High and steep slope geological disaster monitoring method and system based on multi-source data fusion

Abstract

The invention relates to the technical field of slope disaster monitoring, in particular to a high and steep slope geological disaster monitoring method and system based on multi-source data fusion. The method comprises the following steps: tracking surface deformation through PS-InSAR time sequence analysis and atmospheric phase correction to obtain surface deformation characteristics; identifying the apparent diseases of the slope, and extracting spatial and temporal distribution characteristics of the apparent diseases of the slope; analyzing spatio-temporal evolution characteristics under geologic structure constraints by utilizing slope rock mass structure characteristics and the earth surface deformation characteristics; according to underground water level dynamic and potential sliding surface weakening features and the slope apparent disease spatial and temporal distribution features, extracting disaster-causing key factor features after cooperative correction; and combining the disaster-causing key factor characteristics, the true three-dimensional geological environment background characteristics and the spatio-temporal evolution characteristics to analyze the comprehensive early warning grade and the potential instability mode of the landslide. Scientific support is provided for high and steep slope geological disaster prevention and control, and engineering construction and operation safety can be guaranteed.

Description

Technical Field

[0001] This invention relates to the field of slope disaster monitoring technology, specifically to a method and system for monitoring geological disasters on steep slopes based on multi-source data fusion. Background Technology

[0002] Steep slopes are widely distributed in highways, railways, water conservancy projects and mining. Their landslide disasters are characterized by their suddenness, destructiveness and wide impact, which can easily cause casualties, property losses and damage to engineering facilities, seriously threatening the safety of engineering construction and operation.

[0003] Currently, existing landslide early warning methods for steep slopes have significant limitations: First, they mostly rely on single data sources for monitoring, failing to achieve deep integration of multi-source data and making it difficult to comprehensively reflect the landslide evolution patterns of slopes; second, 3D geological modeling often ignores details of rock mass structure and potential slip surfaces, and mechanical parameter assignments use fixed empirical values, which are out of touch with actual geological conditions, resulting in low accuracy of numerical simulations; third, time-series data denoising and feature mining lack geological logic constraints, easily leading to false anomalies and affecting the accuracy of early warnings; fourth, early warning models are mostly black-box-like, with uninterpretable factor weight allocations, and early warning levels are only simply divided without combining spatiotemporal evolution characteristics to clarify the potential spatiotemporal range of instability, resulting in poor applicability.

[0004] Therefore, there is an urgent need for a spatiotemporal accurate early warning method for landslides on steep slopes that can achieve multi-source data collaboration, geological logic adaptation, accurate and interpretable early warning, and feasible disposal, in order to overcome the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing methods and the needs of practical applications, and to solve the aforementioned problems, this invention provides a method for monitoring geological hazards on steep slopes based on multi-source data fusion, comprising the following steps: Surface deformation was tracked using PS-InSAR time-series analysis and atmospheric phase correction to obtain surface deformation characteristics; apparent slope defects were identified, and their spatiotemporal distribution characteristics were extracted; spatiotemporal evolution characteristics under geological structural constraints were analyzed using slope rock mass structure characteristics and the aforementioned surface deformation characteristics; based on groundwater level dynamics, potential slip surface weakening characteristics, and the aforementioned spatiotemporal distribution characteristics of apparent slope defects, key disaster-causing factors were extracted after collaborative correction; and combined with the key disaster-causing factors, true three-dimensional geological environment background characteristics, and the aforementioned spatiotemporal evolution characteristics, the comprehensive landslide early warning level and potential instability mode were analyzed.

[0006] Optionally, the step of tracking surface deformation through PS-InSAR time series analysis and atmospheric phase correction to obtain surface deformation characteristics includes the following steps: Atmospheric phase correction is performed on PS-InSAR data in a hierarchical and partitioned manner; geological laws of slope instability are introduced as a solution constraint, and the solved temporal deformation data is matched with the three-dimensional geological model mesh to generate a multi-dimensional spatiotemporal feature map.

[0007] Optionally, the identification of apparent slope defects and the extraction of the spatiotemporal distribution characteristics of apparent slope defects include the following steps: Prior information on slope aspect and slope gradient is introduced to perform geometric correction on tilted images; a multi-scale disease detection architecture is constructed, and based on the correction results, the spatiotemporal distribution characteristics of apparent diseases on the slope are extracted through the multi-scale disease detection architecture.

[0008] Optionally, the analysis of the spatiotemporal evolution characteristics under geological structural constraints using the slope rock mass structural characteristics and the surface deformation characteristics includes the following steps: A multi-scale feature spatial registration system for steep slopes is constructed; physical information constraints are embedded based on the slope rock mass structure characteristics and the surface deformation characteristics; deep spatiotemporal feature fusion is performed by combining the multi-scale feature spatial registration system and the physical information constraints to obtain spatiotemporal evolution characteristics under geological structure constraints.

[0009] Optionally, obtaining the structural characteristics of the slope rock mass includes the following steps: The point cloud under prior constraints of the slope is adaptively reduced in dimensionality to obtain the most relevant geometric features; natural rough features are obtained by adaptively filtering out noise points; and the slope rock mass structure features are obtained by combining the most relevant geometric features, the natural rough features and the structural surface parameter set.

[0010] Optionally, the step of extracting the collaboratively corrected key disaster-causing factor features based on the dynamic characteristics of groundwater level and potential slip surface weakening, and the spatiotemporal distribution characteristics of the apparent slope damage, includes the following steps: Constrained by the disaster-prone laws of steep slopes, a multi-source characteristic consistency verification system of deformation, disease, and groundwater level is constructed. Based on the dynamic characteristics of groundwater level and potential slip surface weakening and the spatiotemporal distribution characteristics of the apparent disease of the slope, false features are eliminated through conflict analysis and credibility assessment to obtain the key disaster-causing factors.

[0011] Optionally, analyzing the groundwater level dynamics and potential slip surface weakening characteristics includes the following steps: Based on prior knowledge of slope hydrogeology, a groundwater level time-series prediction model is constructed to obtain global time-series prediction data of groundwater level; a slip surface weakening feature inversion model is constructed, and key weakening parameters of potential slip surfaces are inverted through the slip surface weakening feature inversion model; combining the global time-series prediction data of groundwater level and the key weakening parameters, the dynamics of groundwater level and the weakening characteristics of potential slip surfaces are obtained.

[0012] Optionally, the step of combining the characteristics of the key disaster-causing factors, the true three-dimensional geological environment background characteristics, and the spatiotemporal evolution characteristics to analyze the comprehensive landslide early warning level and potential instability mode includes the following steps: Based on a true three-dimensional geomechanical model, key disaster-causing factors and spatiotemporal evolution characteristics are used as real-time input parameters to deduce the stress distribution and strain evolution law of the entire slope area; an instability mode identification rule base is established, and the comprehensive early warning level and potential instability mode of landslides are analyzed by combining the deduction results and the instability mode identification rule base.

[0013] Optionally, the true three-dimensional geological environment background features are extracted, including the following steps: A true three-dimensional geometric skeleton of the real geological morphology of a steep slope is constructed; based on the true three-dimensional geometric skeleton, structural surfaces and geological interfaces are embedded to obtain a structurally controlled three-dimensional model; mechanical parameters are adaptively assigned in the structurally controlled three-dimensional model, and the true three-dimensional geological environment background features are extracted through the assigned structurally controlled three-dimensional model.

[0014] This invention, based on multi-source data of steep slopes, first extracts rock mass structure, surface deformation, and apparent disease characteristics, inverts the underground disaster-prone environment, and constructs a three-dimensional geomechanical model with adaptive assignment of structural surfaces and mechanical parameters. Through multi-physics temporal data denoising, correlation mining, and spatiotemporal fusion, combined with physical constraints and deep networks, multi-source feature synergy is achieved. After multi-dimensional consistency verification and anomaly removal, key disaster-causing factors are extracted. Using XAI to correct factor weights, combined with numerical simulation and instability mode recognition, precise level and spatiotemporal early warnings are achieved. This realizes a closed-loop process from data acquisition, feature extraction, model construction to early warning output, which helps improve the accuracy, interpretability, and applicability of landslide early warnings, effectively predicts slope instability risks, provides scientific support for the prevention and control of geological disasters on steep slopes, further helps reduce casualties and property losses, and ensures the safety of engineering construction and operation.

[0015] Secondly, to efficiently execute the high-steep slope geological hazard monitoring method based on multi-source data fusion provided by this invention, this invention also provides a high-steep slope geological hazard monitoring system based on multi-source data fusion, comprising: an input device, an output device, a processor, and a memory, wherein the input device, output device, processor, and memory are interconnected, and the memory stores program instructions for use in the high-steep slope geological hazard monitoring method based on multi-source data fusion. The high-steep slope geological hazard monitoring system based on multi-source data fusion of this invention has a compact structure and stable performance, and can stably execute the high-steep slope geological hazard monitoring method based on multi-source data fusion provided by this invention, further enhancing the overall applicability and practical application capability of this invention. Attached Figure Description

[0016] Figure 1 A flowchart of a method for monitoring geological hazards on steep slopes based on multi-source data fusion, provided in an embodiment of the present invention; Figure 2 This is a framework diagram of a geological disaster monitoring system for steep slopes based on multi-source data fusion, provided for an embodiment of the present invention. Detailed Implementation

[0017] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0018] Throughout this specification, references to an embodiment, example, or illustration mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, phrases appearing in various places throughout the specification, such as "in one embodiment," "in an embodiment," "an example," or "an illustration," do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in any suitable combination and / or sub-combination in one or more embodiments or examples. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0019] Please see Figure 1 To address the aforementioned problems, this invention provides a method for monitoring geological hazards on steep slopes based on multi-source data fusion, such as... Figure 1 As shown, in one embodiment, the method includes the following steps: S1. By tracking surface deformation through PS-InSAR time series analysis and atmospheric phase correction, the surface deformation characteristics are obtained.

[0020] The method of tracking surface deformation through PS-InSAR time series analysis and atmospheric phase correction to obtain surface deformation characteristics includes the following steps: S11. Perform layered and partitioned atmospheric phase correction on PS-InSAR data.

[0021] Stable slope sections with exposed rock mass, no vegetation cover, and a slope fluctuation within 5° (such as slope shoulder platforms and intact rock face) are prioritized as candidate areas for PS points. Meanwhile, areas prone to generating false scattering signals, such as slope toe accumulation zones, weathered and fractured zones, and areas with dynamic vegetation cover (such as weed and shrub distribution areas), are excluded. This selection method ensures that the selected PS points not only guarantee the long-term stability of the scattering signal but also closely match the distribution characteristics of steep slope rock masses, effectively improving the continuity and accuracy of subsequent time-series deformation extraction.

[0022] Furthermore, based on the slope topographic data, multiple local areas are divided according to slope height (each 50 meters as a layer), slope aspect (divided into 8 directions), and elevation (divided into contour gradients). Then, for each local area, a local atmospheric phase model is constructed by combining contemporaneous meteorological data (such as atmospheric refractive index and temperature gradient) to accurately fit the atmospheric phase delay pattern of the area. Finally, the phase signal of the area is specifically corrected through the local model, effectively reducing the phase error caused by uneven atmospheric stratification and convection differences due to steep terrain, and achieving refined correction of the atmospheric phase of the entire slope area.

[0023] S12. Introduce the geological laws of slope instability as a solution constraint, match the solved temporal deformation data with the three-dimensional geological model mesh, and generate a multi-dimensional spatiotemporal feature map.

[0024] By introducing the geological laws of slope instability as a solution constraint, in the process of deformation sequence solution, combined with the rock mass structure characteristics of high and steep slopes and the dynamic change data of groundwater level, the periodic fluctuation signals related to rainfall and seasonal temperature changes (such as short-term deformation in the rainy season and slight deformation during winter freeze-thaw) are identified through time series data analysis, and then separated from the irreversible accelerated deformation signal.

[0025] Then, the separated irreversible deformation sequence is smoothed to remove instantaneous anomalies caused by sudden disturbances (such as construction disturbances or local blockages) and retain the true geological deformation trend. At the same time, by setting a deformation acceleration threshold, the area with a sudden increase in deformation rate and anomaly in acceleration is captured.

[0026] The calculated temporal deformation data is precisely matched with the 3D geological model mesh to generate multi-dimensional spatiotemporal feature maps. The specific outputs include: deformation rate map (intuitively presenting the deformation rate of different areas of the slope, distinguishing uniform deformation, accelerated deformation, and decelerated deformation areas), cumulative deformation map (showing the total deformation of each area in different time periods, clarifying the degree of deformation accumulation), deformation gradient map (reflecting the spatial change rate of deformation, identifying deformation abrupt change zones, and predicting potential instability boundaries), and deformation anomaly zone distribution map (marking areas where the deformation rate and acceleration exceed the normal range, clarifying key monitoring objects). All feature maps use a unified spatial grid coordinate system.

[0027] S2. Identify apparent slope defects and extract the spatiotemporal distribution characteristics of apparent slope defects.

[0028] The identification of apparent slope defects and the extraction of their spatiotemporal distribution characteristics include the following steps: S21. Introduce prior information on slope aspect and slope gradient to perform geometric correction on the tilted images.

[0029] By obtaining slope aspect and slope parameters of the shooting area through slope topography data, and combining UAV flight attitude data (altitude, pitch angle, heading angle) and monitoring camera installation parameters, a tilt angle correction model is constructed.

[0030] Secondly, pixel-level geometric correction is performed on images taken at an angle. The image viewing angle is adjusted according to the slope aspect to make the slope image more horizontal and to unify the geometric scale of the damage under different shooting angles. For example, for a slope image with a south-facing aspect and a tilt angle of 45°, perspective transformation is used to correct pixel stretching deformation, ensuring that cracks and slabs of the same size present a consistent pixel scale under different shooting angles.

[0031] Meanwhile, the correction process preserves the surface texture and disease details of the rock mass, avoiding blurring of disease characteristics due to correction, and effectively improving the stability and consistency of subsequent disease identification.

[0032] S22. Construct a multi-scale disease detection architecture, and extract the spatiotemporal distribution characteristics of apparent diseases on slopes based on the calibration results through the multi-scale disease detection architecture.

[0033] First, integrate disease samples from different high and steep slopes (covering different geological types such as hard rock, soft rock, and weathered rock) to establish a slope disease sample library with multiple types, shapes, and lighting conditions, including cracks, spalling, water seepage, and bulging. Focus on supplementing difficult-to-identify samples such as micro-cracks (width < 2 mm) and local small-scale spalling to solve the problem of sample imbalance.

[0034] Secondly, using small samples of slope-like scenes (such as slope disease images with similar geological conditions) as the basis for transfer learning, a domain adaptation algorithm is used to align the feature distribution of natural scenes and slope engineering scenes, thereby reducing the domain differences between the two. Specifically, by adjusting the convolution kernel parameters of the feature extraction network, the feature learning of slope rock texture and slope background is strengthened, while redundant features unrelated to the slope in the natural scene are weakened. This ensures that the transferred model can quickly adapt to the complex scenes of steep slopes, avoids the decrease in recognition accuracy caused by scene differences, and achieves efficient transfer learning under small sample conditions.

[0035] To address the issue of large differences in the scale of slope defects (from millimeter-level microcracks to meter-level spalling) and the easy omission of small-target defects, a multi-scale defect detection architecture is constructed based on the YOLO detection head. Specifically, the anchor frame size is dynamically adjusted according to the actual scale distribution of slope defects. Small-sized anchor frames are added to adapt to microcracks and localized small-scale spalling, while large-sized anchor frames are retained to adapt to large spalling and large-area bulging, thus achieving full coverage detection of multi-scale defects.

[0036] In addition, an attention mechanism is added to the detection head to enhance feature extraction of small disease areas, suppress interference from slope background (such as rock texture and shadow), and improve the sensitivity of small target disease identification. During the detection process, not only is the specific location of the disease (represented by both pixel coordinates and actual slope coordinates) and the disease type (cracks, spalling, seepage, bulging, etc.) output, but the actual size of the disease is also accurately calculated (such as crack length and width, spalling area and volume), providing quantitative data for subsequent disease evolution analysis.

[0037] Furthermore, disease detection results from different time periods (such as daily, weekly, and monthly) are integrated according to time series to track the development and changes of the same disease, calculate the disease expansion rate (such as the average daily expansion length of cracks and the average daily expansion area of ​​flaking areas), and identify the evolution trend of the disease (accelerated expansion, uniform expansion, and stable).

[0038] Secondly, the disease detection results are accurately matched with the three-dimensional slope model according to the spatial location, dividing the disease-dense area (such as a large number of cracks and spalling blocks appearing in a certain slope section) and the sparse area, and analyzing the spatial distribution pattern of the disease and its relationship with the rock mass structure (such as the disease-dense area often corresponding to the rock mass fracture zone and the area with dominant joint development).

[0039] Finally, the spatial linkages between different types of diseases are explored (such as the tendency of crack expansion areas to be accompanied by water seepage and spalling), forming a spatiotemporal feature set of apparent diseases that includes temporal evolution, spatial distribution, and type linkages.

[0040] S3. Analyze the spatiotemporal evolution characteristics under geological structural constraints by utilizing the structural characteristics of the slope rock mass and the aforementioned surface deformation characteristics.

[0041] The method of analyzing the spatiotemporal evolution characteristics under geological structural constraints by utilizing the structural characteristics of slope rock mass and the surface deformation characteristics includes the following steps: S311. Construct a multi-scale feature spatial registration system for steep slopes.

[0042] First, determine the scale differences and type classification of multi-source features: point features (such as PS point deformation data, rock mass structure point cloud), surface features (such as structural surfaces, potential slip surfaces, and disease distribution surfaces), and volume features (such as three-dimensional geological model rock mass zoning and groundwater level distribution). Different types of features have significant differences in spatial resolution and coordinate reference, and layered registration is required.

[0043] Secondly, using the three-dimensional geological model mesh as a unified spatial reference, the coordinate systems of all features are first uniformly calibrated to the model mesh coordinate system to eliminate coordinate deviations from different data sources. Then, a differentiated registration strategy is adopted for features of different scales: point features are mapped to model mesh nodes through nearest neighbor interpolation to ensure that fine features such as deformation are not lost; surface features are precisely fitted to the mesh surface through geometric projection to preserve the boundary contours of structural and diseased surfaces; volume features are processed through voxelization to correspond one-to-one with the model mesh voxels, achieving precise matching of volume features such as groundwater level and rock mechanics parameters.

[0044] Meanwhile, prior information about rock mass structure (such as the distribution of dominant joints and the boundaries of rock mass zones) is introduced as registration constraints to ensure that the spatial position of each feature after registration is consistent with the actual geological structure of the slope, avoiding the problems of feature misalignment and spatial contradictions. This achieves accurate unification of point, surface and volume multi-scale features in the three-dimensional geological model mesh, laying a spatial consistency foundation for subsequent physical information constraint embedding and deep fusion.

[0045] S312. Based on the slope rock mass structure characteristics and the surface deformation characteristics, embed physical information constraints.

[0046] First, physical information constraints are set: one is the anisotropic constraint of the rock mass structure, which, combined with the anisotropic characteristics of the rock mass, constrains the deformation and stress distribution in different directions. For example, the deformation in the direction of fracture development should be greater than that in the direction of intact rock mass to avoid unreasonable fusion results of isotropic properties. The second is the constraint of stress diffusion law, which, based on the slope stress transfer mechanism, constrains the evolution of the stress field distribution to ensure that stress diffuses reasonably from the top of the slope to the bottom of the slope and from intact rock mass to broken rock mass, and prohibits situations that violate mechanical laws, such as stress abrupt changes and reverse diffusion. The third is the constraint of deformation rationality, which, combined with the surface deformation characteristics and underground disaster-prone characteristics, constrains the evolution trend of the deformation field. For example, the deformation in the potential slip surface area should be greater than that in the surrounding area, and the deformation direction should be consistent with the slip surface dip to avoid fusion results where the deformation direction contradicts the geological structure and the deformation is abnormally out of proportion.

[0047] At the same time, a constraint verification mechanism is constructed to verify each set of fused data in real time during the fusion process. If a result that violates the above physical constraints occurs, a correction mechanism is immediately triggered to adjust the fusion parameters and ensure that the final fusion result conforms to the actual geomechanical evolution logic of the slope.

[0048] S313. Combine the multi-scale feature space registration system and the physical information constraints to perform deep spatiotemporal feature fusion to obtain spatiotemporal evolution features under geological structure constraints.

[0049] To address the differences in sampling frequency among different types of time series data (surface deformation at the minute level, rainfall at the hour level, and disease at the day level), all multiphysics time series data are first uniformly resampled and normalized into continuous time series with the same time interval, thus eliminating aggregation bias caused by frequency differences.

[0050] Secondly, based on the disaster-prone patterns of slopes, the length of time-series segments is dynamically adjusted: shorter segment lengths are used for key periods such as periods of concentrated rainfall and accelerated deformation to retain more detailed features, while longer segment lengths are used for stable periods to reduce redundant calculations.

[0051] Then, by normalization, the differences in the dimensions of different physical quantities (such as deformation at the millimeter level and groundwater level at the meter level) are eliminated, and the continuous time series data is mapped into a finite number of symbol fragments. The trend change characteristics of the data are highlighted by symbolization transformation, while effectively reducing the interference of high-frequency noise (such as small disturbances of monitoring equipment and instantaneous meteorological fluctuations), thus achieving preliminary denoising and feature simplification of multi-physics time series data.

[0052] Furthermore, considering the geological evolution characteristics of steep slopes, the core differences between the two types of signals were identified: the geological instability trend term is an irreversible, gradual change (such as a steady increase in deformation rate due to the continuous weakening of the potential slip surface), while the fluctuation term is a reversible, periodic change (such as short-term deformation caused by rainfall and freeze-thaw fluctuations caused by seasonal temperature changes). A dual separation method of trend constraint fitting and fluctuation threshold screening was then adopted. First, the trend term was initially extracted through adaptive polynomial fitting. Then, combined with the meteorological data (rainfall, temperature) and environmental monitoring data of the same period, a reasonable threshold range for the fluctuation term was set. Changes that exceed the threshold and conform to the geological instability law were included in the trend term, while changes that exhibit periodic characteristics within the threshold range were classified as fluctuation terms.

[0053] Meanwhile, rock mass structural characteristics (such as the distribution of dominant joints and rock mass integrity) are introduced as auxiliary constraints. If the temporal changes over a certain period of time are highly consistent with the distribution of weak areas in the rock mass structure, it is determined as a trend term, ensuring that the separation results conform to the actual disaster-causing logic of the slope. Finally, irreversible trend signals related to geological instability are retained, while irrelevant fluctuations such as meteorological and environmental factors are eliminated.

[0054] Constrained by rock mass structure characteristics, underground disaster-prone characteristics, and apparent disease characteristics, and based on the potential correlation logic between multi-source data (such as rainfall → rising groundwater level → slip surface weakening → accelerated slope deformation → disease expansion), this serves as the basis for guiding the mining of correlation rules, avoiding the discovery of false correlations without geological significance. Furthermore, considering the temporal characteristics of multi-source data, the focus is on mining three core rules: lagging correlation (such as the groundwater level reaching its peak 12-24 hours after rainfall, or the slope deformation rate suddenly increasing 3-5 days after the groundwater level rises), synchronous correlation (such as the synchronous distribution of severely weakened slip surface areas and densely affected disease areas), and accelerated correlation (such as the deformation rate accelerating synchronously when the disease expansion rate accelerates).

[0055] Meanwhile, by using correlation strength quantification indicators (such as support and confidence), strong correlation rules that are highly correlated with slope instability height are screened out, weak and false correlation rules are eliminated, and the response relationship and influence weight between various multi-physics data are clarified, providing clear correlation basis for subsequent data denoising reconstruction and spatiotemporal fusion.

[0056] Based on the multi-source data association relationships obtained through association rule mining, collaborative denoising is performed on the initially denoised time-series data. By utilizing the mutual corroboration between data from different physical fields, isolated noise points from a single data source are eliminated (e.g., abnormal deformation data at a certain moment, if not corresponding to synchronous changes in groundwater level or disease, are judged as noise and removed). Secondly, the denoised time-series data undergoes structured reconstruction, incorporating trend terms and association rules into the reconstruction process to restore the true evolutionary patterns of the multi-physical field data, forming a regular and continuous pure time-series sequence. Finally, multi-physical field trend characteristics are obtained: first, pure time-series data (denoised deformation, groundwater level, disease, and rainfall time-series sequences, which can be directly used for subsequent analysis); second, evolutionary trend characteristics (the changing trends, rates of change, and acceleration / deceleration characteristics of each physical field data, such as the daily average change in deformation rate and the rising and falling trends of groundwater level); and third, feature association strength (the support and confidence of association rules between various multi-physical field data, clarifying the degree of influence between different factors).

[0057] Furthermore, a spatiotemporal fusion network for slopes is constructed, which incorporates the temporal patterns of slope disaster formation (such as the lag response relationship between rainfall, deformation, and disease) and spatial distribution patterns (such as the spatial correlation between disease and rock fracture zones). The network's convolution kernel and temporal attention mechanism are optimized to enhance the capture of key periods and key regional characteristics of slope instability.

[0058] Secondly, the spatially registered multi-source features (rock mass structure features, surface temporal deformation features, multi-physics field trend features, and three-dimensional geological model features) are used as network inputs. The spatial dimension incorporates the geometric and structural constraints of the three-dimensional geological model, and the temporal dimension incorporates the temporal evolution trend of the multi-physics field. Through the network's spatiotemporal attention mechanism, the network automatically mines the temporal lag correlations and spatial complementary correlations of different features (such as the spatial correspondence between rock mass fracturing and deformation acceleration in a certain area, and the temporal correlation between groundwater level and deformation after rainfall).

[0059] During the fusion process, real-time guidance is provided by combining physical information constraints to ensure that the fusion process does not deviate from the logic of geomechanics. The final output is a high-resolution spatiotemporal correlation feature map. This feature map not only integrates the core information of each source feature, but also clearly presents the linkage relationship of each feature in the spatiotemporal dimension, such as the spatiotemporal coupling distribution of deformation field and stress field, and the spatial correspondence between deterioration field and disease distribution.

[0060] For deformation fields, the spatiotemporal evolution characteristics of deformation rate, cumulative deformation, and deformation acceleration in different regions and time periods are extracted to determine the spatial distribution differences of deformation (e.g., significant deformation at the toe of the slope and in the fractured rock mass area, and weak deformation in the intact rock mass area) and the temporal evolution trend (e.g., uniform deformation, accelerated deformation, and stable deformation). The spatiotemporal characteristics of sudden increases in deformation rate and expansion of deformation range are captured to predict the development direction of potential unstable areas.

[0061] Secondly, for the stress field, the spatiotemporal characteristics of stress distribution, stress concentration area and stress evolution rate are extracted to clarify the spatial relationship between stress concentration and rock mass structure and potential slip surface (such as stress concentration often occurs at the top of potential slip surface and at the intersection of rock mass fractures), as well as the evolution law of stress over time (such as the expansion of stress concentration area as groundwater level rises and deformation accelerates).

[0062] Finally, for the deterioration field, the spatiotemporal evolution characteristics of rock mass deterioration degree, deterioration range, and deterioration rate are extracted. Combined with underground disaster-inducing characteristics and apparent disease characteristics, the spatiotemporal linkage relationship between the deterioration field and groundwater level and disease expansion is clarified (e.g., the rock mass deterioration rate accelerates in areas where groundwater level rises, and areas with dense disease correspond to areas with higher deterioration degree).

[0063] By integrating the spatiotemporal evolution characteristics of the three major fields, a synergistic evolution law of deformation-stress-deterioration is formed, clearly presenting the spatiotemporal evolution process of the slope from stability to potential instability, providing accurate evolutionary basis for subsequent anomaly identification and early warning level determination.

[0064] In this embodiment, obtaining the structural characteristics of the slope rock mass includes the following steps: S321. Adaptive dimensionality reduction of point cloud under slope prior constraints to obtain the most relevant geometric features.

[0065] Using the actual slope parameters of steep slopes as the core constraints, the slope aspect and slope grade of the three-dimensional laser scanning point cloud are first accurately processed by the slope topographic data. The slope aspect is divided into 8 directions (east, south, west, north, northeast, southeast, northwest, and southwest), and the slope is divided into three levels according to the common standards for steep slopes: gentle slope (<30°), steep slope (30°-60°), and very steep slope (>60°).

[0066] Based on this, the dimensionality reduction dimension is dynamically adjusted to address the differences in point cloud density across different slope sections (e.g., lower point cloud density at the top and shoulder of the slope, and higher point cloud density at the foot and middle of the slope): the dimensionality reduction dimension is appropriately increased in areas with high point cloud density to retain more of the fine structural features of the rock mass; the dimensionality reduction dimension is appropriately decreased in areas with low point cloud density to avoid excessive retention of noise information.

[0067] This method accurately preserves the geometric features related to the rock mass structure (joints, fissures, bedding planes), effectively eliminates redundant information caused by topographic undulations, and ultimately achieves structure-sensitive dimensionality reduction that fits the high and steep slope scenario.

[0068] S322. Obtain natural roughness features by adaptively filtering out noise points, and combine the most relevant geometric features, the natural roughness features and the structural surface parameter set to obtain the slope rock mass structural features.

[0069] First, calculate the local slope angle change rate within a certain range around each point cloud. If the local slope angle change rate of a point cloud is much higher than the average level of the surrounding area, and the consistency between its normal vector and the overall normal vector of the slope segment is lower than the set threshold (this threshold is dynamically adjusted according to the slope rock mass type; the threshold is slightly higher for hard rock mass and slightly lower for soft rock mass), then it is determined to be a noise point.

[0070] Meanwhile, a special distinction is made between the natural rough surface of the slope and scanning noise and vegetation points: by the difference in point cloud reflection intensity (the reflection intensity of vegetation points is lower than that of rock mass points, and the reflection intensity of scanning noise is chaotic) and spatial distribution characteristics (vegetation points are mostly discretely distributed, while natural rough surfaces are continuously distributed), the noise judgment logic is further optimized to ensure that while filtering out scanning noise and vegetation interference points, the natural rough features of the real rock mass surface are fully preserved, and the loss of rock mass structure information due to noise filtering is avoided.

[0071] The direction of the free face of the slope and the direction of the stress dominance of the rock mass are introduced as clustering guiding conditions. The specific direction of the free face is determined by the slope topographic data, and the direction of the stress dominance of the rock mass is determined by the on-site rock mass stress test data (the stress dominance direction of high and steep slopes is mostly along the slope direction). These two directions are used as the initial guiding vectors for clustering.

[0072] During the regional growth process, point clouds with an angle between the normal vector and the guiding vector within a reasonable range (set to 15° to 30° according to the complexity of the slope rock mass structure) are used as seed points to gradually grow and cluster, automatically identifying different types of rock mass structural surfaces such as joints, fissures, and bedding planes.

[0073] After clustering, the key attitude parameters of each structural surface are solved through spatial geometric calculations. The strike is represented by the azimuth (0° to 360°), the dip is the direction perpendicular to the strike, and the dip angle is the angle between the structural surface and the horizontal plane. At the same time, the spacing of the structural surfaces (the average distance between adjacent structural surfaces of the same type) and the extension length (the continuous extension distance of the structural surface on the slope) are accurately calculated to form a complete set of structural surface parameters.

[0074] Based on this, the rock mass integrity coefficient is calculated by the spacing and extension length of structural planes. Larger spacing and shorter extension lengths indicate better rock mass integrity; conversely, smaller spacing and shorter extension lengths indicate poorer rock mass integrity. Secondly, the anisotropy characteristics of the rock mass are analyzed by combining the strike and dip of the structural planes with the slope aspect. If the dominant structural plane strike is parallel to the slope aspect, the rock mass exhibits significant anisotropy and is prone to instability. Finally, by selecting dominant structural planes with dip angles between 30° and 60° (a common dip angle range for unstable structural planes on steep slopes), long extension lengths, and small spacings, the potential dominant instability structures of the slope are determined, thus obtaining the rock mass structural characteristics, including rock mass integrity, anisotropy, and dominant instability structures.

[0075] S4. Based on the dynamic characteristics of groundwater level and potential slip surface weakening, and the spatiotemporal distribution characteristics of the apparent slope damage, extract the characteristics of the key disaster-causing factors after collaborative correction.

[0076] In this embodiment, the step of extracting the collaboratively corrected key disaster-causing factor features based on the dynamic characteristics of groundwater level and potential slip surface weakening, and the spatiotemporal distribution characteristics of the apparent slope damage, includes the following steps: S411. Based on the disaster-prone laws of steep slopes, a multi-source characteristic consistency verification system for deformation, disease, and groundwater level is constructed.

[0077] Rising groundwater levels can weaken potential slip surfaces, leading to accelerated slope deformation and ultimately exacerbating apparent damage (crack expansion, increased rockfall), exhibiting a co-evolutionary trend.

[0078] Based on this, the following trends were extracted: surface deformation evolution trend (deformation rate, cumulative deformation change), apparent disease evolution trend (disease expansion rate, disease range change), and groundwater level evolution trend (water level rise and fall rate, peak value change). The three types of trend data were then uniformly calibrated to the same time scale and spatial coordinates.

[0079] Secondly, a quantitative analysis of trend similarity is adopted to calculate the correlation coefficient of the evolution trend of any two types of features. A reasonable similarity threshold is set (dynamically adjusted according to the slope rock mass type, with a slightly higher threshold for hard rock mass and a slightly lower threshold for soft rock mass). If the similarity of any pair of the three types of features is higher than the threshold, it is judged as a trend match; if the similarity of any two types of features is lower than the threshold and does not conform to the disaster-pregnancy response logic of groundwater level-deformation-disease, it is judged as a trend mismatch and marked as a suspected anomaly, providing a clear basis for subsequent conflict identification.

[0080] S412. Based on the dynamic characteristics of groundwater level and potential slip surface weakening, and the spatiotemporal distribution characteristics of the apparent slope defects, false features are eliminated through conflict analysis and credibility assessment to obtain the key disaster-causing factors.

[0081] First, identify the conflict types among various features, which are mainly divided into three categories: one is a single feature abnormal conflict (such as a sudden increase in deformation data at a certain moment, but no synchronous change in groundwater level and disease); the second is a conflict between two types of features in opposite directions (such as groundwater level rising and disease expanding, but no significant change in deformation); and the third is a chaotic conflict among three types of features (such as groundwater level falling, deformation accelerating, and disease stabilizing).

[0082] Secondly, a credibility assessment index system is established, with the stability of features, multi-source corroboration, and geological rationality as the core assessment dimensions: isolated features from a single data source have the lowest credibility, features partially corroborated by multiple sources have medium credibility, and features fully corroborated by multiple sources and conform to the geological disaster-causing laws have the highest credibility.

[0083] By using quantitative scoring (confidence score 0-100), features with a confidence score below 40 are identified as low-confidence false anomalies. The feature data corresponding to these anomalies (such as isolated deformation anomalies and disease data that contradicts geological logic) are highlighted and directly removed. Features with a confidence score of 40-70 are identified as suspicious anomalies and are temporarily retained for subsequent collaborative correction. Features with a confidence score above 70 are identified as high-confidence valid features and serve as the core basis for subsequent correction and factor extraction, avoiding early warning deviations due to the accidental deletion of valid features.

[0084] For suspicious anomalies (confidence level 40-70 points), the anomaly characteristics are corrected by referring to high-confidence features in the same spatial location and time period, and by combining the response lag relationship between groundwater level, deformation, and disease. For example, if a small anomaly appears in the deformation data of a certain area, and the groundwater level has risen but the disease has not expanded synchronously (consistent with the lag response law), the deformation data is corrected to a reasonable range based on the groundwater level change trend. If the disease data is abnormal, and there is no corresponding change in deformation or groundwater level, the abnormal disease data is corrected by combining the distribution pattern of high-confidence disease features in the surrounding area.

[0085] Furthermore, to fill data gaps after removing low-confidence anomalies, interpolation based on the spatiotemporal correlation patterns of high-confidence features is employed to ensure data continuity and integrity. For example, if groundwater level data for a certain period is missing due to equipment malfunction, the missing data can be filled by interpolation based on the groundwater level change trends of previous and subsequent periods, combined with concurrent rainfall and deformation data. The entire correction process is constrained by slope geomechanical laws, ensuring that the corrected feature data are both mutually synergistic and consistent with the actual disaster-causing evolution logic of slopes.

[0086] Furthermore, the screening criteria for key disaster-causing factors were determined as follows: First, multi-source corroboration, which must be simultaneously corroborated by three types of characteristics: deformation, disease, and groundwater level, and conform to the disaster-causing logic of groundwater level-slip surface weakening-deformation-disease; Second, correlation, which is highly correlated with slope instability and can directly reflect the evolution trend of slope instability (such as deformation rate acceleration, peak groundwater level, slip surface softening coefficient, and disease expansion rate); Third, stability, which shows a continuous changing trend in the time-series evolution without frequent fluctuations, and can be used as the core basis for early warning.

[0087] Next, through correlation strength analysis, the correlation between each candidate factor and slope instability was quantified, and core factors with high correlation strength and good stability were screened out, mainly including: groundwater level rise and fall rate and peak value, potential slip surface softening coefficient and saturation, surface deformation rate and acceleration, apparent disease expansion rate and density, and rock mass integrity deterioration rate.

[0088] Finally, the extracted key disaster-causing factors are structured and organized to clarify the quantitative indicators, spatiotemporal distribution characteristics and evolution trends of each factor, forming a standardized set of key disaster-causing factors.

[0089] In this embodiment, the analysis of the groundwater level dynamics and potential slip surface weakening characteristics includes the following steps: S421. Based on prior knowledge of slope hydrogeology, a groundwater level time series prediction model is constructed to obtain groundwater level time series prediction data for the entire region.

[0090] First, the timestamp benchmark for all time-series data is unified. Meteorological rainfall data (hourly), surface deformation data (minutely), apparent disease monitoring data (daily), and groundwater level measurement data (hourly) are all uniformly calibrated to millisecond-level timestamps to eliminate time deviations caused by differences in sampling frequencies of different monitoring equipment.

[0091] At the same time, based on the spatial grid of the three-dimensional geological model, the spatial coordinates of all data are precisely registered to ensure that each set of time series data can correspond to the specific spatial location of the slope (such as a certain slope section or a certain elevation range), thus achieving the unification of the spatiotemporal dual benchmarks.

[0092] Based on this, the temporal response correlation between rainfall, seepage, deformation, and disease was established, and the lag effect between various data was clarified (such as the lag time of groundwater level rise after rainfall and the lag time of slope deformation acceleration after water level rise). By calculating the correlation coefficient of each data sequence, effective data pairs with strong correlation were screened out, and redundant data without correlation were eliminated, laying a precise and collaborative multi-source data foundation for subsequent groundwater level prediction and slip surface inversion.

[0093] Furthermore, using historical groundwater level data as labels and concurrent meteorological rainfall data (rainfall amount, rainfall duration, rainfall intensity), surface deformation data (deformation rate, cumulative deformation), and apparent disease data (disease expansion rate, disease type) as input features, we focus on supplementing sample data during periods of concentrated rainfall and periods of accelerated disease expansion to address the issues of scarce and unevenly distributed samples.

[0094] Secondly, the deep learning models (such as LSTM and GRU) are improved for slope scenarios by embedding prior knowledge of slope hydrogeology (such as rock permeability coefficient, aquifer distribution, and slope seepage path), optimizing the model's loss function, and focusing on improving the prediction accuracy during periods of sudden water level changes (such as after heavy rain) to avoid excessive prediction deviations of general models under complex seepage conditions on slopes.

[0095] During the training process, cross-validation was used, and the model parameters were continuously adjusted in combination with on-site measured data to ensure that the model could accurately capture the dynamic response of groundwater level with rainfall, deformation and disease, and finally realize the dynamic extrapolation of groundwater level in different spatial locations and time periods, and achieve full-domain time-series prediction of groundwater level.

[0096] S422. Construct a slip surface weakening feature inversion model, and invert the key weakening parameters of the potential slip surface through the slip surface weakening feature inversion model.

[0097] Using groundwater level change data (water level rise and fall, duration), surface deformation acceleration data (deformation rate increase, acceleration), and apparent disease expansion data (crack length, spalling area growth) as core inputs, and combining them with rock mass structural characteristics (distribution of dominant joints, rock mass integrity), a slip surface weakening characteristic inversion model is constructed.

[0098] Secondly, the key weakening parameters of the potential slip surface are inverted through the model: First, the slip surface saturation, based on changes in groundwater level and slope seepage distribution, the degree of water saturation in different areas of the slip surface is inverted. The higher the saturation, the lower the slip surface strength. Second, the slip surface softening degree, combined with deformation rate and disease propagation rate, the degree of strength softening of the slip surface rock mass caused by water erosion and stress is inverted. Third, the slip surface strength deterioration trend, by tracking the changes in slip surface saturation and softening degree through time series data, the deterioration rate and development direction of slip surface strength are predicted.

[0099] Meanwhile, the geological laws of slope instability are introduced as constraints to ensure that the inversion results conform to the physical logic of slip surface evolution, avoid inversion results that contradict the rock mass structure and seepage laws, and realize the full-domain, dynamic and quantitative inversion of potential slip surface weakening characteristics.

[0100] S423. By combining the global time-series prediction data of groundwater level and the key weakening parameters, the dynamic characteristics of groundwater level and potential slip surface weakening are obtained.

[0101] The dynamic characteristics of groundwater level and the weakening characteristics of potential slip surfaces obtained from the inversion are structured and organized to form a standardized and reusable set of groundwater disaster-inducing features. The specific output includes: First, dynamic characteristics of groundwater level, covering the time-series change curves of water level in different spatial regions, peak water level, rate of rise and fall, and response lag time with rainfall, clarifying the spatiotemporal distribution pattern of groundwater level; Second, location characteristics of potential slip surfaces, accurately marking the spatial coordinates, extension range, dip angle, and thickness of potential slip surfaces, and realizing spatial visualization of slip surfaces by combining with a three-dimensional geological model; Third, characteristics of slip surface weakening degree, quantitatively outputting the saturation, softening coefficient, and strength degradation degree of different areas of slip surfaces, and classifying the weakening level (mild, moderate, severe); Fourth, characteristics of slip surface development speed, calculating the daily average rate of change of slip surface weakening degree and cumulative weakening amount, and predicting the evolution trend of slip surfaces (stable, slow weakening, rapid weakening).

[0102] S5. Combining the characteristics of the key disaster-causing factors, the true three-dimensional geological environment background characteristics, and the spatiotemporal evolution characteristics, analyze the comprehensive early warning level and potential instability mode of landslides.

[0103] The analysis of landslide comprehensive early warning levels and potential instability modes, combining the characteristics of the key disaster-causing factors, the true three-dimensional geological environment background characteristics, and the spatiotemporal evolution characteristics, includes the following steps: S511. Based on a true three-dimensional geomechanical model, key disaster-causing factors and spatiotemporal evolution characteristics are used as real-time input parameters to deduce the stress distribution and strain evolution law of the entire slope.

[0104] First, key disaster-causing factors (groundwater level rise and fall rate and peak value, slip surface softening coefficient, etc.) are input into an interpretable AI model (such as LIME and SHAP algorithms). By decomposing the model decision-making process layer by layer, the contribution weight of each disaster-causing factor to slope instability is quantified and the priority of different factors is clarified. For example, the contribution weight of potential slip surface softening coefficient and surface deformation rate acceleration is significantly higher than that of other factors, and they are used as core early warning factors.

[0105] Secondly, an artificial verification mechanism for slope geological patterns is introduced. Combining the characteristics of rock mass structure of steep slopes and the evolution of underground disaster-prone environment, the initial weights of XAI output are corrected: if the contribution weight of a certain factor (such as peak groundwater level) in the fractured rock mass area is lower than expected by geological patterns, its weight is appropriately increased; if the weight of a certain factor (such as disease density) in the intact rock mass area is too high, its weight is reduced in combination with actual geological conditions. This ensures that the weight allocation conforms to both data-driven patterns and the logic of real geological evolution, ultimately forming a dynamic weight system that provides accurate factor weight support for subsequent early warning level determination.

[0106] S512. Establish an instability mode identification rule base, and combine the inference results with the instability mode identification rule base to analyze the comprehensive early warning level of landslides and potential instability modes.

[0107] Based on a true three-dimensional geomechanical model, the corrected key disaster-causing factors (such as groundwater level dynamics and slip surface softening coefficient) and spatiotemporal evolution characteristics (such as deformation field and stress field evolution data) are used as real-time input parameters to achieve synchronous updates of simulation parameters and actual slope evolution status.

[0108] Secondly, numerical simulation algorithms suitable for steep slopes (such as FLAC3D and Phase2) are selected, and the anisotropic characteristics of the slope rock mass and the evolution law of slip surface weakening are embedded to optimize the stress and strain calculation model, focusing on improving the stress and strain calculation accuracy of potential slip surface areas and rock mass fracture areas.

[0109] Meanwhile, a dynamic simulation step size is set, and the simulation frequency is adaptively adjusted according to the slope evolution speed. Short step sizes (such as 1 hour / step) are used during the deformation acceleration period and the rapid weakening period of the slip surface, while long step sizes (such as 12 hours / step) are used during the stabilization period, which ensures both simulation accuracy and computational efficiency.

[0110] This simulation system accurately predicts the stress distribution and strain evolution of the entire slope, calculates the slope stability coefficients for different time periods and regions, and predicts potential slope failure paths (such as sliding paths along potential slip surfaces and collapse paths in fractured rock areas), providing quantitative mechanical support for instability mode identification and early warning level determination.

[0111] Furthermore, multi-source feature data are integrated as the basis for identification: rock mass structural features (dominant unstable structures, rock mass integrity), surface deformation features (deformation rate, deformation direction, deformation gradient), apparent disease features (disease type, distribution density, expansion trend), spatiotemporal evolution features (stress concentration areas, deterioration field distribution), as well as stress-strain data and failure path inference results obtained from numerical simulation.

[0112] Secondly, an instability mode identification rule base was established. Based on the differences in the core characteristics of different instability modes of high and steep slopes, the identification criteria for each mode were determined: the slip mode corresponds to a clear potential slip surface, uniform deformation rate, and the main damage being slope-directed cracks, with stress concentrated at the top of the slip surface; the collapse mode corresponds to severely fractured rock mass, sudden increase in deformation rate, and the main damage being rockfall and bulging, with stress concentrated at the slope shoulder or toe; the runoff mode corresponds to a continuous rise in groundwater level, severe softening of the slip surface, rapid expansion of the deformation range, and the main damage being large-area water seepage and crack expansion.

[0113] Meanwhile, through multi-feature collaborative matching and geological logic verification, the instability modes of different areas of the slope are accurately identified, and the evolution trend of the instability modes in each area is marked (such as whether the slip mode is transforming into the collapse mode), providing clear model support for subsequent early warning level determination and treatment recommendations.

[0114] Based on this, the judgment dimensions are determined, and quantitative indicators are used for assignment: First, the stability dimension, with the slope stability coefficient (Fs) obtained from numerical simulation as the core indicator, is divided into three basic levels: stable (Fs≥1.3), basically stable (1.1≤Fs<1.3), and unstable (Fs<1.1); Second, the evolution rate dimension, with the surface deformation rate, slip surface weakening rate, and disease expansion rate as the core indicators, is divided into three levels: slow evolution, uniform evolution, and accelerated evolution; Third, the degree of danger dimension, combined with the instability mode, the spatial range of the potential instability area, and the distribution of surrounding personnel and equipment, is divided into three levels: low danger, medium danger, and high danger.

[0115] Secondly, based on the factor weights corrected by XAI, the quantitative indicators of the three dimensions are weighted and summed to obtain a comprehensive early warning score (0-100 points), and correspondingly divided into four levels of early warning: blue warning (80-100 points, stable, only routine monitoring is required), yellow warning (60-79 points, basically stable, increase monitoring frequency), orange warning (40-59 points, unstable, activate early warning plan), and red warning (0-39 points, extremely unstable, immediately activate emergency response).

[0116] At the same time, by combining the spatiotemporal evolution characteristics, the spatiotemporal range corresponding to each warning level is clarified: the specific spatial coordinates of potential instability (such as a certain slope section or a certain elevation range) and the expected time window of instability (such as the possibility of instability within 72 hours during the accelerated evolution period) are marked, so as to achieve accurate determination of both level and spatiotemporal range.

[0117] Furthermore, a standardized and implementable early warning output system will be constructed to comprehensively cover core early warning information and handling guidance, ensuring that early warning results can be directly applied to engineering practice. Specific output content includes: 1. Early warning level: clearly defining the current early warning level (blue / yellow / orange / red) for the entire slope area and each zone, and marking key early warning areas; 2. Potential instability information: accurately marking the spatial location of potential instability (3D coordinates, slope range), the expected instability time window, and the instability mode (sliding / collapse / breakdown), presenting the potential failure path in conjunction with numerical simulation results; 3. Details of key disaster-causing factors. The document outlines five key aspects of the emergency response plan: First, it lists the current main disaster-causing factors, their contribution weights, and real-time monitoring data, clarifying the core reasons for triggering the warning. Second, it provides tiered response recommendations, developing differentiated and actionable measures for different warning levels and instability modes—such as increasing the frequency of drone inspections for yellow warnings, evacuating personnel and equipment from the foot of slopes for orange warnings, and activating emergency rescue plans for red warnings, while also specifying responsibilities and time requirements. Third, it provides recommendations for subsequent monitoring, identifying key monitoring indicators, monitoring frequencies, and monitoring areas based on spatiotemporal evolution trends, providing guidance for dynamic updates of warnings and tracking of response effectiveness.

[0118] All output results can be synchronously connected to the slope monitoring platform to achieve visual display and information push, ensuring rapid transmission of early warning information and timely implementation of response measures.

[0119] In this embodiment, the extraction of the true three-dimensional geological environment background features includes the following steps: S521. Construct a true three-dimensional geometric framework for the real geological morphology of steep slopes.

[0120] First, taking the characteristics of rock mass structure (distribution of dominant joints and zoning of rock mass integrity) as the core constraint, we integrate high-precision slope topographic data (digital elevation model and 3D laser scanning topographic point cloud). By accurately registering the spatial coordinates of topographic data and rock mass structure data, we clarify the geometric boundaries of slope top, shoulder, toe, and slope surface, as well as the spatial range of different rock mass zones.

[0121] Secondly, by combining the spatial variation patterns of slope aspect and slope, the topographic data is reconstructed in detail, focusing on restoring the geometric morphology of special terrains such as steep slopes, slope turning points, and slope foot accumulation areas. At the same time, rock mass layering information (such as weathered layers, intact rock mass layers, and aquifers) is embedded to ensure that the constructed three-dimensional geological framework not only conforms to the actual topographic undulations of the slope, but also reflects the internal structural layering characteristics of the rock mass.

[0122] S522. Based on the true three-dimensional geometric skeleton, embed the structural surface and geological interface to obtain a structurally controlled three-dimensional model.

[0123] All joints and fissures (including millimeter-level micro-fissures and short-length local fissures) are precisely embedded into a three-dimensional geological framework according to their true spatial coordinates and attitude parameters (strike, dip, and dip angle), clarifying the spatial interlacing relationship, extension range, and distribution density of different structural planes, and restoring the degree of fragmentation and structural integrity of the rock mass.

[0124] Secondly, the potential slip surface (including regions with different levels of weakening) obtained by inversion is embedded into the model as the core control structural surface. The spatial location, dip angle, thickness and spatial relationship with surrounding joints and fractures of the slip surface are marked, and the cross relationship between the slip surface and rock mass stratification and aquifer is clarified.

[0125] At the same time, key geological interfaces of the slope (such as the interface between the weathered layer and the intact rock mass, and the interface between the aquifer and the impermeable layer) are embedded to form an integrated structural control three-dimensional model of topography-rock mass stratification-structural surface-potential slip surface, ensuring that the model can accurately reflect the structural control characteristics of the slope.

[0126] S523. Adaptive assignment of mechanical parameters is performed on the structural control type three-dimensional model, and the true three-dimensional geological environment background features are extracted through the assigned structural control type three-dimensional model.

[0127] First, a correlation mapping system of structural characteristics, degree of weakening, and mechanical parameters is established, with rock mass integrity coefficient, anisotropic characteristics, dominant unstable structure, and the inverted degree of slip surface weakening (saturation, softening coefficient) as the core assignment basis.

[0128] Secondly, differentiated values ​​are assigned according to rock mass zoning, structural plane type, and weakening level: higher elastic modulus, internal friction angle, and cohesion are adopted for intact rock mass areas; mechanical parameters are appropriately reduced in fractured rock mass areas and areas with dense fractures, and the mechanical parameters in the direction of fracture development are adjusted according to anisotropic characteristics; the shear strength parameters of potential slip surface areas are reduced according to the weakening level (slightly reduced in slightly weakened areas and significantly reduced in heavily weakened areas), and the mechanical parameters of aquifer areas are corrected in combination with the dynamic characteristics of groundwater level (e.g., the parameters of saturated rock mass are lower than those of unsaturated rock mass).

[0129] A comprehensive feature extraction process is performed on the constructed 3D geomechanical model to form a standardized and reusable 3D geological environment background feature set. Specific extraction contents include: First, three-dimensional geometric features, including the overall geometric shape of the slope, the aspect / gradient of each slope segment, the elevation range of the slope, and the spatial coordinates and geometric parameters of special terrain features (slope shoulders, slope toes, and turning points), to achieve a quantitative description of the slope's geometric shape. Second, structural distribution features, including the spatial distribution coordinates, attitude parameters, distribution density, and extension length of joints, fissures, and potential sliding surfaces, as well as the interrelationship of different structural planes, to clarify the structural control role. Third, mechanical property features, including the elastic modulus, internal friction angle, cohesion, shear strength, and other mechanical parameters of each rock mass zone, structural plane, and potential sliding surface, as well as the spatial distribution differences of these parameters, to provide accurate parameter support for numerical simulation. Fourth, zoning features, based on the integrity of the rock mass, the degree of weakening, and differences in mechanical parameters, the slope is divided into zones, and the stability level (stable, basically stable, unstable) of each zone is marked, clearly identifying key monitoring areas.

[0130] Please see Figure 2 In this embodiment, to efficiently execute the high-steep slope geological hazard monitoring method based on multi-source data fusion provided by the present invention, the present invention also provides a high-steep slope geological hazard monitoring system based on multi-source data fusion, comprising: an input device 1, an output device 2, a processor 3, and a memory 4, wherein the input device 1, output device 2, processor 3, and memory 4 are interconnected, and the memory 4 stores program instructions for executing the steps of the high-steep slope geological hazard monitoring method based on multi-source data fusion. The high-steep slope geological hazard monitoring system based on multi-source data fusion of the present invention has a compact structure and stable performance, and can stably execute the high-steep slope geological hazard monitoring method based on multi-source data fusion provided by the present invention, further enhancing the overall applicability and practical application capability of the present invention.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the present invention.

Claims

1. A method for monitoring geological hazards on steep slopes based on multi-source data fusion, characterized in that, Includes the following steps: Surface deformation characteristics were obtained by tracking surface deformation through PS-InSAR time series analysis and atmospheric phase correction. Identify apparent slope defects and extract their spatiotemporal distribution characteristics. By utilizing the structural characteristics of the slope rock mass and the aforementioned surface deformation characteristics, the spatiotemporal evolution characteristics under the constraints of geological structure are analyzed; Based on the dynamic characteristics of groundwater level and potential slip surface weakening, and the spatiotemporal distribution characteristics of the apparent slope damage, the key disaster-causing factors were extracted after collaborative correction. Based on the characteristics of the key disaster-causing factors, the true three-dimensional geological environment background characteristics, and the spatiotemporal evolution characteristics, the comprehensive early warning level and potential instability mode of landslides are analyzed.

2. The method for monitoring geological hazards on steep slopes based on multi-source data fusion according to claim 1, characterized in that, The method of tracking surface deformation through PS-InSAR time series analysis and atmospheric phase correction to obtain surface deformation characteristics includes the following steps: Layered and partitioned atmospheric phase correction is performed on PS-InSAR data; By introducing the geological laws of slope instability as a solution constraint, the solved temporal deformation data is matched with the three-dimensional geological model mesh to generate a multi-dimensional spatiotemporal feature map.

3. The method for monitoring geological hazards on steep slopes based on multi-source data fusion according to claim 1, characterized in that, The identification of apparent slope defects and the extraction of their spatiotemporal distribution characteristics include the following steps: Incorporate prior information on slope aspect and gradient to perform geometric correction on tilted images; A multi-scale disease detection architecture is constructed, and based on the calibration results, the spatiotemporal distribution characteristics of apparent diseases on slopes are extracted through the multi-scale disease detection architecture.

4. The method for monitoring geological hazards on steep slopes based on multi-source data fusion according to claim 1, characterized in that, The method of analyzing the spatiotemporal evolution characteristics under geological structural constraints by utilizing the structural characteristics of slope rock mass and the surface deformation characteristics includes the following steps: Constructing a multi-scale feature spatial registration system for steep slopes; Based on the structural characteristics of the slope rock mass and the surface deformation characteristics, physical information constraints are embedded. By combining the multi-scale feature space registration system and the physical information constraints, deep spatiotemporal feature fusion is performed to obtain spatiotemporal evolution features under geological structure constraints.

5. The method for monitoring geological hazards on steep slopes based on multi-source data fusion according to claim 4, characterized in that, Obtaining the structural characteristics of the slope rock mass includes the following steps: Adaptive dimensionality reduction of point clouds under slope prior constraints to obtain the most relevant geometric features; Natural roughness features are obtained by adaptive filtering of noise points. The most relevant geometric features, natural roughness features and structural surface parameter set are combined to obtain the structural features of the slope rock mass.

6. The method for monitoring geological hazards on steep slopes based on multi-source data fusion according to claim 1, characterized in that, The process of extracting the key disaster-causing factors after collaborative correction based on the dynamic characteristics of groundwater level and potential slip surface weakening, and the spatiotemporal distribution characteristics of apparent slope damage, includes the following steps: Based on the disaster-prone laws of steep slopes, a multi-source characteristic consistency verification system for deformation, disease, and groundwater level is constructed. Based on the dynamic characteristics of groundwater level and potential slip surface weakening, as well as the spatiotemporal distribution characteristics of the apparent slope damage, false features were eliminated through conflict analysis and credibility assessment to obtain the key disaster-causing factors.

7. The method for monitoring geological hazards on steep slopes based on multi-source data fusion according to claim 6, characterized in that, The analysis of the groundwater level dynamics and potential slip surface weakening characteristics includes the following steps: Based on prior knowledge of slope hydrogeology, a groundwater level time series prediction model is constructed to obtain time series prediction data of groundwater level across the entire region. Construct a slip surface weakening feature inversion model, and invert the key weakening parameters of potential slip surfaces using the slip surface weakening feature inversion model; By combining the global time-series prediction data of groundwater level with the key weakening parameters, the dynamic characteristics of groundwater level and potential slip surface weakening are obtained.

8. The method for monitoring geological hazards on steep slopes based on multi-source data fusion according to claim 1, characterized in that, The analysis of landslide comprehensive early warning levels and potential instability modes, combining the characteristics of the key disaster-causing factors, the true three-dimensional geological environment background characteristics, and the spatiotemporal evolution characteristics, includes the following steps: Based on a true three-dimensional geomechanical model, key disaster-causing factors and spatiotemporal evolution characteristics are used as real-time input parameters to deduce the stress distribution and strain evolution law of the entire slope area; A rule base for identifying instability patterns was established. Based on the simulation results and the rule base, the comprehensive early warning level of landslides and potential instability patterns were analyzed.

9. The method for monitoring geological hazards on steep slopes based on multi-source data fusion according to claim 8, characterized in that, Extracting the true 3D geological environment background features includes the following steps: Construct a true three-dimensional geometric framework for the realistic geological morphology of steep slopes; Based on the true 3D geometric skeleton, structural surfaces and geological interfaces are embedded to obtain a structurally controlled 3D model. In the structurally controlled 3D model, mechanical parameters are adaptively assigned, and the true 3D geological environment background features are extracted from the assigned structurally controlled 3D model.

10. A geological hazard monitoring system for steep slopes based on multi-source data fusion, characterized in that, The high and steep slope geological disaster monitoring system based on multi-source data fusion includes: an input device, an output device, a processor, and a memory. The input device, output device, processor, and memory are interconnected. The memory stores program instructions, which are used to execute the high and steep slope geological disaster monitoring method based on multi-source data fusion as described in any one of claims 1-9.

Images