Slope monitoring data processing and stability evaluation method based on multi-agent cooperation

Abstract

The application discloses a kind of based on multi-agent cooperation's slope monitoring data processing and stability evaluation method, the method includes: data agent drives data pre-processing, to the geometry parameter of slope, geotechnical physical and mechanical parameters, meteorological data, hydrological data and historical monitoring data are washed, missing filling, abnormal value is excluded, standardization, time alignment, window aggregation and static / dynamic feature fusion, form unified input dataset;Analysis agent unified dispatch bishop method stability calculation and XGBoost stability prediction, so that the current and historical safety factor sequence output by bishop method is used as machine learning model training target, verification basis or feature reference;Decision agent carries out joint determination based on current safety factor and future predicted safety factor, and outputs graded early warning and disposal suggestion.The application can improve the real-time performance, accuracy, interpretability and early warning lead time of slope stability evaluation.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring and slope engineering safety assessment technology, and in particular to a method for slope monitoring data processing and stability assessment based on multi-agent collaboration. Background Technology

[0002] Slopes are common engineering geological features in highways, railways, water conservancy projects, mining, and urban construction. Their stability directly affects the safety of engineering structures, the safety of people and property, and the stability of regional operations. Affected by factors such as rainfall infiltration, groundwater fluctuations, seismic disturbances, excavation unloading, and long-term weathering, slopes are prone to deformation, instability, and even landslides. Therefore, real-time monitoring, stability assessment, and risk early warning of slopes are of great significance.

[0003] Among existing slope stability analysis methods, the limit equilibrium method is widely used. The Bishop method, in particular, is widely used for calculating slope safety factors due to its relatively simple calculation process and good engineering applicability. This type of method typically analyzes the stability state of a slope at a specific moment based on slope geometric parameters, soil and rock physical and mechanical parameters, and groundwater conditions, and can reflect the slope's safety status to a certain extent.

[0004] With the development of monitoring technology, various methods such as GNSS, InSAR, displacement monitoring, pore water pressure monitoring, rainfall monitoring, and groundwater level monitoring have been gradually introduced into slope engineering. These methods can continuously acquire information on slope deformation, soil and rock conditions, and changes in the external environment. Meanwhile, machine learning methods have strong advantages in modeling complex nonlinear relationships. They can use historical monitoring data, meteorological data, and soil and rock parameters to predict the future stability of slopes, providing a new technical approach for dynamic slope safety assessment.

[0005] However, existing technologies still have the following shortcomings: On the one hand, traditional slope stability analysis based on the limit equilibrium method is mostly static analysis, which is difficult to effectively reflect the time-varying characteristics of slopes under dynamic environmental effects such as rainfall, groundwater changes and long-term deformation, and is insufficient in assessing dynamic stability under complex working conditions; on the other hand, existing slope monitoring often involves multi-source heterogeneous data, and different data sources have significant differences in sampling frequency, dimensions, accuracy and time scale. Existing technologies lack a unified data access, cleaning, standardization and time sequence alignment mechanism, resulting in low efficiency of multi-source monitoring data fusion and utilization.

[0006] Furthermore, existing technologies typically implement monitoring data processing, stability calculation, predictive analysis, and early warning decision-making in a decentralized manner, lacking a collaborative mechanism capable of automatically completing data processing, physical mechanism analysis, trend prediction, and early warning output. This makes it difficult to meet the application requirements of real-time, intelligent, and automated slope safety monitoring in complex engineering scenarios. In particular, relying solely on traditional theoretical models has limited ability to characterize complex nonlinear relationships; while relying solely on data-driven models suffers from insufficient physical interpretability, making it difficult to balance the accuracy and interpretability of stability assessments. Summary of the Invention

[0007] The purpose of this invention is to provide a method for slope monitoring data processing and stability assessment based on multi-agent collaboration, which can automatically complete multi-source monitoring data processing, stability calculation, predictive analysis and early warning output, thereby improving the automation level and operational efficiency of slope safety monitoring and early warning decision-making.

[0008] To achieve the above objectives, the present invention provides the following solution: A method for slope monitoring data processing and stability assessment based on multi-agent collaboration includes the following steps: S1. Obtain multi-source monitoring data of the slope monitoring area, including slope geometric parameters, physical and mechanical parameters of soil and rock, meteorological data, hydrological data, and historical monitoring data; S2. Construct a collaborative working mechanism among data intelligence agents, analytical intelligence agents, and decision-making intelligence agents; S3. Performed by the data intelligence agent: Cleaning, missing value processing, outlier removal, standardization, time alignment, and feature fusion of multi-source monitoring data to form an input dataset for slope stability assessment; S4. Performed by the analytical agent: Based on the slope's geometric parameters and the physical and mechanical parameters of the soil and rock mass, the Bishop method is used to calculate the safety factor of the slope under its current state; based on the input dataset and historical monitoring data, a pre-trained XGBoost model is used to dynamically predict the stability state of the slope at future moments. S5. Executed by the decision-making intelligent agent: Based on the safety factor under the current state and the stability state at future moments, jointly determine the stability state of the slope and output the risk warning level and disposal suggestions; S6. Outputs a graphical display of the slope safety factor variation trend, stability prediction results, and risk warning level.

[0009] Preferably, in S2, time alignment uses a preset time step Δt as a unified time reference, and performs resampling, timestamp matching, interpolation completion, and window aggregation on monitoring data with different sampling frequencies; feature fusion splices slope geometric parameters, soil and rock physical and mechanical parameters with rainfall, groundwater level, pore water pressure, displacement rate, displacement acceleration, and cumulative displacement into a fused feature vector.

[0010] Preferably, in S3, the Bishop method is used to calculate the safety factor of the slope under its current state, specifically including: The potential sliding body of the slope is divided into multiple slices. A moment balance relationship is established based on the geometric dimensions, weight, normal pressure, water pressure, and shear strength parameters of each slice, and the safety factor is iteratively solved. The weight of the i-th slice is... for:

[0011] Water pressure of the i-th slice for:

[0012] Slope safety factor satisfy:

[0013] in,

[0014] Where γ is the unit weight of the rock and soil mass, b i h is the width of the i-th slice. i Let γ be the height of the i-th slice. w z is the unit weight of water w,i Let c be the water level depth corresponding to the i-th slice. i Let φ be the cohesion of the i-th slice. i Let α be the internal friction angle of the i-th slice. i Let m be the inclination angle of the sliding surface of the i-th slice. i The correction coefficient is n, which is the total number of slices. The calculation is iterated until the difference between the safety factors obtained from two adjacent calculations is less than the preset error threshold. The minimum safety factor is obtained by traversing different sliding circle center positions and radii, which is used as the slope stability assessment result.

[0015] Preferably, in S4, the XGBoost model is used to dynamically predict the stability state of the slope at future times, and the objective function is as follows:

[0016] in, (·) represents the loss function, y i For the true value, i (t-1) This is the predicted value after the (t-1)th iteration. For the t-th tree, the input feature x i The predicted output, is the regularization term for the t-th tree, and n is the number of training samples; the XGBoost model takes the preprocessed slope geometric parameters, soil and rock physical and mechanical parameters, meteorological data, hydrological data and historical monitoring data as input features, and the stable state label as output to obtain the stability state prediction results of the slope at future times.

[0017] Preferably, S4 also includes pre-training of the XGBoost model: The input dataset is divided into training, validation, and test sets in chronological order and fed into the XGBoost model for training and validation. The root mean square error, mean absolute error, coefficient of determination, or accuracy of warning level identification are used as evaluation metrics. When the validation error exceeds the preset error threshold, the slope condition changes abruptly, or the number of new samples reaches the preset number, the model is retrained or incrementally updated. The model update mechanism includes three methods: scheduled batch update, event-triggered update, and error-triggered update. Scheduled batch update is used to retrain or fine-tune the model weekly, monthly, or quarterly using new samples. Event-triggered update is triggered after heavy rainfall, earthquake disturbance, excavation unloading, rapid rise and fall of reservoir water level, or replacement of monitoring equipment. Error-triggered update is triggered when the MAE or RMSE exceeds the preset error threshold for N consecutive prediction times, or when the number of misjudgments of the warning level exceeds the preset number.

[0018] Preferably, in S5, data flow and control flow interfaces are set between the data agent, the analysis agent, and the decision agent; the data agent outputs the input dataset, including time alignment rules and fused feature vectors; the analysis agent outputs the current safety coefficient, historical safety coefficient sequence, future stable state and prediction confidence information; the decision agent outputs the risk warning level, warning triggering reason and handling suggestions, and feeds back the warning results and model errors to the data agent and the analysis agent to optimize the next round of data fusion strategy and model update strategy.

[0019] Preferably, S5 also includes: Data flow closed loop: The data agent sends the input dataset to the analysis agent; the analysis agent calls the Bishop method and XGBoost models based on the fused feature vectors in the input dataset to generate the current safety factor, historical safety factor sequence, future stable state, and prediction error; the historical safety factor sequence and the corresponding fused feature vectors are written together into the training set cache for continuous updates of the XGBoost model; the analysis agent feeds back the safety factor results, error results, and feature importance to the data agent, which adjusts the next round of time alignment window, outlier identification threshold, missing value imputation method, or feature fusion weights accordingly; Control flow closed loop: When the decision agent recognizes that the warning level has increased, the prediction error has exceeded the threshold, the continuous rainfall process has started, or the monitoring data has changed abruptly, it sends control instructions to the data agent to increase the sampling frequency, shorten the time step Δt, or expand the historical window length, and sends control instructions to the analysis agent to recalculate the safety factor, retrain the XGBoost model, or increase the prediction frequency.

[0020] Preferably, in S5, the slope stability state is jointly determined based on the safety factor under the current state and the stability state at future times, specifically including: When the safety factor F in the current state s (t) has not yet reached the high-level warning threshold, but the stable state F at future moments s (t+k) When the value drops to a lower range or the rate of decline exceeds a preset threshold at multiple consecutive prediction times, the decision-making agent triggers a forward-looking warning of the corresponding level in advance; the warning levels are classified as follows: F s (t)>1.25 indicates a stable state with no warning; 1.15 <F s A yellow alert is issued if (t) ≤ 1.25, and 1.00 <F s (t)≤1.15 indicates an orange alert, F s A red alert is issued if (t) ≤ 1.00.

[0021] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a slope monitoring data processing and stability assessment method based on multi-agent collaboration as described above.

[0022] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: (1) By introducing a multi-source monitoring data acquisition and fusion processing mechanism, this invention can uniformly process slope geometric parameters, soil and rock physical and mechanical parameters, meteorological data, hydrological data and historical monitoring data, thereby improving the utilization efficiency of multi-source heterogeneous data in slope stability assessment. By combining the Bishop method with the XGBoost model, on the one hand, the Bishop method is used to calculate the safety factor based on physical mechanisms, thereby enhancing the interpretability of slope stability assessment. On the other hand, the XGBoost model is used to dynamically predict the future stability state of the slope, thereby improving the prediction accuracy and real-time performance under complex environments and time-varying conditions. (2) The present invention constructs a collaborative working mechanism composed of a data intelligence agent, an analysis intelligence agent and a decision intelligence agent, which can automatically complete the processing of multi-source monitoring data, stability calculation, prediction analysis and early warning output, thereby improving the automation level and operating efficiency of slope safety monitoring and early warning decision-making; it can automatically output graded early warning information based on the safety factor calculation results and prediction results, and display the slope stability status and change trend in a graphical manner, which makes it easy for engineers to quickly understand the monitoring results and take targeted measures, and has high engineering application value. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating the slope monitoring data processing and stability assessment method based on multi-agent collaboration provided by the present invention. Figure 2 This is a schematic diagram comparing the change of safety factor over time in an embodiment of the present invention; Figure 3 This is a scatter plot of the actual and predicted values ​​in an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0026] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] like Figure 1 As shown, the present invention provides a method for slope monitoring data processing and stability assessment based on multi-agent collaboration, comprising the following steps: S1. Obtain multi-source monitoring data of the slope monitoring area, including slope geometric parameters, physical and mechanical parameters of soil and rock, meteorological data, hydrological data, and historical monitoring data; S2. Construct a collaborative working mechanism among data intelligence agents, analytical intelligence agents, and decision-making intelligence agents; S3. Performed by the data intelligence agent: Cleaning, missing value processing, outlier removal, standardization, time alignment, and feature fusion of multi-source monitoring data to form an input dataset for slope stability assessment; S4. Performed by the analytical agent: Based on the slope's geometric parameters and the physical and mechanical parameters of the soil and rock mass, the Bishop method is used to calculate the safety factor of the slope under its current state; based on the input dataset and historical monitoring data, a pre-trained XGBoost model is used to dynamically predict the stability state of the slope at future moments. S5. Executed by the decision-making intelligent agent: Based on the safety factor under the current state and the stability state at future moments, jointly determine the stability state of the slope and output the risk warning level and disposal suggestions; S6. Outputs a graphical display of the slope safety factor variation trend, stability prediction results, and risk warning level.

[0028] Specifically, the method of the present invention includes: I. Overall Work Process The system acquires slope geometric parameters, soil and rock physical and mechanical parameters, meteorological data, hydrological data, and historical monitoring data. A data preprocessing module cleans the acquired data, handles missing values, removes outliers, standardizes, aligns the data over time, and fuses features to form an input dataset for subsequent analysis. Based on the slope geometric and soil physical and mechanical parameters, the Bishop method is used to calculate the safety factor of the slope under its current state. Based on the preprocessed multi-source data and historical monitoring data, the XGBoost model is used to dynamically predict the safety factor or stability state of the slope at future times. The system classifies the slope stability state into risk levels based on the safety factor calculation and prediction results and outputs corresponding early warning information. The trend of safety factor changes, stability prediction results, and early warning levels are displayed in the form of charts or graphical interfaces.

[0029] A collaborative working mechanism is constructed among data intelligence agents, analytical intelligence agents, and decision-making intelligence agents. The data intelligence agent receives multi-source slope monitoring data from monitoring equipment, survey data, historical databases, or manual input, and calls the data preprocessing module to complete data organization and input construction. The analytical intelligence agent calls the stability calculation module and stability prediction module to calculate and predict the current safety factor and future stability state of the slope. The decision-making intelligence agent completes risk identification, early warning level classification, generation of disposal suggestions, and result distribution based on the analysis results. Through this collaborative mechanism, automatic linkage of slope monitoring data processing, stability analysis, trend prediction, and early warning output can be achieved. The system design section of the paper explicitly adopts the collaborative working method of data intelligence agents, analytical intelligence agents, and decision-making intelligence agents.

[0030] The workflow of this embodiment is as follows: First, static parameters and dynamic monitoring information of the slope monitoring area are acquired; second, data from different sources, frequencies, and units are processed uniformly to form a structured input dataset; then, the Bishop method is used to calculate the current safety factor of the slope based on the current input parameters, and the XGBoost model is used to output the predicted safety factor or stability state prediction results for future times based on historical samples and the latest monitoring data; subsequently, the risk level is determined based on the numerical range of the current safety factor and the predicted safety factor; finally, the current stability state, future change trend, and early warning information of the slope are graphically displayed. Through the above process, a complete automated operation from data input to analysis and decision-making to result output can be achieved.

[0031] Compared to traditional slope analysis methods that rely solely on a single theoretical model or single monitoring data, this embodiment integrates multi-source monitoring data processing, physical mechanism calculation, machine learning prediction, and intelligent agent collaborative decision-making into a unified system framework. This allows it to better adapt to the dynamic stability assessment needs of slopes under complex environmental conditions such as rainfall infiltration, groundwater fluctuations, and long-term deformation, and improves the real-time performance, accuracy, and automation level of slope safety monitoring and early warning response. The conclusion of the paper also demonstrates that the system can automatically complete the analysis and display the prediction results and early warning information through a graphical interface, thereby lowering the barrier to entry for slope stability assessment.

[0032] II. Multi-source monitoring data processing and input construction process This embodiment describes the acquisition, processing, and input construction process of multi-source monitoring data. Raw data from different sources, formats, and time scales within the slope monitoring area are converted into a structured input dataset that can be used by the stability calculation module and the stability prediction module. Dynamic assessment of slope stability requires comprehensive consideration of slope geometric parameters, soil and rock physical and mechanical parameters, meteorological data, hydrological data, and historical monitoring data to fully reflect the changes in the slope's stability under complex environmental conditions.

[0033] In this embodiment, the acquired data includes slope geometric parameters, soil and rock physical and mechanical parameters, meteorological data, hydrological data, and historical monitoring data. Slope geometric parameters characterize the basic morphology and potential sliding body geometry of the slope, preferably including slope height, slope angle, slope surface morphology parameters, and related geometric information of the sliding surface. Soil and rock physical and mechanical parameters characterize the shear strength and self-weight characteristics of the slope materials, preferably including cohesion, internal friction angle, and unit weight. Meteorological data reflects the impact of external environmental changes on slope stability, preferably including rainfall, temperature, and humidity. Hydrological data characterizes the groundwater and internal water content of the slope, preferably including groundwater level and pore water pressure. Historical monitoring data reflects the deformation and evolution characteristics of the slope over time, preferably including displacement data, deformation data, and corresponding time series data. The data can be sourced from on-site monitoring equipment, engineering survey data, historical databases, remote sensing monitoring results, or manually entered information.

[0034] Furthermore, after acquiring the raw data, it is first categorized and organized according to data type, monitoring object identifier, and timestamp, and then transmitted to the data preprocessing module. Since different monitoring data differ in sampling frequency, units, precision, and data format, the raw data undergoes sequential data cleaning, missing value processing, outlier removal, standardization, time alignment, and feature fusion to form a unified input data structure. Through these processes, structural differences between multi-source heterogeneous data can be eliminated, improving the usability of the data in subsequent slope stability assessment and dynamic prediction.

[0035] In this embodiment, data cleaning is used to identify and remove duplicate, erroneous, and non-standard formatted records, and to uniformly convert the field names, unit expressions, and numerical formats of data from different sources. Missing value processing is used to fill in missing items. For static engineering parameters, this can be done by backfilling with survey data, inheriting historical valid records, or manually entering data. For time-series monitoring data, this can be done by interpolating adjacent time points, estimating historical trends, or using preset rules. Outlier removal is used to identify monitoring values ​​that significantly deviate from the normal engineering range or time-series variation patterns, and to remove or correct abnormal observations based on threshold rules, comparison with adjacent time points, or statistical distribution characteristics. Through these steps, the authenticity and completeness of the original monitoring data can be effectively improved.

[0036] Furthermore, the cleaned multi-source monitoring data undergoes standardization processing to eliminate differences in dimensions and orders of magnitude among different indicators. Preferably, the standardization processing employs range normalization, expressed as:

[0037] Where x is the original monitoring value, and x' is the standardized monitoring value.max x represents the maximum value of the corresponding monitoring indicator. min This represents the minimum value of the corresponding monitoring indicator. For data that conforms to specific statistical distribution characteristics, mean-standard deviation standardization can also be used to improve the stability of subsequent model training and prediction. Standardization should be performed before inputting multiple types of monitoring data into the model to improve model performance.

[0038] Since different monitoring data typically have different sampling time intervals, this embodiment also performs time alignment. Specifically, using a preset time step as a unified time reference, rainfall monitoring, groundwater level monitoring, pore water pressure monitoring, displacement monitoring, and other time-series data are resampled, timestamp matched, and windowed, so that various monitoring data form a corresponding relationship at a unified time or within a unified time window. Through time alignment, a synchronous input structure among multiple monitoring indicators can be established, improving the effectiveness of the joint characterization of slope status by different features.

[0039] Specifically, when displacement monitoring data is hourly, rainfall data is minutely, groundwater level data is hourly or daily, and soil and rock physical and mechanical parameters are daily or event-driven data, a unified time axis is established with a preset time step Δt. For high-frequency data, the mean, maximum, cumulative value, or rate of change is calculated within the corresponding window. For low-frequency data, nearest neighbor preservation, linear interpolation, or engineering event update methods are used to fill in the missing data. For missing time periods, linear interpolation, mean of the same period, or removal are selected based on the length of the missing data.

[0040] Specifically, the preset time step Δt is determined based on the response speed of the monitored object and the data sampling frequency, and can be 5 min, 10 min, 30 min, 1 h, 6 h, 12 h or 24 h; for rapidly changing data such as rainfall and pore water pressure, Δt is preferably 5 min to 1 h; for the comprehensive evaluation of conventional stable slopes, Δt is preferably 1 h to 24 h.

[0041] When the timestamps of the monitoring data are not completely consistent with the unified time axis, if the interval between adjacent valid data is less than 2Δt, linear interpolation or spline interpolation is used; if the interval between adjacent valid data is greater than or equal to 2Δt and less than the preset missing data threshold, the average of the same historical period, correlation correction of nearby monitoring points, or forward maintenance method is used; if the missing data duration exceeds the preset missing data threshold, the samples of that period are marked with low confidence or removed from the training samples. For cumulative indicators such as rainfall, window summation or effective rainfall attenuation model is used for aggregation; for state indicators such as groundwater level, pore water pressure, and displacement, the window mean, final value, maximum value, and rate of change are used to represent them.

[0042] After standardization and time alignment, feature fusion is performed on the multi-source monitoring data to construct a structured input dataset. This structured input dataset includes a set of basic parameters for stability calculation and a set of feature vectors for stability prediction. The set of basic parameters for stability calculation preferably includes slope geometric parameters, soil and rock physical and mechanical parameters, and hydrological parameters to meet the requirements of Bishop's method for calculating the safety factor. The set of feature vectors for stability prediction preferably includes slope geometric features, soil and rock shear strength features, meteorological features, hydrological features, and historical deformation and displacement features to meet the requirements of XGBoost model training and prediction. This input construction process enables a unified expression of static engineering attributes and dynamic monitoring information, providing a data foundation for subsequent slope stability calculations and future stability predictions.

[0043] Specifically, the feature fusion result is represented as Xt=[S,Dt], where S is a static feature sub-vector, including at least slope height, slope angle, slope morphology, cohesion, internal friction angle, unit weight, and potential sliding surface geometric parameters; Dt is a dynamic feature sub-vector, including at least current rainfall intensity, previous cumulative rainfall, groundwater level, pore water pressure, current displacement, displacement rate, displacement acceleration, cumulative displacement, and historical safety factor. By jointly expressing static engineering attributes and dynamic monitoring responses on the same time reference, the XGBoost model's ability to identify the coupling effects of rainfall infiltration, groundwater fluctuations, and slope deformation can be improved.

[0044] The fused feature vector can be represented as: X t =[H,β,G,c,φ,γ,P t ,P t-24h ,P t-72h Z t U t ,d t ,v t ,a t D t ,F s (t-1),F s (t-2),q t ] Where H is the slope height, β is the slope angle, G is the geometric parameter of the sliding surface, c is the cohesion, φ is the internal friction angle, γ is the unit weight of the rock and soil, and P t P represents the current rainfall intensity. t-24h P represents the cumulative rainfall over the past 24 hours. t-72h Z represents the cumulative rainfall over the past 72 hours. t For groundwater level, U t d is the pore water pressure. t v is the current displacement. t =(d t-d t-1 ) / Δt is the displacement rate, a t =(v t -v t-1 ) / Δt is the displacement acceleration, D t For cumulative displacement, F s (t-1), F s (t-2) represents the safety factor for the first two time points, q t This serves as a data quality identifier.

[0045] Specifically, static features are determined and standardized in terms of dimensions based on engineering survey or design data before model training, while dynamic features are updated on a rolling basis according to a unified time step Δt. For indicators with significant differences in dimensions, range normalization, Z-score standardization, or normalization according to engineering thresholds are used to ensure that each feature is at a comparable numerical scale before entering the XGBoost model.

[0046] Furthermore, after the input construction is completed, the processing results are transmitted to the stability calculation module and the stability prediction module respectively, and processing log information can be saved simultaneously. The log information includes the number of original data entries, records of missing value handling, records of outlier removal, standardized parameters, and time alignment rules, etc., for subsequent result tracking, manual review, and system maintenance. Through the multi-source monitoring data processing and input construction process in this embodiment, the present invention realizes the standardized transformation of original slope monitoring data into input data for stability assessment and dynamic prediction, thereby laying the foundation for the automated operation of the slope safety monitoring system.

[0047] III. Slope Stability Calculation and Dynamic Prediction Process This embodiment describes the process of slope stability calculation and dynamic prediction. Based on the acquisition and preprocessing of multi-source monitoring data, the safety factor of the slope under its current state is first calculated, and then the safety factor or stability state of the slope at future times is dynamically predicted, thus combining the current assessment and future judgment of slope stability. By combining the Bishop method and the XGBoost algorithm, the system can comprehensively evaluate slope stability and output the safety factor and corresponding warning level.

[0048] In this embodiment, the Bishop method is used to calculate the stability of the slope in its current state based on slope geometric parameters, soil and rock physical and mechanical parameters, and hydrological parameters. Specifically, firstly, a potential sliding body model is constructed based on the slope geometric parameters, and the potential sliding body is divided into multiple slices along the sliding direction; then, for each slice, its weight, water pressure, and shear strength-related parameters are calculated, and an overall moment balance relationship is established to iteratively solve for the slope safety factor. The Bishop method obtains the overall slope safety factor through slice analysis, moment balance, iterative solution, and sliding surface optimization.

[0049] After calculating the current safety factor, the XGBoost model is used to dynamically predict the safety factor or stability state of the slope at future times, based on preprocessed multi-source input features and historical monitoring data. The preferred input features include slope geometric parameters, soil and rock physical and mechanical parameters, meteorological data, hydrological data, and historical monitoring data. Historical monitoring data may include displacement, deformation, safety factor sequences, and other time-series features. The output can be a predicted safety factor value for future times, or a stability state category after threshold mapping. XGBoost can utilize more complex input features such as soil properties, meteorological data, and historical records for dynamic stability prediction and has strong generalization ability.

[0050] Furthermore, the two steps described above are interconnected: on the one hand, they provide the safety factor results for the current and historical moments, offering the stability prediction module a monitoring target, feature reference, and verification basis; on the other hand, they output the predicted safety factor values ​​or stability state prediction results for future moments, providing the early warning decision module with a forward-looking judgment basis. Through the coordinated operation of current calculations and future predictions, the system can identify the continuous deterioration trend of slope stability, improving the lead time for risk identification and the timeliness of early warning decisions.

[0051] Furthermore, the training samples are constructed using a sliding window approach. Let the window length be L, the sliding step size be s, and the prediction step size be k. The input to the t-th sample is the fused feature sequence from time t-L+1 to time t, and the output is the safety coefficient or stable state label at time t+k. The training set, validation set, and test set are divided chronologically, with an optimal ratio of 70%:15%:15% or adjusted according to the engineering sample size to avoid random partitioning that could lead to future information leakage. During model training, grid search, random search, or Bayesian optimization can be used to optimize the maximum depth, learning rate, subsampling ratio, column sampling ratio, regularization coefficient, and number of trees. The root mean square error, mean absolute error, coefficient of determination, and accuracy of warning level recognition are used as validation metrics.

[0052] Specifically, the window length L is set to 12h, 24h, 48h, 72h, 7d or 30d according to the slope response hysteresis characteristics, the sliding step s is Δt, 3Δt, 6Δt or 24h, and the prediction step k is 1h, 6h, 12h, 24h, 48h or 72h. For rainfall-sensitive shallow slopes, the window length is preferably 24h to 72h, and the prediction step is preferably 1h to 24h. For reservoir bank slopes or large open-pit mine slopes, the window length is preferably 7d to 30d, and the prediction step is preferably 24h to 72h.

[0053] When the sample size is sufficient and the distribution of warning levels is uneven, five-fold or ten-fold cross-validation can be used for initial screening of hyperparameters within the training set, but the validation set and test set are still retained as posterior evaluation data in chronological order. For engineering applications with strong requirements for time continuity, rolling start validation or forward chain validation is preferred to simulate the gradual update process of the model in real operation.

[0054] Preferably, the hyperparameter optimization range includes: maximum depth of 3 to 10, learning rate of 0.01 to 0.3, number of trees of 50 to 1000, subsampling ratio of 0.5 to 1.0, column sampling ratio of 0.5 to 1.0, minimum child node weight of 1 to 10, L1 regularization parameter α of 0 to 10, and L2 regularization parameter λ of 0.1 to 20. Validation metrics preferably include RMSE, MAE, R², accuracy of warning level, recall rate of red / orange warnings, and early warning time.

[0055] This embodiment organically combines Bishop's method stability calculation with XGBoost dynamic prediction, and further realizes the trend prediction of future state based on the current safety status analysis of the slope. It not only retains the engineering interpretability of the traditional physical model, but also enhances the dynamic analysis capability under complex nonlinear conditions, providing a reliable basis for subsequent early warning decisions and result output.

[0056] IV. Early Warning Decision-Making and Result Output Process This embodiment describes the early warning decision-making and result output process. Based on the current slope safety factor calculation results and future stability prediction results, the system automatically completes risk identification, early warning level determination, disposal suggestion generation, and result display, thereby realizing the conversion of slope stability assessment results into early warning decision information. The system can automatically generate stability, warning, alert, or alarm levels according to different values ​​of the slope safety factor, and display the prediction results and early warning information through a graphical interface.

[0057] In this embodiment, a collaborative working mechanism is constructed between a data intelligence agent, an analytical intelligence agent, and a decision-making intelligence agent. The data intelligence agent receives data and processing results output from the multi-source monitoring data acquisition module and the data preprocessing module, and transmits them to the analytical intelligence agent according to a preset format. The analytical intelligence agent receives the current safety factor result and the output future safety factor prediction value or stable state prediction result, and transmits the results to the decision-making intelligence agent. The decision-making intelligence agent completes risk identification, early warning level classification, disposal suggestion generation, and result distribution based on the current safety factor and prediction results. Through the above division of labor and collaboration among the intelligence agents, automatic linkage between data processing, analysis and judgment, and early warning output can be achieved. The collaborative working method between the data intelligence agent, the analytical intelligence agent, and the decision-making intelligence agent is described.

[0058] Furthermore, based on the slope safety factor Fs The warning level is determined based on the numerical range of F. In this embodiment, the preferred rule for classifying the warning level is as follows: when F... s When the slope strength is greater than 1.25, the slope is considered stable and at a warning level of no warning is required; when it is greater than 1.15... <F s When the slope strength is ≤ 1.25, it is determined to be in a warning state, at a yellow warning level; when it is 1.00... <F s When F ≤ 1.15, the slope is judged to be in a warning state, at the orange warning level; when F s When the slope value is ≤ 1.00, the slope is considered unstable and at risk of landslide, which is classified as a red alert. The above threshold values ​​for slope alert levels and their corresponding colors are listed above.

[0059] In this embodiment, not only can the current safety factor output by the stability calculation module be used for immediate judgment, but the predicted safety factor for future moments output by the stability prediction module can also be combined for forward-looking early warning. When the predicted safety factor falls into a lower level range or shows a continuous downward trend, the decision-making agent can trigger the corresponding level of early warning signal in advance, thereby issuing a risk warning before the slope reaches a significant instability state. By combining the current state assessment with future trend prediction, the lead time for early warning and the proactiveness of risk prevention can be improved. This system can adjust the slope safety assessment according to real-time changes and provide more accurate stability prediction and early warning.

[0060] Specifically, the decision-making agent will use the current safety factor F s (t), Predicted safety factor F s (t+k), minimum safety factor F within the prediction interval s,min Together with the rate of decrease in the safety factor, it serves as a warning trigger condition; when F s (t) is at a higher level but minF s When the safety factor drops to a lower level range, or when the rate of decrease of the safety factor exceeds a preset threshold for m consecutive prediction times, a lower level or corresponding level warning is triggered in advance, and the triggering factor is marked in the warning information as at least one of the following: current low safety factor, predicted low safety factor, continuous downward trend, abnormal rainfall accumulation, groundwater level rise or abnormal displacement rate.

[0061] Furthermore, after the warning level is determined, the decision-making agent can automatically generate corresponding risk warning information and response suggestions based on the warning level. Preferably, when the slope is at no warning level, a normal monitoring prompt is output; when the slope is at a yellow warning level, a prompt to strengthen patrols and pay close attention is output; when the slope is at an orange warning level, a prompt to increase monitoring frequency, initiate on-site inspections, and implement local protection measures is output; when the slope is at a red warning level, a high-level response suggestion is output, indicating a landslide risk and requiring immediate emergency response or personnel evacuation. By converting the safety factor results into engineering-executable response suggestions, the application value of the system in actual slope safety management can be enhanced.

[0062] In this embodiment, the current safety status, future stability prediction results, and warning levels of the slope are graphically displayed. Preferably, the visualization output includes a slope safety factor variation curve over time, a future safety factor prediction curve, risk warning level information, a historical monitoring data trend chart, and corresponding time series charts. Different warning levels can be distinguished using different colors, icons, or text labels to enhance users' intuitive ability to identify risk levels. The system can display prediction results and warning information through a graphical interface, thereby lowering the barrier to entry for users.

[0063] Furthermore, the current safety factor, predicted safety factor, stability status determination results, and early warning level information can be integrated into a unified monitoring interface to achieve integrated display of slope stability assessment results. For application scenarios requiring data retention or remote sharing, the visualization output module can also export the assessment results as chart files, log files, or early warning reports and send them to monitoring terminals, management platforms, or early warning receivers for further analysis and management decisions by engineering technicians. Through these result output methods, the system can achieve a complete closed loop from data analysis to risk transmission.

[0064] This embodiment introduces an intelligent agent collaboration mechanism and combines early warning judgment with graphical result output, enabling the invention not only to complete slope stability calculation and prediction, but also to automatically generate graded early warning and auxiliary decision-making results, thereby improving the automation, practicality and engineering promotion capability of the slope safety monitoring system.

[0065] Example: Application case of a slope in a transitional landform between low hills and terraces This embodiment uses a typical slope in a hilly-plateau transition zone as an example to illustrate the practical application of the method of the present invention. This slope is located in a transitional geomorphological unit between low hills and plateaus, with significant overall topographic relief. The upper part of the slope is relatively gentle, the middle part forms local slope break zones, and the lower part is relatively steep, exhibiting typical natural slope morphology characteristics. The strata and lithology of the study area are mainly granite and its weathering products. The shallow part of the slope contains residual layers, colluvial layers, and strongly weathered rock masses, with locally developed moderately weathered bedrock. Due to long-term weathering, the surface structure of the slope is relatively loose, and the rock-soil contact zone and weathering interface are quite obvious. Joints and fissures are well-developed in the area, and fissure water activity has a certain controlling effect on the internal water content of the slope and the evolution of local weak zones. Under conditions of continuous heavy rainfall, short-duration rainstorms, and groundwater level fluctuations, the slope is prone to shallow slippage, local deformation accumulation, and decreased stability. Therefore, it is suitable as the application object of the slope monitoring data processing and stability assessment system of the present invention. The above description is consistent with the common low mountain and hilly-plateau landforms, granite weathering crust, joint and fissure development, and weathering structural fissure groundwater conditions in the Yangjiang area.

[0066] In this embodiment, the data required for slope monitoring and assessment are first collected. The collected data includes slope geometric parameters, soil and rock physical and mechanical parameters, meteorological data, hydrological data, and historical monitoring data. Specifically, slope geometric parameters include slope height, slope angle, slope surface morphology parameters, and geometric information related to the potential sliding surface; soil and rock physical and mechanical parameters include natural unit weight, cohesion, and internal friction angle; meteorological data includes rainfall, temperature, and humidity; hydrological data includes groundwater level, pore water pressure, and fissure water response information; and historical monitoring data includes slope displacement, local deformation, fissure development records, and previous monitoring results. Considering that rainfall in this type of area is concentrated during the flood season, and frontal rain and tropical cyclone rain processes are common, the system focuses on the impact of continuous rainfall, short-duration heavy rainfall, and post-rain groundwater level rise on slope stability. Relevant regional data show that groundwater in this area is mostly weathered tectonic fissure phreatic water, and the groundwater level depth often varies with topography and season, with rainfall highly concentrated during the flood season.

[0067] In other embodiments, the present invention can also be applied to high-cut rock slopes, open-pit mine slopes, reservoir bank slopes, and highway cutting slopes. For high-cut rock slopes, key input parameters may include the orientation of structural planes, joint spacing, rock mass integrity, and anchorage status; for open-pit mine slopes, key input parameters may include bench height, platform width, blasting disturbance, and stripping progress; for reservoir bank slopes, key input parameters may include reservoir water level fluctuation rates, phreatic line location, and slope seepage response. Stability assessment for all of the above scenarios can be achieved through a unified multi-source data fusion, Bishop's method safety factor calculation, XGBoost trend prediction, and multi-agent collaborative early warning process.

[0068] Specifically, typical parameters for high-cut rock slopes include slope height of 20m to 120m, slope angle of 45° to 80°, structural plane dip angle of 20° to 75°, rock mass unit weight of 18kN / m³ to 28kN / m³, equivalent cohesion of 20kPa to 500kPa, and equivalent internal friction angle of 20° to 45°; typical parameters for open-pit mine slopes include total slope height of 50m to 300m, bench height of 10m to 30m, platform width of 5m to 20m, and slope angle of 35° to 65°. The typical parameters for reservoir bank slopes include: slope height 10m to 100m, slope angle 20° to 55°, daily reservoir water level variation 0.1m / d to 5m / d, groundwater level depth 0m to 30m, and pore water pressure 0kPa to 300kPa; and typical parameters for highway cutting slopes include: slope height 5m to 80m, slope angle 25° to 70°, soil cohesion 5kPa to 80kPa, and internal friction angle 10° to 40°.

[0069] The above parameter range is used to illustrate typical engineering scenarios to which the present invention can be adapted and does not constitute a limitation on the scope of protection. In actual application, the parameter range, early warning threshold and model input dimension can be adjusted according to the survey report, design documents, monitoring equipment range and local engineering experience.

[0070] Furthermore, the original monitoring data is cleaned, missing values ​​are processed, outliers are removed, standardized, and time-aligned, and an input feature set is constructed. In this embodiment, considering the large variation in the thickness of the shallow weathering zone and the spatial differences in the hydraulic conductivity of fissures in this type of slope, the system performs synchronous matching of rainfall processes, groundwater level changes, and deformation responses at a unified time scale during the input construction process to enhance the model's ability to identify the coupling relationship between rainfall infiltration, groundwater fluctuations, and slope deformation. For continuous monitoring data, daily or hourly scales are preferred for resampling and window aggregation; for sudden anomalies, the data is removed or corrected based on the changing trends and statistical distribution characteristics of adjacent time periods.

[0071] In this embodiment, the Bishop method is used to calculate the safety factor of the slope under its current state, based on the slope's geometric parameters, soil and rock physical and mechanical parameters, and hydrological parameters. Specifically, the potential sliding body is divided into multiple slices, and the weight, water pressure, and shear strength parameters of each slice are calculated. The overall safety factor of the slope is then solved iteratively. Simultaneously, the safety factor is calculated for multiple candidate sliding surfaces at different sliding center positions and radii, and the sliding surface corresponding to the minimum safety factor is selected as the most dangerous sliding surface. The minimum safety factor is used as the stability assessment result of the slope under its current working condition. For slopes with a clear transition zone between weathered granite residual layer and strongly weathered rock, the system can prioritize the potential sliding surface near the interface between the shallow overburden layer and the underlying weathered bedrock, which better reflects the shallow instability characteristics of this type of slope under heavy rainfall conditions. Publicly available data from similar areas indicate that under conditions of well-developed granite weathering layers, residual colluvial layers, and joints and fissures, rainfall and groundwater variations are important factors affecting slope stability.

[0072] like Figure 2 As shown, after obtaining the safety factor at the current moment, the XGBoost regression model is used to dynamically predict the safety factor of the slope at future moments. The training set consists of historical monitoring samples, and the input variables include slope geometric features, soil and rock mechanical features, rainfall features, hydrological features, and displacement and deformation sequence features; the output variable is the predicted value of the safety factor at future moments or the stable state category after threshold mapping. In this embodiment, the system focuses on strengthening the learning of factors such as rainfall duration, cumulative rainfall, previous water content, and groundwater level change rate to improve the model's ability to identify continuous rainfall and sudden rainstorm-induced stability decline processes. Since the rainfall process in this type of area is highly phased and localized, and the slope response during the flood season often has a certain lag, the system enhances the expression of temporal evolution characteristics by introducing historical time window features and sliding window statistics. For application areas with concentrated flood season rainfall, localized heavy rainfall, or significant tropical cyclone influence, the system expresses the lag effect of rainfall infiltration and slope response through historical time window features and sliding window statistics.

[0073] The displacement sequence features in the input variables of the XGBoost model include the current displacement d. t Displacement increment d over the past 24 hours t -d t-24h Average displacement rate over the past 24 hours, cumulative displacement increment over the past 72 hours, and displacement acceleration a. t The duration of continuous over-threshold displacement rates; rainfall sequence characteristics include current rainfall intensity, cumulative rainfall over the past 1 hour, 6 hours, 24 hours and 72 hours, previous effective rainfall and post-rain recovery time; hydrological sequence characteristics include current groundwater level, changes over the past 24 hours, current pore water pressure and pore water pressure change rate.

[0074] Furthermore, in this embodiment, the prediction result is based on the safety factor F. s The threshold is used for graded early warning. When F s When the slope strength is greater than 1.25, the slope is considered stable and at the level of no warning. When the slope strength is less than 1.15, the slope is considered stable and at the level of no warning. s When the slope strength is ≤1.25, the slope is considered to be in a warning state, at a yellow warning level; when 1.00 < F s When the slope strength is ≤1.15, it is determined to be in a warning state, at the orange warning level; when F s When the slope safety factor is ≤1.00, it is determined to be in an unstable state with a risk of landslide, which is classified as a red alert level. The early warning decision module automatically determines the slope risk level based on the interval between the current safety factor and the predicted safety factor, and outputs corresponding early warning information. For situations where the safety factor continues to decrease, the groundwater level continues to rise, or the local displacement rate increases significantly during continuous rainfall, the decision-making agent can improve the sensitivity of the early warning and trigger the corresponding level of early warning in advance, thereby enhancing the initiative in risk prevention and control.

[0075] The aforementioned safety factor thresholds can be set or calibrated according to the slope engineering safety evaluation specifications, industry technical guidelines, project design safety levels, and on-site management requirements. In the absence of specific thresholds, 1.00 can be used as the critical stability criterion, and 1.15 and 1.25 can be used as tiered thresholds between different risk buffer zones. For important traffic arteries, densely populated areas, or deformation-sensitive slopes, the yellow or orange warning trigger thresholds can be appropriately increased to enhance the conservatism of the warnings.

[0076] Furthermore, the threshold is a warning discrimination interval rather than a design safety factor that is fixedly applicable to all projects. When setting the threshold, it can be calibrated based on the slope type, project safety level, operating conditions, and instability consequences, combined with the provisions on landslide stability analysis and design safety factors in the "Code for Design of Landslide Prevention" (GB / T 38509-2020), as well as the slope engineering design, monitoring, and early warning requirements applicable to the project. When the safety reserve required by the engineering design is higher than the threshold in this embodiment, a more stringent control value is used as the basis for triggering the early warning.

[0077] F s ≤1.00 is considered a critical instability or limiting equilibrium state, and 1.00 is... <F s ≤1.15 is considered a warning interval approaching critical stability. <F s ≤1.25 is used as a warning range for a reduction in safety reserve, and F s >1.25 is considered a stable range with a relatively safe reserve; for slopes with a high design safety level or serious consequences of instability, the yellow warning threshold can be raised to 1.30 or the orange warning threshold can be raised to 1.20.

[0078] After model training, the predictive capability of the system was tested using validation set data. The results show that the predicted safety factor output by the system maintains a consistent overall trend with the actual safety factor, and the prediction error remains within a small range, effectively reflecting the stability changes of the slope under rainfall, post-rain recovery, and groundwater fluctuation conditions. From the time series results, the predicted curve shows high consistency with the actual curve in most periods, and the peak and trough positions are basically synchronized with the stage fluctuations, indicating that the model can effectively track the gradual changes in slope safety status and short-term disturbance responses. Figure 3 As shown in the scatter plot comparison results, the actual values ​​and predicted values ​​generally follow the trend... The distribution near the reference line indicates that the model's predictions are highly consistent with the actual results, demonstrating good fitting ability and generalization performance.

[0079] In this embodiment, the processes of multi-source monitoring data processing, stability calculation, and predictive analysis are uniformly scheduled. The data agent is responsible for receiving multi-source monitoring data and completing preprocessing; the analysis agent is responsible for using the Bishop method and XGBoost models to calculate the current safety factor and predict future stability; and the decision agent automatically determines the risk level and outputs early warning information based on the warning threshold. For slope responses triggered by heavy rainfall, the system can combine current conditions with future trends for joint judgment, thereby achieving automated integration from real-time monitoring and dynamic assessment to forward-looking early warning.

[0080] By applying this invention to weathered rock slopes in the transition zone of hilly terraces, a complete process has been achieved, from acquiring multi-source monitoring data, constructing input features, calculating the Bishop's method safety factor, dynamic prediction using XGBoost, to determining the warning level and displaying graphical results. The results show that this invention can adapt well to complex slope scenarios affected by weathering stratification, joint and fissure development, rainfall infiltration, and groundwater fluctuations. It exhibits small prediction errors on most samples, accurately depicts the temporal variation of the safety factor, and automatically adjusts the risk warning level based on stability changes, demonstrating good engineering applicability and promotional value.

[0081] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a slope monitoring data processing and stability assessment method based on multi-agent collaboration as described in any of the preceding claims.

[0082] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0083] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for slope monitoring data processing and stability assessment based on multi-agent collaboration, characterized in that, Includes the following steps: S1. Obtain multi-source monitoring data of the slope monitoring area, including slope geometric parameters, physical and mechanical parameters of soil and rock, meteorological data, hydrological data, and historical monitoring data; S2. Construct a collaborative working mechanism among data intelligence agents, analytical intelligence agents, and decision-making intelligence agents; S3. Performed by the data intelligence agent: Cleaning, missing value processing, outlier removal, standardization, time alignment, and feature fusion of multi-source monitoring data to form an input dataset for slope stability assessment; S4. Performed by the analytical agent: Based on the slope's geometric parameters and the physical and mechanical parameters of the soil and rock mass, the Bishop method is used to calculate the safety factor of the slope under its current state; based on the input dataset and historical monitoring data, a pre-trained XGBoost model is used to dynamically predict the stability state of the slope at future moments. S5. Executed by the decision-making intelligent agent: Based on the safety factor under the current state and the stability state at future moments, jointly determine the stability state of the slope and output the risk warning level and disposal suggestions; S6. Outputs a graphical display of the slope safety factor variation trend, stability prediction results, and risk warning level.

2. The method for slope monitoring data processing and stability assessment based on multi-agent collaboration as described in claim 1, characterized in that, In S2, time alignment uses a preset time step Δt as a unified time reference, and performs resampling, timestamp matching, interpolation completion and window aggregation on monitoring data with different sampling frequencies; feature fusion splices slope geometric parameters, soil and rock physical and mechanical parameters with rainfall, groundwater level, pore water pressure, displacement rate, displacement acceleration and cumulative displacement into a fused feature vector.

3. The method for slope monitoring data processing and stability assessment based on multi-agent collaboration according to claim 1, characterized in that, In S3, the Bishop method is used to calculate the safety factor of the slope under its current state, specifically including: The potential sliding body of the slope is divided into multiple slices. A moment balance relationship is established based on the geometric dimensions, weight, normal pressure, water pressure, and shear strength parameters of each slice, and the safety factor is iteratively solved. The weight of the i-th slice is... for: Water pressure of the i-th slice for: Slope safety factor satisfy: in, Where γ is the unit weight of the rock and soil mass, b i h is the width of the i-th slice. i Let γ be the height of the i-th slice. w z is the unit weight of water w,i Let c be the water level depth corresponding to the i-th slice. i Let φ be the cohesion of the i-th slice. i Let α be the internal friction angle of the i-th slice. i Let m be the inclination angle of the sliding surface of the i-th slice. i The correction coefficient is n, which is the total number of slices. The calculation is iterated until the difference between the safety factors obtained from two adjacent calculations is less than the preset error threshold. The minimum safety factor is obtained by traversing different sliding circle center positions and radii, which is used as the slope stability assessment result.

4. The method for slope monitoring data processing and stability assessment based on multi-agent collaboration as described in claim 1, characterized in that, In step S4, the XGBoost model is used to dynamically predict the stability state of the slope at future times. The objective function is as follows: in, (·) represents the loss function, y i For the true value, i (t-1) This is the predicted value after the (t-1)th iteration. For the t-th tree, the input feature x i The predicted output, is the regularization term for the t-th tree, and n is the number of training samples; the XGBoost model takes the preprocessed slope geometric parameters, soil and rock physical and mechanical parameters, meteorological data, hydrological data and historical monitoring data as input features, and the stable state label as output to obtain the stability state prediction result of the slope at future times.

5. The method for slope monitoring data processing and stability assessment based on multi-agent collaboration according to claim 1, characterized in that, S4 also includes the pre-training of the XGBoost model: The input dataset is divided into training, validation, and test sets in chronological order and fed into the XGBoost model for training and validation. The root mean square error, mean absolute error, coefficient of determination, or accuracy of warning level identification are used as evaluation metrics. When the validation error exceeds the preset error threshold, the slope condition changes abruptly, or the number of new samples reaches the preset number, the model is retrained or incrementally updated. The model update mechanism includes three methods: scheduled batch update, event-triggered update, and error-triggered update. The scheduled batch update is to retrain or fine-tune the model using new samples every week, month, or quarter. The event-triggered update is triggered after heavy rainfall, earthquake disturbance, excavation unloading, rapid rise and fall of reservoir water level, or replacement of monitoring equipment. The error-triggered update is triggered when the MAE or RMSE exceeds a preset error threshold for N consecutive prediction times, or when the number of misjudgments of the warning level exceeds a preset number.

6. The method for slope monitoring data processing and stability assessment based on multi-agent collaboration according to claim 1, characterized in that, In step S5, data flow and control flow interfaces are set up between the data agent, the analysis agent, and the decision agent. The data agent outputs the input dataset, including time alignment rules and fused feature vectors. The analysis agent outputs the current safety coefficient, historical safety coefficient sequence, future stable state, and prediction confidence information. The decision agent outputs the risk warning level, warning triggering reason, and handling suggestions, and feeds back the warning results and model errors to the data agent and the analysis agent to optimize the next round of data fusion strategy and model update strategy.

7. The method for slope monitoring data processing and stability assessment based on multi-agent collaboration according to claim 1, characterized in that, S3-S5 also include: Data flow closed loop: The data agent sends the input dataset to the analysis agent; the analysis agent calls the Bishop method and XGBoost models based on the fused feature vectors in the input dataset to generate the current safety factor, historical safety factor sequence, future stable state, and prediction error; the historical safety factor sequence and the corresponding fused feature vectors are jointly written into the training set cache for continuous updates of the XGBoost model; the analysis agent feeds back the safety factor results, error results, and feature importance to the data agent, which adjusts the next round of time alignment window, outlier identification threshold, missing value imputation method, or feature fusion weights accordingly; Control flow closed loop: When the decision agent recognizes that the warning level has increased, the prediction error has exceeded the threshold, the continuous rainfall process has started, or the monitoring data has changed abruptly, it sends control instructions to the data agent to increase the sampling frequency, shorten the time step Δt, or expand the historical window length, and sends control instructions to the analysis agent to recalculate the safety factor, retrain the XGBoost model, or increase the prediction frequency.

8. The method for slope monitoring data processing and stability assessment based on multi-agent collaboration according to claim 1, characterized in that, In step S5, the slope stability state is jointly determined based on the safety factor under the current state and the stability state at future times, specifically including: When the safety factor F in the current state s (t) has not yet reached the high-level warning threshold, but the stable state F at future moments s (t+k) When the value drops to a lower range or the rate of decline exceeds a preset threshold at multiple consecutive prediction times, the decision-making agent triggers a forward-looking warning of the corresponding level in advance; the warning levels are classified as follows: F s (t)>1.25 indicates a stable state with no warning; 1.15 <F s A yellow alert is issued if (t) ≤ 1.25, and 1.00 <F s (t)≤1.15 indicates an orange alert, F s A red alert is issued if (t) ≤ 1.

00.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a slope monitoring data processing and stability assessment method based on multi-agent collaboration as described in any one of claims 1-8.

Images