terms are merely for the purpose of facilitating the description of the present invention or simplifying the description in specific embodiments, and for enabling those skilled in the art to quickly understand the solution, and do not indicate or imply that a particular device / component / element must have a specific orientation, or be constructed and operated in a specific positional relationship. Therefore, they should not be construed as limitations on the present invention.

[0022] Furthermore, the use of terms such as "horizontal," "vertical," "suspended," "parallel," and "coaxial" does not imply that the corresponding device / component / element must be absolutely horizontal, vertical, suspended, parallel, or coaxial. Slight tilt or deviation is permissible, as long as it does not affect the normal function of the relevant component. For example, "horizontal" simply means that its direction is more horizontal relative to "vertical," not that the structure must be perfectly horizontal; a slight tilt is acceptable. "Coaxial" means that two components are arranged as coaxially as possible, allowing them to move coaxially or approximately coaxially when their relative positions change. Alternatively, it can be simplified to mean that the corresponding device / component / element, when arranged in "horizontal," "vertical," "suspended," "parallel," or "coaxial" directions, can have an error / deviation of ±10% relative to the corresponding direction, more preferably within ±8%, more preferably within ±6%, more preferably within ±5%, and more preferably within ±4%. For example, the deviation in the "coaxial" direction is controlled within 0.2-1mm, preferably within 0.2-0.5mm. As long as the corresponding device / component / element is within the error / deviation range, it can still achieve its function in the solution of the present invention.

[0023] Furthermore, the use of terms such as "first," "second," and "third" in terminology is merely for distinguishing descriptions of identical or similar components and should not be interpreted as emphasizing or implying the relative importance of a particular component.

[0024] Furthermore, in the description of the embodiments of the present invention, "several", "more than", and "a number of" represent at least two. The number can be any number, such as two, three, four, five, six, seven, eight, or nine, and can even exceed nine.

[0025] Furthermore, in the description of the technical solution of this invention, unless otherwise explicitly specified / limited / restricted, the terms "set up," "install," "connect," "link," "provided with," "laid out," and "arranged" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to connection methods commonly used in the art, such as welding, riveting, bolting, and threaded connections. Such connections can be mechanical, electrical, or communication connections; they can be direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components.

[0026] Example 1 like Figure 1 and Figure 2 The diagram shows a flowchart and detailed flowchart of a method for generating slope protection measures based on multi-source monitoring data, including the following steps: S1: Acquire multi-source monitoring data of the slope, extract and fuse monitoring features from the multi-source monitoring data of the slope, and generate monitoring features; In optional embodiments, the multi-source monitoring data for the slope includes, but is not limited to: surface displacement data, deep displacement data, stress-strain data, and environmental data. Specifically, the surface displacement data includes horizontal and vertical displacements to reflect the overall deformation trend of the slope; the deep displacement data includes pore water pressure and internal displacement of the soil and rock mass to capture minute deformations near potential sliding surfaces; the stress-strain data includes anchor cable stress and earth pressure to analyze the impact of changes in anchor cable stress and earth pressure on slope stability; and the environmental data includes rainfall, groundwater level, and soil moisture content to analyze the impact of rainfall, groundwater level, and moisture content on slope stability.

[0027] In an optional implementation, the method further includes preprocessing the multi-source slope monitoring data, including data cleaning, missing value imputation, and standardization. This preprocessing transforms the raw multi-source slope monitoring data into unified, complete, and high-quality structured data, providing reliable input for subsequent slope stability level assessment and prevention measures generation.

[0028] (1) Data cleaning For high-frequency data (5-10Hz) such as displacement / stress, Sym8 wavelet basis layered thresholding is used to preserve the true deformation trend and filter out electromagnetic interference. This can be expressed by the following formula:

[0029] in, This is the estimated value of the denoised signal. The original, noisy input signal. This is a scaling function used to capture approximate information about a signal at specific low frequencies. Wavelet functions are used to capture detailed information about a signal at specific high frequencies. These represent scale and translation, respectively. For signal In the scaling function Projection coefficients on For signal In wavelet function Projection coefficients on It is a contraction function. For threshold parameters, This represents the number of decomposition layers.

[0030] Specifically, for high-frequency data of 5-10Hz, a 5-level decomposition is performed using the Sym8 wavelet basis. A soft threshold function is applied to the high-frequency coefficients (levels 4-5), while the original trend is preserved for the low-frequency coefficients (levels 1-3) to avoid excessive smoothing that could lead to the loss of deformation features.

[0031] For sudden outlier data, a median filter with dynamically adjusted window size is used to protect the statistical distribution stability of the data required by the random forest.

[0032] (2) Missing value imputation If the missing data rate at a single point is less than 10%, time series linear interpolation is used; if the missing data rate is greater than 10%, data from adjacent measurement points are introduced, and spatial correlation is used to fill the gaps. The filling method can be expressed by the following formula:

[0033] in, 0 is the target point, Z( 0) is the target point 0 padding value, These are the weighting coefficients. Given neighboring points, Z( ) are known neighboring points The fill value, This represents the total amount of data.

[0034] (3) Standardization The timestamps and spatial coordinates of multi-source slope monitoring data must be consistent, as follows: Time alignment: Resample non-uniformly spaced data to a uniform time step using linear interpolation or cubic spline interpolation; Spatial alignment: Project the coordinates of all monitoring points to the same coordinate system, and combine the spatial location of each point with the slope DEM model to establish a three-dimensional data matrix of "space-time-attribute", which can be represented by the following formula:

[0035] Where θ is the slope strike angle, (ΔX, ΔY, ΔZ) is the coordinate translation, and (X, Y, Z) are the original coordinates of the monitoring point. , , () represents the coordinates of the aligned monitoring points.

[0036] Specifically, in the time alignment mechanism: in addition to BeiDou time synchronization, an NTP network time protocol server is added to perform time calibration on devices without built-in BeiDou modules, ensuring that the timestamp deviation of all data is less than 10ms.

[0037] In the spatial alignment mechanism: For GNSS monitoring points, precise coordinates are obtained through static measurement (continuous observation for 4 hours). These coordinates are then compared with the pixel coordinates matched by UAV imagery. If the deviation is >0.3m, an affine transformation model is used to correct the DEM model coordinates to ensure the accuracy of the "space-attribute" mapping.

[0038] In optional implementations, the monitoring features include, but are not limited to: temporal dynamic features, spatial correlation features, and environmental coupling features.

[0039] (1) Temporal dynamic feature extraction For time-series data such as displacement and seepage pressure, statistical features reflecting deformation rate and trend are extracted and comprehensively calculated: Absolute displacement:

[0040] in, For at any time The collected displacement values, To be at the selected initial reference time The collected displacement values, It represents the absolute displacement at time t (the cumulative displacement relative to the initial reference time).

[0041] Displacement rate:

[0042] in, For at any time instantaneous rate, For the current moment The displacement value; Before the current moment Displacement value at time t. The time interval used for calculation rate.

[0043] Displacement acceleration:

[0044] in, For at any time Instantaneous acceleration, For the current moment The instantaneous rate value; Before the current moment The instantaneous velocity value at time t. The time interval used to calculate acceleration.

[0045] variance:

[0046] in, The variance of the displacement samples, The total number of sample points within the time window used to calculate the variance. The first within the time window A relative displacement sample value, It is the arithmetic mean of all relative displacement sample values ​​within the time window.

[0047] (2) Spatial correlation feature extraction Combining the slope DEM model with the spatial coordinates of monitoring points, the spatial clustering of deformation is quantified using the spatial autocorrelation method, which can be expressed by the following formula:

[0048] in, These are characteristic values ​​representing the strength of the spatial autocorrelation of slope displacement. This is the spatial weight matrix. For the number of monitoring points, For monitoring points The cumulative displacement value, For monitoring points The cumulative displacement value, For variables In all The average value at each monitoring point. This indicates a positive correlation (adjacent points show consistent deformation trends). This indicates a negative correlation (there may be a shear failure zone).

[0049] (3) Environmental coupling feature extraction The hysteresis correlation between environmental factors such as rainfall and water level and deformation is analyzed, as shown in the following formula:

[0050] in, The lag correlation between rainfall, water level, and deformation. It is a rainfall sequence. It is a displacement sequence. The lag step size is in hours, and N is the total number of data samples collected.

[0051] S2: Based on the monitoring characteristics, use a time-series prediction model to obtain future time-series characteristics; In an optional implementation, the time-series prediction model is a long short-term memory network-attention hybrid model, used to predict future displacement trends and instability probabilities based on the monitoring features.

[0052] Specifically, to provide early warning of instability risks, a hybrid Long Short-Term Memory (LSTM) network-attention model is used to predict displacement trends over the next three days: (1) Input: Time series characteristics of the past 30 days (displacement rate, seepage pressure, rainfall, etc.); (2) LSTM layer: captures the long-term dependencies of time series and outputs the hidden state; specifically, LSTM is responsible for learning the "trend-period-abrupt change" features of slope deformation from time series data and outputting the predicted displacement value and hidden state feature vector for the next 3 days. (3) Attention layer: Calculate the attention weights at each time step and focus on key deformation stages; (4) Fully connected layer: output predicted displacement And the probability of instability.

[0053] Specifically, it also includes LSTM time-series feature pre-training: using transfer learning to initialize the LSTM model—first pre-training with a public slope dataset, and then fine-tuning with a small amount of measured data from the target slope to solve the problem of model underfitting in small sample scenarios.

[0054] S3: Based on the monitoring characteristics and future time series characteristics, use a multimodal fusion model to determine the stability level of the slope and generate corresponding prevention and control measures; In an optional implementation, the multimodal fusion model adopts a hierarchical architecture, including data layer weighted fusion, feature layer Transformer fusion, and decision layer classifier fusion.

[0055] Specifically, since the monitoring data includes structured data (displacement, seepage pressure), unstructured data (video, image) and spatial data (DEM), a layered fusion architecture is adopted: (1) Bottom layer fusion (data layer): The data of the same type of sensor are weighted and averaged, and the weights are dynamically adjusted by the sensor accuracy and historical error, as shown in the following formula:

[0056] in, For the first The fusion weight of each data source For the first The root mean square error of each data source For the first The root mean square error of each data source The total number of data sources involved in the fusion. (2) Mid-level fusion (feature layer): Input time-domain, frequency-domain, and spatial features into the Transformer model to capture the dependencies between different features, as shown in the following formula:

[0057] in, (Query) (key), The value is obtained by linear transformation of the characteristic matrix. For feature dimension, The output of the attention function is a weighted and fused feature representation. This is the normalization function.

[0058] Specifically, it also includes the embedded deployment of Transformer models: the Transformer model used in the mid-layer fusion compresses the number of parameters through knowledge distillation—using a large Transformer as the teacher model to train a small model, while maintaining 95% accuracy, the model size is reduced by 60%, making it suitable for edge computing terminals.

[0059] (3) High-level fusion (decision layer): The fused features are input into a support vector machine (SVM) or random forest (RF) to output the stability level of the slope. The stability level of the slope includes stable, basically stable, understability, and unstable.

[0060] Specifically, the evaluation of random forest fusion includes: extracting the temporal features from the LSTM output. By combining the static features with random forest, the stability level of the slope is determined and the warning level is output.

[0061] a) Feature concatenation: Constructing the input feature vector for a random forest ,in, For the final multi-dimensional fused feature vector, These are predicted values ​​for future multi-step displacements. These are real-time monitoring values ​​for specific monitoring points. This represents the total height of the slope. The internal friction angle of the rock and soil mass. The unit weight of the rock and soil mass. The friction angle of the potential sliding surface. The cohesion of the soil and rock mass, The length of the potential slip surface.

[0062] b) Random Forest Classification: The node splitting criterion uses a weighted index. ,in, This is the sample subset after feature splitting. The total number of classes in the dataset. For category The percentage of samples that are (stable / mostly stable / slightly unstable / unstable).

[0063] c) Early warning level mapping: Based on the displacement acceleration output by the random forest ( ,in For displacement acceleration, (Predicted future multi-step displacement values), based on displacement acceleration, determine the stability level and output an early warning, with the early warning level divided according to the following rules: Blue alert (i.e., stability level is "stable"): ; Yellow alert (i.e., stability level is "basically stable"): ; Orange alert (i.e., stability level is "understood"): ; Red alert (i.e., stability level is "unstable"): .

[0064] The warning level output by the algorithm directly triggers the prevention and control response process: a) Blue Alert: Level I Response Measures; b) Yellow Alert: Level II Response Measures; c) Orange alert: Level III response measures; d) Red Alert: Level IV Response Measures.

[0065] In an optional implementation, the step of generating corresponding prevention and control measures using a multimodal fusion model (i.e., triggering a prevention and control measure response process based on the warning level) specifically includes: Calculate the corresponding importance score based on the Gini index of each monitored feature; Based on the monitoring feature with the highest importance score, a corresponding prevention and control measure library is matched; The objective function is constructed with the goals of maximizing the safety factor increment, minimizing the total cost, and minimizing the construction period. The Pareto optimal solution set is obtained by using a non-dominated sorting genetic algorithm. Finally, the comprehensive optimal solution is selected from the preliminary prevention and control measures as the prevention and control measures corresponding to the stability level by using the fuzzy comprehensive evaluation method.

[0066] Specifically, prevention and control measures are generated based on the slope type, failure mode, and slope stability level, as shown in Table 1.

[0067] Table 1. Measures Type Library

[0068] Specifically, the measure recommendation logic is based on the matching of "Gini importance - measure" output by the algorithm. a) Importance score calculation: Calculate the importance score for each input feature. in, For split nodes, Features At the node During the split The index decreased. For the total quantity, No. Importance scores for each feature.

[0069] b) Matching of prevention and control measures library: If the Gini importance of rainfall is the greatest, then the measures library for drainage should be selected first; if the Gini importance of displacement acceleration is the greatest, then the measures library for support should be selected first.

[0070] It should be noted that each of Level I, Level II, Level III, and Level IV corresponds to a measure library, and each measure library can be further divided into a corresponding feature library based on its characteristics.

[0071] c) Multi-objective optimization to generate the optimal solution The objective function is constructed with the goals of maximizing the safety factor increment, minimizing the total cost, and minimizing the construction period. The Pareto optimal solution set is obtained by using a non-dominated sorting genetic algorithm. Finally, the comprehensive optimal solution is selected from the preliminary prevention and control measures as the prevention and control measures corresponding to the stability level by using the fuzzy comprehensive evaluation method.

[0072] Specifically, prevention and control measures need to balance three objectives: safety, economy, and construction difficulty, and construct an objective function: Safety objective: Maximize the increase in safety factor ; Economic objective: Minimize total cost ; Construction difficulty objective: Minimize construction period .

[0073] A non-dominated sorting genetic algorithm is used to find the Pareto optimal solution set, and finally, the comprehensive optimal solution is selected by fuzzy comprehensive evaluation method.

[0074] in, The target weights (determined by expert experience and owner needs) (i=1, 2, 3) For the goal The membership degree (range 0-1, 1 represents the optimal option, i=1, 2, 3).

[0075] For example, for the initial matching combination of measures, the parameters are optimized using the NSGA-II algorithm, and the objective function is:

[0076] in: The safety factor is C (>1.15), C is the cost of the measures (ten thousand yuan), and T is the construction period (days). These are the weights (default values ​​are 0.5, 0.3, and 0.2, respectively, and can be dynamically adjusted).

[0077] Based on the above methods, the prevention and control measures according to the warning level response can be: a) Blue alert: Regular monitoring, no action taken; b) Yellow alert: LSTM is used to predict the displacement in the next 7 days. If the displacement continues to exceed the threshold, the monitoring will be automatically intensified. c) Orange alert: Triggers the "Gini importance analysis" of random forest, prioritizing the recommendation of low-level measures; d) Red alert: Use the Bishop method to calculate the safety factor. ,like This triggered emergency anchoring measures.

[0078] Taking the Yibin-Yiliang Expressway slope pilot project as an example, the above method achieves an early warning accuracy rate of 95% (compared to 78% for the traditional threshold method); a cost reduction of 30% compared to experience-based solutions (slope protection schemes selected by experts or project leaders based on past engineering experience); and a response time reduction from 2 hours for manual assessment to 5 minutes for automatic algorithm output.

[0079] The warning threshold is linked to the emergency plan: When a blue warning is issued, the system automatically sends a "regular monitoring" instruction to the field terminal; when a yellow warning is issued, the system automatically sends a "encrypted monitoring (sampling frequency increased to every 10 minutes)" instruction to the field terminal; when an orange warning is issued, the system calls up the Level III (corresponding to the understability level) risk alternative plan (active protection net + intercepting ditch) in the measures library and generates a "site investigation within 3 days" task order; when a red warning is issued, the system directly triggers an emergency contact person's SMS + telephone notification and locks the material procurement list of the Level IV (corresponding to the unstable level) plan (anchor bolt + lattice beam) in the measures library.

[0080] It also includes a false alarm tracking mechanism: after each warning, the triggering characteristics are recorded. If no instability occurs in the following 7 days, it is automatically marked as a "false alarm". The cause of the false alarm is located by analyzing the SHAP value and the weight coefficient of the characteristic is updated.

[0081] In optional implementations, dynamic screening of monitoring features by random forest is also included: among 20+ features, features with importance <0.05 are eliminated every 10 iterations by combining recursive feature elimination (RFE) with the Gini importance score of random forest, and finally 8-10 core features are retained to improve the computational efficiency of the model.

[0082] In optional implementations, a fast call interface for mechanical verification is also included: in high-level fusion, the stability level output by the random forest synchronously calls the Bishop calculation engine, inputs the slope geometric parameters and geological parameters, and returns the safety factor within 5 seconds, realizing millisecond-level linkage of "AI assessment + mechanical verification".

[0083] In an optional implementation, the remaining static features in the future time-series features and monitoring features are normalized using Min-Max to eliminate the influence of dimensions: , in, At the point of time The standardized value, For at a certain point in time The collected monitoring values, This is the minimum value in historical data statistics. The maximum values ​​are statistically analyzed for historical data; for occasionally missing displacement data, an LSTM autoencoder is used to reconstruct the data (inputting 5 days of data before and after, and outputting missing values).

[0084] In an optional implementation, the engineering geological parameters of the slope (such as unit weight) are considered. internal friction angle Cohesion Based on historical disaster data, the safety factor is calculated using the limit equilibrium method. To verify the reliability of the slope stability level assessment:

[0085] in, For safety reasons, The weight of the soil strip is _____. The angle of inclination of the bottom surface of the soil strip. Pore ​​water pressure, The width of the soil strip. This is the effective stress intensity parameter. If... If so, it is determined to be unstable; It is not stable; For stability.

[0086] In optional implementations, the method further includes optimizing prevention and control measures and multimodal fusion models through a scheme dynamic modulation mechanism, wherein the scheme dynamic modulation mechanism includes: If the slope stability level does not meet the preset requirements after implementing prevention and control measures, optimize the prevention and control measures so that the slope stability level meets the preset requirements. Based on the optimized prevention and control measures, the multimodal fusion model is optimized. Specifically, post-construction monitoring data is input into the LSTM in real time to reassess its stability. a) If, one month after the implementation of the measures, the displacement rate predicted by LSTM drops to <1 mm / day and the random forest determines that it is "stable", then the current measures will be maintained; b) If the displacement rate is still >2mm / day, the algorithm automatically marks "measures failed", analyzes the cause of failure through SHAP value, recommends a higher level of prevention and control measures, and thus optimizes the output results of the multimodal fusion model.

[0087] This embodiment provides a method for generating slope prevention and control measures based on multi-source monitoring data. By extracting multi-dimensional monitoring features from multi-source slope monitoring data, it overcomes the information limitations of single monitoring data. The constructed monitoring features can more comprehensively and accurately depict the evolution of slope state, laying a solid data foundation for subsequent stability assessment. Relying on a time-series prediction model to obtain future time-series features, it realizes the transformation of slope stability from passive response to active prediction, which can capture potential instability trends in advance and significantly improve the timeliness and foresight of prevention and control measures. By using a multi-modal fusion model to simultaneously complete the determination of slope stability level and the generation of prevention and control measures, it organically connects state analysis and decision generation, reduces intermediate links of manual intervention, and improves the efficiency and accuracy of scheme generation. The whole process constructs an intelligent decision-making closed loop from multi-source monitoring and feature fusion to forward-looking prediction and judgment generation, which greatly improves the adaptability of prevention and control schemes, effectively reduces the risk of slope instability, and combines engineering safety and economy.

[0088] Example 2 Based on the same inventive concept, the present invention also provides a slope prevention and control measure generation system based on multi-source monitoring data, comprising: The multi-source monitoring module includes displacement sensors, stress-strain sensors, deformation monitoring sensors, and environmental factor sensors, which are used to collect multi-source monitoring data of slopes; Specifically, slope stability is influenced by multiple factors, including geological conditions, hydrological environment, meteorological factors, and human activities, making it difficult for a single type of data to comprehensively reflect the slope's condition. Therefore, this invention designs a multi-source integrated monitoring system (i.e., a multi-source monitoring module) covering multi-dimensional data acquisition from the surface, underground, stress-strain, and surrounding environment. Detailed data and hardware deployment are shown in the table below. Figure 2 .

[0089] Table 2 Multi-source monitoring data collection table

[0090] Hardware deployment should be carried out in accordance with the following requirements: all sensors should have a built-in BeiDou high-precision clock module to ensure that data timestamps are consistent to the nanosecond level.

[0091] (1) GNSS monitoring station: Deployed in stable areas at the top of the slope or the rear edge of the slope, avoiding vegetation and buildings, with a spacing of 50-100m (key points need to be densified to 20m) to ensure that there are no dead angles in satellite signal reception.

[0092] (2) Crack gauge: Deployed at both ends and the middle of surface cracks (spacing ≤ 5m), perpendicular to the crack direction; for potential cracks, denser deployment (spacing ≤ 2m).

[0093] (3) Strain gauges: 2-3 measuring points (mid-span and both ends) are arranged on the main reinforcing bars of the anchor rod (cable) and retaining wall steel skeleton.

[0094] (4) Stress gauges: symmetrically arranged along the axis of tension at the end of the free section of the anchor bolt and near the anchor head of the anchor cable (1-2m from the anchor head).

[0095] (5) Inclinometer: near the potential sliding surface inside the slope (depth ≥ 1.5 times the slope height), at the base of the support structure or in the area of ​​potential deformation weakness.

[0096] (6) Flexible displacement gauges: The potential sliding direction of the slope surface or deep part is to be set up every 5-10m along the sliding trajectory (each group has ≥3 measuring points).

[0097] (7) Vibrating wire piezometer: buried in the aquifer inside the slope, in the drainage blind ditch or intercepting ditch behind the support structure, at a depth of ≥1m (avoiding surface water accumulation areas).

[0098] (8) Earth pressure gauge: The vertical distance between the soil behind the retaining wall, the anchor plate or the stress surface of the soil nail wall and the surface of the support structure is ≤5cm.

[0099] (9) Inclinometers: top, bottom and middle of concrete retaining walls (horizontal spacing ≤ 5m, vertical spacing ≤ 3m per layer); flexible retaining structures (such as sheet piles) are equidistantly arranged along the pile body (spacing ≤ 2m).

[0100] (10) Rainfall monitoring station: open area at the top of the slope (≥10m away from any obstruction), avoiding the area affected by local topographic rainfall.

[0101] (11) Groundwater level monitoring equipment: The main runoff direction of the slope aquifer corresponds vertically to the surface monitoring point (depth difference ≤ 20m).

[0102] Optional implementation methods also include anti-interference and adaptive deployment measures for the multi-source monitoring module, including: (1) Sensor anti-interference enhancement design Electromagnetic shielding and signal enhancement: For equipment such as vibrating wire piezometers and strain gauges that are susceptible to electromagnetic interference, a double-layer metal shielding shell is used, and an adaptive filter chip is embedded in the signal transmission line to suppress 50Hz power frequency interference.

[0103] (2) Dynamic densification and cropping mechanism for monitoring points Adaptive deployment based on deformation trends: After initial deployment, if the displacement rate of a certain area is greater than 3 mm / day for 7 consecutive days, the "sensor cluster" will be automatically encrypted; if there is no significant deformation in a certain area for 2 months, redundant equipment will be removed to reduce operation and maintenance costs.

[0104] It also includes a data processing module, which is used to process the multi-source monitoring data of the slope using a slope prevention and control measure generation method based on multi-source monitoring data as described in Example 1, output the stability level of the slope, and generate corresponding prevention and control measures.

[0105] Specifically, this data processing module can be implemented in various physical forms and deployment methods, and its implementation does not depend on any specific hardware architecture or software environment. This module can be deployed on cloud servers, edge computing devices, dedicated computing platforms, or any combination of the above devices to form a flexible and scalable data processing unit.

[0106] For example, the computing platform on which this module relies may include one or more general-purpose or special-purpose processing units. For instance, the processor may include a central processing unit (CPU), a graphics processing unit (GPU), a tensor processing unit (TPU), a neural network processor (NPU), a field-programmable gate array (FPGA), or an application-specific integrated circuit (ASIC) customized for a specific algorithm. Different processing units can operate independently or be integrated into one or more processing chips through collaborative computing to optimize data processing efficiency and power consumption.

[0107] This module relies on a storage system for operation. The storage system stores executable program code, intermediate processing data, historical monitoring data, and related parameter models and rule bases. Storage media can include, but are not limited to, dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, hard disks, and newer non-volatile memory. The processor completes the entire process from data input and analysis to result output by calling and executing instructions and data from the storage system.

[0108] This module can exchange data and transmit commands with the front-end monitoring sensor network, the back-end early warning and release platform, and user interaction terminals via wired or wireless communication interfaces. It can be implemented as a standalone software application, a packaged service interface, or as a functional component embedded in a larger-scale monitoring system.

[0109] This data processing module combines the core logic of the method with a programmable hardware platform, enabling the transformation from method to physical device. This design allows the module to be integrated as a software solution into various information systems, or embedded in dedicated hardware to form an integrated intelligent processing terminal. This ensures stable, reliable, and efficient slope stability assessment and prevention measure generation services under different application scenarios and engineering needs. This embodiment provides a slope prevention measure generation system based on multi-source monitoring data. By integrating multiple types of sensors to form a multi-source monitoring module, it can comprehensively collect slope displacement, stress-strain, deformation, and environmental status data, providing complete and reliable data support for subsequent monitoring feature fusion and model analysis. Simultaneously, the data processing module is deeply adapted to the method described in Embodiment 1, efficiently undertaking the entire intelligent decision-making chain from data acquisition to measure generation, reducing manual intervention, ensuring the timeliness and accuracy of prevention measure generation, effectively reducing slope instability risk, and combining engineering safety and economy.

[0110] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for generating slope prevention measures based on multi-source monitoring data, characterized in that, Includes the following steps: Acquire multi-source monitoring data of slopes, and extract multi-dimensional monitoring features from the multi-source monitoring data of slopes; Based on the monitoring characteristics, future time-series characteristics are obtained using a time-series prediction model; Based on the monitoring characteristics and future time series characteristics, a multimodal fusion model is used to determine the stability level of the slope and generate corresponding prevention and control measures.

2. The method for generating slope prevention measures based on multi-source monitoring data according to claim 1, characterized in that, The multi-source monitoring data for the slope includes: surface displacement data, deep displacement data, stress and strain data, and environmental data.

3. The method for generating slope prevention measures based on multi-source monitoring data according to claim 1, characterized in that, It also includes preprocessing the multi-source monitoring data of the slope, including: data cleaning, missing value imputation, and standardization.

4. The method for generating slope prevention measures based on multi-source monitoring data according to claim 1, characterized in that, The monitoring features include: temporal dynamic features, spatial correlation features, and environmental coupling features.

5. The method for generating slope prevention measures based on multi-source monitoring data according to claim 4, characterized in that, The time-series prediction model is a hybrid model of long short-term memory network and attention, used to predict future displacement trends and instability probabilities based on the monitoring features.

6. The method for generating slope prevention measures based on multi-source monitoring data according to claim 1, characterized in that, The multimodal fusion model adopts a hierarchical architecture, including data layer weighted fusion, feature layer Transformer fusion, and decision layer classifier fusion.

7. The method for generating slope prevention measures based on multi-source monitoring data according to claim 1, characterized in that, The stability levels of the slopes include stable, basically stable, slightly unstable, and unstable.

8. The method for generating slope prevention measures based on multi-source monitoring data according to claim 7, characterized in that, The generation of corresponding prevention and control measures using a multimodal fusion model specifically includes: Calculate the corresponding importance score based on the Gini index of each monitored feature; Based on the monitoring feature with the highest importance score, a corresponding prevention and control measure library is matched; The objective function is constructed with the goals of maximizing the safety factor increment, minimizing the total cost, and minimizing the construction period. The Pareto optimal solution set is obtained by using a non-dominated sorting genetic algorithm. Finally, the comprehensive optimal solution is selected from the prevention and control measure library as the prevention and control measure corresponding to the stability level by using the fuzzy comprehensive evaluation method.

9. The method for generating slope prevention measures based on multi-source monitoring data according to claim 1, characterized in that, It also includes optimizing prevention and control measures and multimodal fusion models through a scheme dynamic modulation mechanism, wherein the scheme dynamic modulation mechanism includes: If the slope stability level does not meet the preset requirements after implementing prevention and control measures, optimize the prevention and control measures so that the slope stability level meets the preset requirements. Based on the optimized prevention and control measures, the multimodal fusion model is optimized.

10. A slope prevention and control measure generation system based on multi-source monitoring data, characterized in that, include: The multi-source monitoring module includes displacement sensors, stress-strain sensors, deformation monitoring sensors, and environmental factor sensors, which are used to collect multi-source monitoring data of slopes; The data processing module is used to process the multi-source monitoring data of the slope using the slope prevention and control measures generation method based on multi-source monitoring data as described in any one of claims 1 to 9, output the stability level of the slope, and generate corresponding prevention and control measures.

Images