Method for monitoring instability of soil-rock interface slope under water immersion effect

Abstract

The invention discloses a method for monitoring instability of a soil-rock interface slope under the action of water immersion, and belongs to the field of geological disaster monitoring and geotechnical engineering.The method comprises the following steps that a three-dimensional geological model containing a soil-rock interface is constructed based on a slope geological survey result; arranging a multi-type sensor monitoring network based on the three-dimensional geologic model to collect multi-source monitoring data; calculating a key stability index based on the multi-source monitoring data; on the basis of the key stability indexes, a comprehensive evaluation index is established by adopting a multi-parameter fusion method, a graded early warning threshold value is set, and real-time early warning of the slope stability state is achieved; meanwhile, constructing a multi-dimensional time sequence characteristic matrix based on historical and real-time monitoring data, and predicting the slope instability trend based on the multi-dimensional time sequence characteristic matrix by adopting a gating cycle unit GRU model to obtain advanced early warning data; and carrying out evaluation and graded early warning on the slope state based on real-time early warning and advanced early warning data.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of geological disaster monitoring and geotechnical engineering, and particularly relates to a method for monitoring the instability of a soil-rock interface slope under the action of water immersion. BACKGROUND

[0002] The soil-rock binary structure slope, i.e., the slope with loose soil on the upper part and relatively hard bedrock on the lower part, is widely distributed in nature and engineering activities. For example, such slopes are very common in mountainous areas of highways, railways, water conservancy and hydropower projects, and mining areas. The soil-rock interface, as a transition zone between two media with different physical and mechanical properties, has complex mechanical properties and is often a weak link of slope deformation and failure and a potential sliding surface. Under the action of water immersion such as natural rainfall, reservoir water level fluctuation, canal seepage or artificial irrigation, water will infiltrate into the soil-rock interface along the slope fissures or soil pores. The infiltration of water will significantly change the water content, pore water pressure and effective stress state of the soil in the interface area, resulting in a significant reduction in the shear strength of the soil. On the other hand, the softening and lubricating effects of water and the generation of dynamic water pressure will further weaken the shear resistance of the interface, inducing interface sliding, soil peeling and even overall instability. Currently, landslides, collapses and other geological disasters caused by the instability of such slopes under the action of water immersion are common, posing a serious threat to people's life and property safety and engineering facilities.

[0003] At present, the monitoring methods for slope stability mainly include surface displacement monitoring (such as GPS, InSAR, laser scanning), deep displacement monitoring (such as inclinometer, multi-point displacement meter), hydrological monitoring (pore water pressure gauge, osmometer) and geophysical exploration (such as resistivity method, ground penetrating radar) and the like. However, when the existing monitoring methods are applied to soil-rock interface binary slopes under the action of water immersion, the following shortcomings still exist: (1) First, the monitoring information is single and it is difficult to fully reflect the state of the interface. Most of the current monitoring methods focus on the response of a certain aspect of the slope, such as only displacement or only hydrological changes. For the complex mechanical structure of the soil-rock interface, a single information source cannot fully and accurately capture the subtle changes and deterioration process under the influence of water immersion.

[0004] (2) Then, it is not strong enough for the soil-rock interface. The traditional monitoring arrangement may pay more attention to the surface of the slope or the interior of the homogeneous soil layer, and the direct monitoring and fine analysis capability of the core parameters such as stress, strain, water migration and strength deterioration of the key part of the soil-rock interface is limited.

[0005] (3) Secondly, the dynamic influence of water immersion is not fully evaluated. The influence of water immersion on the stability of the slope is a dynamic development process, involving water infiltration path, saturation zone expansion, pore water pressure growth and decline, and material parameter degradation. The existing method often fails to track these dynamic changes in real time and effectively associate them with the stability evolution of the interface.

[0006] (4) In addition, the reliability and timeliness of the early warning model need to be improved. The current early warning model based on a single indicator (such as displacement threshold) is prone to false negatives or false positives when facing complex slopes and variable water immersion conditions. Therefore, a more reliable and sensitive early warning model needs to be established by integrating multi-source monitoring information.

[0007] (5) Finally, the characteristics of the binary structure are not fully considered. Due to the huge differences in permeability, deformation modulus, and strength parameters between soil and rock, the hydraulic coupling and mechanical interaction between them under water immersion conditions are very complex, such as differential settlement and interface crack expansion. These characteristics have not been fully reflected in conventional monitoring.

[0008] Therefore, developing a new monitoring method that can comprehensively consider the effects of water immersion and the characteristics of the soil-rock binary structure, effectively monitor the stability of the soil-rock interface, and provide timely and reliable early warning information is of great theoretical significance and engineering application value for ensuring the safety of such slopes. SUMMARY

[0009] To solve the above technical problems, the present application provides a monitoring method for the instability of a soil-rock interface slope under water immersion.

[0010] To achieve the above purpose, the present application provides a monitoring method for the instability of a soil-rock interface slope under water immersion, comprising: constructing a three-dimensional geological model containing a soil-rock interface based on the results of slope geological survey; laying out a multi-type sensor monitoring network based on the three-dimensional geological model to collect multi-source monitoring data; calculating key stability indicators based on the multi-source monitoring data; based on the key stability indicators, using a multi-parameter fusion method to establish a comprehensive evaluation index, and setting a graded early warning threshold to realize real-time early warning of the stability state of the slope; At the same time, based on historical and real-time monitoring data, a multi-dimensional time series feature matrix is constructed, and based on the multi-dimensional time series feature matrix, a gated recurrent unit (GRU) model is used to predict the slope instability trend to obtain early warning data; Based on the real-time warning and the early warning data, the state of the slope is evaluated and graded.

[0011] Optionally, the three-dimensional geological model is used to deploy a multi-type sensor monitoring network, including: deploying a sensor array at key positions of the spatial undulations of the soil-rock interface under the guidance of the three-dimensional geological model; the sensor array includes a pore water pressure sensor, a soil pressure sensor, a strain sensor, and a displacement sensor. The pore water pressure sensor is arranged in a three-dimensional grid form on the upper and lower regions of the soil-rock interface, the soil pressure sensor and the strain sensor are arranged along the normal and tangential directions of the soil-rock interface, and the displacement sensor forms a monitoring network on the surface of the slope and is arranged in a vertical direction in a deep borehole.

[0012] Optionally, the key stability indicators include: interface effective stress, hydraulic gradient, seepage force, interface shear strength reduction factor, cumulative displacement, displacement rate, and shear strain.

[0013] Optionally, based on the key stability indicators, a multi-parameter fusion method is used to establish a comprehensive evaluation index, including: The key stability indicators are normalized into dimensionless parameters using a multi-parameter fusion algorithm; The weight coefficients of each parameter are determined based on the analytic hierarchy process; Based on the dimensionless parameters and the weight coefficients of each parameter, a comprehensive evaluation index is generated by weighted summation.

[0014] Optionally, the calculation expression of the comprehensive evaluation index is: ; In the formula, FSI is the comprehensive evaluation index, is the weight of different monitoring parameters in the comprehensive evaluation, is the normalized value of the kth parameter.

[0015] Optionally, the process of constructing a multi-dimensional time sequence feature matrix based on historical and real-time monitoring data includes: Collecting external driving factor sequences, internal state response sequences, and static structural feature sequences; The monitoring data in the external driving factor sequences, internal state response sequences, and static structural feature sequences are standardized by using the maximum-minimum normalization method, and the processed data are time-aligned and feature-fused to form a multi-dimensional time sequence feature matrix.

[0016] Optionally, based on the multi-dimensional time sequence feature matrix, a gated recurrent unit (GRU) model is used to predict the slope instability trend to obtain early warning data, including: The normalized multi-dimensional time sequence feature matrix is input into a multi-layer GRU neural network, and long-term dependence in the time sequence data is learned through reset gate and update gate mechanisms in the GRU unit; wherein, the reset gate vector is calculated based on the input vector of the current time step and the hidden state of the previous time step, to control the forgetting degree of the historical information; the update gate vector is calculated based on the same input, to control the update degree of the new information on the hidden state; the candidate hidden state is calculated based on the output of the gating mechanism, and the final hidden state of the current time step is obtained in combination with the result of the update gate; high-order time sequence features are extracted layer by layer through the multi-layer GRU network, the hidden state of the last time step is input into a fully connected layer, and the prediction value of the key stability index of the slope in the future period is mapped to form the early warning data.

[0017] Compared with the prior art, the present application has the following advantages and technical effects: (1) The present application can comprehensively monitor the changes of physical quantities such as displacement, pore water pressure, soil pressure and environmental temperature and humidity of the soil-rock interface slope under the action of water immersion through the cooperative arrangement of various types of sensors, and obtain rich monitoring information.

[0018] (2) The present application realizes the comprehensive monitoring of hydrology, stress, strain, displacement and internal damage, and the information sources are diverse, which can be verified with each other, reduce the misjudgment possibly caused by a single information source, and improve the comprehensiveness and reliability of the monitoring results.

[0019] (3) The present application realizes the real-time tracking of the dynamic response process of the slope under the influence of water immersion, considers the hysteresis effect of seepage influence, combines the multi-parameter fusion early warning and prediction model, and can identify the potential risk earlier before the slope deforms significantly or the whole instability, so as to gain valuable time for emergency disposal. BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of the present application illustrated in the drawings, and their description thereto, are presented to provide the applicant's best contemplation of the present application, but not to form an improper limitation to the present application. In the drawings: Figure 1 The flow chart of the slope instability monitoring method of the embodiment of the present application; Figure 2 The monitoring point layout schematic diagram of the embodiment of the present application; Figure 3 The data prediction flow chart of the embodiment of the present application; Figure 4 The data prediction schematic diagram of the embodiment of the present application; Figure 5 The early warning grade and prediction linkage mechanism schematic diagram of the embodiment of the present application. DETAILED DESCRIPTION

[0021] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other in the case of no conflict. The present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0022] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown.

[0023] As Figure 1 shown, a method for monitoring the instability of a soil-rock interface slope under the action of water immersion is provided in the embodiment, which realizes early and accurate early warning of slope instability through multi-parameter and multi-dimensional monitoring and analysis of soil-rock interface characteristics, and specifically includes the following steps: a three-dimensional geological model containing a soil-rock interface is constructed based on the results of slope geological survey; a multi-type sensor monitoring network is laid out based on the three-dimensional geological model to collect multi-source monitoring data; key stability indexes are calculated based on the multi-source monitoring data; a comprehensive evaluation index is established by using a multi-parameter fusion method based on the key stability indexes, and a graded early warning threshold is set to realize real-time early warning of the slope stability state; at the same time, a multi-dimensional time sequence feature matrix is constructed based on historical and real-time monitoring data, and a GRU (Gated Recurrent Unit) model is used to predict the slope instability trend based on the multi-dimensional time sequence feature matrix to obtain early warning data; the slope state is evaluated and graded based on the real-time early warning and early warning data.

[0024] S1: Slope geological survey and three-dimensional model construction The topography, stratum lithology, structural characteristics, hydrogeological conditions and historical deformation and damage of the slope to be monitored are investigated in detail. Through drilling, geophysical prospecting and other means, the thickness, nature, spatial distribution of the overlying soil layer, and the lithology, occurrence, weathering degree and structure surface development of the underlying bedrock are found out. The burial depth, shape, contact relationship (conformity or unconformity) and interface material characteristics of the soil-rock interface are highlighted. On this basis, a refined three-dimensional geological model containing terrain, soil layer, bedrock and key structure surface (especially soil-rock interface) is constructed by using three-dimensional geological modeling software. The model provides a spatial basis for subsequent sensor optimization and data interpretation.

[0025] S2: Multi-parameter monitoring system layout To address the characteristics of the soil-rock interface and the potential impacts of immersion and water level changes, multiple types of sensors are deployed at key locations on the slope (such as the top, middle, and toe of the slope, and along and near the soil-rock interface) to form an integrated, automated monitoring network. First, hydrological dynamic monitoring: pore water pressure sensors (piezometers) are buried near the soil-rock interface and at different depths in the overlying soil layer to monitor changes in pore water pressure in real time. Soil moisture sensors are deployed in the soil layer to monitor changes in soil saturation. Second, stress-strain monitoring: earth pressure cells are buried at or near the soil-rock interface to monitor changes in the interface's normal and tangential stresses (or total stress). Interface shear displacement gauges or distributed fiber optic strain sensors are deployed along the dip of the soil-rock interface to directly monitor the relative slip and strain distribution of the interface. Finally, displacement monitoring: high-precision GPS receivers or total station reflecting prisms are deployed on the slope surface to monitor three-dimensional surface displacement. Install fixed inclinometers or multi-point displacement gauges inside the borehole to monitor deep displacements near the soil-rock interface and potential slip zones. Use UAV photogrammetry or terrestrial 3D laser scanning technology to periodically acquire high-density slope point cloud data and analyze the overall slope deformation field.

[0026] S3: Monitoring data acquisition, transmission and processing All sensors are connected to the data acquisition unit and remote transmission unit via wired or wireless means to achieve automated, real-time acquisition and remote transmission of monitoring data. A monitoring database is established, and the raw data is preprocessed, including data filtering, noise removal, unit conversion, and error correction. Based on the monitoring data, the following key data are calculated: effective stress at the soil-rock interface; hydraulic gradient near the soil-rock interface based on pore water pressure gauge readings and elevation differences at different locations; seepage force along the potential sliding direction under immersion; estimation of interface shear strength parameters; calculation of cumulative displacement, three-dimensional displacement vector, average displacement rate, and acceleration at each monitoring point; calculation of shear strain near the soil-rock interface based on distributed fiber optic or displacement gauge data; and estimation of soil saturation based on moisture content monitoring data and soil-water characteristic curves.

[0027] Calculate the effective normal stress at the interface based on the monitored values ​​of pore water pressure and total normal earth pressure at (or immediately adjacent to) the soil-rock interface. :

[0028] In the formula: The total normal stress at the interface is u; the pore water pressure at the interface is u. The dynamic evaluation of the shear strength parameters of the soil-rock interface can be used as a sensitive indicator for early warning. The effective cohesion and the effective internal friction angle of the soil-rock interface are obtained by combining the results of the shear tests on the material taken from the soil-rock interface in the laboratory under saturated and natural conditions, and the shear strength of the interface is evaluated according to the real-time monitoring of the effective normal stress of the interface.

[0029] In the formula: and are the shear strength parameters of the soil-rock interface, and they are significantly dynamically reduced due to the water-softening effect, being functions of the water content w and the water immersion duration t In practical applications, the relationship curve between the strength parameters in the natural state and the water content / saturation can be established by laboratory tests to correct the strength parameters in the natural state.

[0030] The dynamic change of the hydraulic gradient can also be used as a sensitive indicator for early warning by monitoring the readings of the pore water pressure gauges and at different positions on and below the soil-rock interface, the elevation differences and , and the distance L between the two points perpendicular to the interface, to calculate the hydraulic gradient i perpendicular to the interface.

[0031] In the formula: and are the readings of the pore water pressure gauges at different positions on and below the soil-rock interface; and are the elevation differences at the corresponding positions; L is the distance between the two points perpendicular to the interface, which is an approximation of the length of the seepage path in that direction for the water flow; is the specific weight of water.

[0032] The cumulative displacement amount, the three-dimensional displacement vector, the average displacement rate, the acceleration, etc. of each monitoring point, especially the interface shear displacement meter and the inclinometer at the interface depth, are calculated.

[0033] In the formula: v is the interface shear displacement rate, is the interface shear displacement increment in a certain period of time.

[0034] Subsequently, the average shear strain near the soil-rock interface is calculated according to the distributed optical fiber or interface shear displacement meter group data.

[0035] wherein: is the interface shear displacement, is the effective thickness of the interface shear band.

[0036] S4: Instability early warning model establishment based on multi-parameter fusion Firstly, single parameter early warning threshold can be set, and according to the geological conditions of the slope, engineering experience, laboratory test results and numerical simulation analysis, the graded early warning threshold (attention level, alert level, danger level) is set for each key monitoring parameter (such as pore water pressure, displacement rate, interface shear strain).

[0037] Since different monitoring parameters have different physical units (such as kPa, mm / d, dimensionless hydraulic gradient, etc.) and numerical ranges, it is meaningless to directly add these original values. At the same time, the original value of some parameters and the risk degree they represent may not be a simple linear relationship. For example, the risk is very low when the displacement rate is very small, but once it exceeds a certain inflection point, the risk may increase sharply. Therefore, a comprehensive index can be established to quantitatively evaluate the overall stability or risk degree of the soil-rock interface under the condition of immersion. Its main purpose is to combine multiple monitoring parameters from different sensors that reflect different aspects of the slope state into a single quantitative index that can comprehensively evaluate the current stability state of the soil-rock interface.

[0038] By analyzing historical data and instability mechanism, several key parameter combinations that play a dominant control role in the stability of the soil-rock interface under the action of immersion are determined, and finally the comprehensive evaluation index FSI is constructed:

[0039] wherein: is the weight of different monitoring parameters in the comprehensive evaluation, which can be obtained by using the analytic hierarchy process (AHP) or expert scoring method, and dynamically updated by machine learning method; is the normalized value of the kth parameter or a function of its proximity to the threshold value.

[0040] Based on the key stability index, the process of establishing a comprehensive evaluation index by using multi-parameter fusion method includes: using multi-parameter fusion algorithm to normalize the key stability index into a dimensionless parameter; determining the weight coefficient of each parameter based on the analytic hierarchy process; based on the dimensionless parameter and the weight coefficient of each parameter, the comprehensive evaluation index is generated by weighted summation.

[0041] For the complex problem of soil-rock interface slope instability under water immersion, the sensitivity and indication of different monitoring parameters are different. Parameters that directly reflect the interface hydraulic state (such as interface pore water pressure) and parameters that directly reflect the interface deformation (such as interface shear displacement rate) may be more important than other indirect parameters. By giving these key parameters a higher weight, FSI can more accurately reflect the true dangerous state of the slope. Finally, the model is trained by historical data (including instability cases and stable cases), allowing the model to automatically learn the contribution of each parameter to predicting instability, thereby continuously updating the weights.

[0042] The dimensionless value obtained after processing the kth monitoring parameter, which reflects the risk level or the degree of contribution to instability represented by the current state of the parameter. It is not the original monitoring data value. First, the original monitoring parameter value mapped to the interval [0, 1] by linear normalization, 0 may represent the safest state, and 1 represents the most dangerous state (reaching or exceeding the critical threshold).

[0043] In the formula: P is the original monitoring parameter value; Lk is a lower threshold value set for the kth monitoring parameter, which is based on long-term background value observation, data range in stable state, and relevant specification standards; Ck is an upper critical threshold value set for the kth monitoring parameter, when the actual monitoring value P of the parameter reaches or exceeds this threshold value, it is considered that the state indicated by the parameter has entered a critical or dangerous state.

[0044] The higher the value of FSI, the more unstable the soil-rock interface, and the greater the risk of instability and failure; on the contrary, the lower the value, the more stable the interface. The absolute value of this index may not have a physical unit, but its relative change and comparison with the preset threshold have practical significance. Similarly, different thresholds can be set to achieve graded early warning of slope instability risk.

[0045] When a single parameter reaches its threshold value, or the comprehensive evaluation index reaches the set warning level, the system automatically triggers an audible and visual alarm, and sends warning information to management personnel through SMS, APP push, network platform, etc., including warning level, main overrun parameters, and possible instability areas. According to the accumulation of monitoring data and feedback from the actual situation, the parameters, weights, and thresholds of the warning model are continuously modified and optimized to improve the accuracy and reliability of the warning.

[0046] S5: Multi-parameter, multi-feature fusion instability prediction model establishment The process of obtaining the early warning data by predicting the slope instability trend based on the multi-dimensional time sequence feature matrix using the gated recurrent unit GRU model includes: inputting the normalized multi-dimensional time sequence feature matrix into a multi-layer GRU neural network, and learning the long-term dependence relationship in the time sequence data through the reset gate and update gate mechanism in the GRU unit; wherein, the reset gate vector is calculated based on the input vector of the current time step and the hidden state of the previous time step, to control the forgetting degree of historical information; the update gate vector is calculated based on the same input, to control the updating degree of the hidden state by new information; the candidate hidden state is calculated based on the output of the gating mechanism, and the final hidden state of the current time step is obtained in combination with the result of the update gate; high-order time sequence features are extracted layer by layer through the multi-layer GRU network, the hidden state of the last time step is input into the full connection layer, and the prediction value of the key stability index of the slope in the future period is mapped to form the early warning data.

[0047] Firstly, a multi-dimensional dynamic and static input feature matrix is constructed. In order to realize accurate prediction of the slope instability trend, this step aims to construct a multi-dimensional input feature matrix which can comprehensively reflect external driving factors, internal state response and static structural feature sequence. The matrix is composed of the following three types of data: ① External driving factor sequence: quantitative rainfall forecast data of future M time steps are obtained and processed, and combined with historical and current real-time rainfall, reservoir water level, environmental temperature and humidity and other external environmental monitoring values to form an external driving factor time sequence. This is the key input to realize "early" prediction.

[0048] ② Internal state response sequence: internal multi-source monitoring data of the slope in the past T time steps are collected, including but not limited to: pore water pressure near the soil-rock interface, soil moisture content, soil pressure, interface shear displacement, deep displacement rate, slope displacement, etc., to form an internal state response time sequence.

[0049] ③ Static structural feature sequence: due to the time effect of rainfall and water level changes, there is a lag in the change of water level and water pressure in the soil. The prediction index of the day may have strong or strong correlation with the data of a certain period of time in the past, therefore, feature construction is needed. At the same time, the immediate deformation state of the slope should also be considered, and the displacement rate of the slope is taken as the construction feature.

[0050] For each monitoring sensor point, the static geological and geometric feature parameters of its location are extracted, including: slope, slope direction, overburden thickness, soil-rock interface burial depth, rock-soil permeability coefficient, initial value of cohesion and internal friction angle, etc. The vector remains unchanged in the time dimension, providing physical environmental constraints for the model. The above three types of data are time-aligned and normalized by maximum and minimum, as shown in equation (9). Finally, a multi-dimensional time sequence feature matrix for model input is fused.

[0051] wherein: wherein: X scaled is the normalized data; X is the monitoring data; X min is the minimum value in the monitoring data; X max is the maximum value in the monitoring data.

[0052] The processed time feature sequence (containing all features) is input into a gated recurrent unit (hereinafter referred to as GRU), and the GRU selects different layers according to the data complexity, thereby effectively learning the long-term dependence relationship in the time series data and then performing key data prediction. The reset gate gives the model the control ability of short-term memory. When there is a structural breakpoint in the sequence or it is necessary to start accumulating information again, the model can learn the reset gate r t is set to a vector close to 0, effectively ignoring the historical context, thereby generating a new state only according to the current input x t The update gate is responsible for the maintenance of long-term memory. The time feature sequence gated recurrent unit first calculates the reset gate r t and the update gate z t as shown in equations (10) and (11).

[0053] wherein: x t is the input vector of time step t; h t-1 is the hidden state vector of time step t-1; W r and W z are the weight matrices of the reset gate and the update gate, respectively; b r and b z are the corresponding bias terms, and σ is the Sigmoid activation function.

[0054] The candidate hidden state and the final hidden state are then calculated through the above gating mechanism.

[0055] wherein: W h is the weight matrix of the candidate state.

[0056] When the input data has many features and is complex, multi-layer GRUs can be used to further capture the correlations between data. Building upon the patterns extracted by the first layer, more advanced and abstract temporal dependencies can be learned. At each time step t, the hidden state output of the first-layer GRU is... h t(1) This will serve as the input to the second-layer GRU at this time step. Subsequently, a fully connected layer maps the final hidden state of the last time step to the desired output dimension, thus completing the prediction.

[0057] S6: Slope Condition Assessment By combining real-time monitoring data and advanced early warning data, the condition of the slope is evaluated, and an early warning is issued in a timely manner when the real-time data assessment or advanced prediction reaches the critical value.

[0058] Furthermore, as another specific implementation method of this embodiment, the three-dimensional spatial morphology, undulation characteristics, approximate slope changes, physical and mechanical properties of the soil near the interface, and the degree of weathering of the bedrock were first identified and finely characterized through on-site surveying, five geological exploration boreholes (the boreholes penetrated the soil-rock interface and entered the bedrock for no less than 5m, with key core sampling and characterization of the interface), and high-density resistivity detection along multiple representative profiles. By establishing a detailed three-dimensional geological model of the slope, the spatial occurrence state of the soil-rock interface can be displayed.

[0059] Based on the analysis of the soil-rock interface morphology and potential water infiltration paths revealed by the 3D model, monitoring sections were selected to be set up in the interface depression area, the interface exposure area at the rear edge of the historical landslide, and the interface section at the toe of the slope affected by the stream, such as... Figure 2 As shown. Pore water pressure gauges 1 (3 in total) are installed in the soil 1m above the soil-rock interface at each monitoring section, at the interface (using special drilling and sensor implantation techniques), and layer by layer. At the soil-rock interface of selected key sections, miniature earth pressure cells 2 (3 sets in total) and vibratory strain gauges 4 are installed using a combination of boreholes and small horizontal pits, placed close to or very close to the soil-rock interface for direct monitoring of interface strain changes. Rain gauges 5 and meteorological monitoring stations 6 are then installed. Three GNSS monitoring points 7 are set up on the slope (located on the slope, in the middle of the slope, and at the toe of the slope, respectively). A fixed inclinometer 8 is then installed in the borehole (through the interface) to observe the rate and direction of displacement changes at the depth of the soil-rock interface. During the rainy season, the frequency of oblique photography by UAVs 9 is increased to obtain high-precision DEMs for difference analysis, focusing on whether slope deformation is related to deep interface activity, while also recording water level changes. All sensors are models with automatic data acquisition and wireless transmission capabilities, and data is uploaded to a cloud server in real time via a 4G network module 10.

[0060] The system is set to automatically collect data every hour during the flat water period, and to collect data every 10-15 minutes during heavy rainfall or when the water content rises significantly. After the data is preliminarily processed by the cloud platform, the derived indicators are calculated according to the formulas (1)-(6) of the invention. For the interfacial pore water pressure: more than 30% of the average stable seepage value during the rainy season for many years is the attention level; more than 80% of the highest measured value in history is the warning level. The hydraulic gradient: greater than 0.3 is the attention level; greater than 0.5 (may be close to the critical value of seepage failure) is the warning level. The shear displacement rate: greater than 0.1 mm / d is the attention level; greater than 0.5 mm / d is the warning level; greater than 1 mm / d is the danger level. Then, the comprehensive evaluation index FSI is constructed based on formula (7), and the direct related parameters are given weights. FSI greater than 0.4 is the attention level, greater than 0.65 is the warning level, and greater than 0.85 is the danger level. Finally, real-time monitoring data evaluation is performed.

[0061] Considering that the seepage of water in the soil has a lagging effect on the slope soil, and the gradual failure process of the slope also has a certain historical influence effect, there is limitation in evaluating or predicting the slope state only from real-time monitoring data. By constructing new characteristic data to enrich the data source, the data information can be further mined. For example, by constructing: lagging rainfall and cumulative slope displacement, etc. After processing and normalizing the monitoring data, a total of 26 characteristic data are obtained: slope top settlement value A1, slope top horizontal displacement A2, slope middle settlement value B1, slope middle horizontal displacement B2, slope foot settlement value C1, slope foot horizontal displacement C2, pore water pressure D1, pore water pressure D2, pore water pressure D3, soil pressure E1, soil pressure E2, soil pressure E3, strain F1, deep displacement G1, water level H1, current day rainfall K1, lagging 5-day rainfall K2, lagging 10-day rainfall K3, lagging 15-day rainfall K4, forecasted tomorrow's rainfall M1, slope top lagging 5-day cumulative displacement N1, slope middle lagging 5-day cumulative displacement N2, slope top lagging 5-day cumulative displacement N3, slope top lagging 10-day cumulative displacement O1, slope middle lagging 10-day cumulative displacement O2, and slope top lagging 10-day cumulative displacement O3.

[0062] The above multi-feature data time series along the time scale is divided into n time step sample data for rolling prediction, as shown in Figure 3 The slope surface displacement tomorrow is predicted by the above sample data (selecting the required key data), as shown in Figure 4 Finally, the warning level and prediction linkage are completed based on the above process, as shown in Figure 5 According to the current warning level, the system automatically sends a warning SMS to the on-duty engineer's mobile phone when the warning is triggered.

[0063] The above merely provides the preferred embodiments of the present application, and the protection scope of the present application is not limited thereto, and any changes or substitutions within the technical scope disclosed by the present application should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for monitoring slope instability at the soil-rock interface under water immersion, characterized in that, Includes the following steps: A three-dimensional geological model including the soil-rock interface was constructed based on the results of slope geological investigation; Based on the aforementioned three-dimensional geological model, a multi-type sensor monitoring network is deployed to collect multi-source monitoring data; Key stability indicators are calculated based on the multi-source monitoring data. Based on the aforementioned key stability indicators, a comprehensive evaluation index is established using a multi-parameter fusion method, and graded early warning thresholds are set to achieve real-time early warning of slope stability. Meanwhile, a multi-dimensional time-series feature matrix is ​​constructed based on historical and real-time monitoring data. Based on the multi-dimensional time-series feature matrix, a gated cyclic unit (GRU) model is used to predict the slope instability trend and obtain early warning data. The slope condition is evaluated and graded based on the real-time early warning and the advanced early warning data.

2. The method for monitoring slope instability at the soil-rock interface under water immersion as described in claim 1, characterized in that, The three-dimensional geological model is used to deploy a multi-type sensor monitoring network, which includes: under the guidance of the three-dimensional geological model, sensor arrays are deployed at key locations along the spatial undulations of the soil-rock interface; the sensor arrays include pore water pressure sensors, soil pressure sensors, strain sensors and displacement sensors; The pore water pressure sensor is arranged in a three-dimensional grid in the upper and lower regions of the soil-rock interface. The soil pressure sensor and the strain sensor are arranged along the normal and tangential directions of the soil-rock interface. The displacement sensor forms a monitoring network on the slope surface and is arranged vertically in deep boreholes.

3. The method for monitoring slope instability at the soil-rock interface under water immersion as described in claim 1, characterized in that, The key stability indicators include: effective interfacial stress, hydraulic gradient, permeability, interfacial shear strength reduction factor, cumulative displacement, displacement rate, and shear strain.

4. The method for monitoring slope instability at the soil-rock interface under water immersion as described in claim 1, characterized in that, The process of establishing a comprehensive evaluation index based on the aforementioned key stability indicators using a multi-parameter fusion method includes: The key stability indexes are normalized into dimensionless parameters using a multi-parameter fusion algorithm. The weight coefficients of each parameter are determined based on the analytic hierarchy process (AHP). A comprehensive evaluation index is generated by weighted summation based on dimensionless parameters and the weighting coefficients of each parameter.

5. The method for monitoring slope instability at the soil-rock interface under water immersion as described in claim 4, characterized in that, The formula for calculating the comprehensive evaluation index is as follows: ； In the formula, FSI is the comprehensive evaluation index. Assigning weights to different monitoring parameters in the comprehensive evaluation. This is the normalized value of the k-th parameter.

6. The method for monitoring slope instability at the soil-rock interface under water immersion as described in claim 1, characterized in that, The process of constructing a multidimensional time-series feature matrix based on historical and real-time monitoring data includes: Collect external driving factor sequences, internal state response sequences, and static construction feature sequences; The monitoring data in the external driving factor sequence, internal state response sequence and static constructed feature sequence are standardized by the max-min normalization method. The processed data are then time-aligned and feature-fused to form a multi-dimensional time series feature matrix.

7. The method for monitoring slope instability at the soil-rock interface under water immersion as described in claim 1, characterized in that, The process of predicting slope instability trends and obtaining early warning data based on the multidimensional time-series feature matrix using a gated cyclic unit (GRU) model includes: The normalized multidimensional temporal feature matrix is ​​input into a multilayer GRU neural network. The long-term dependencies in the temporal data are learned through the reset and update gate mechanisms in the GRU units. Specifically, the reset gate vector is calculated based on the input vector of the current time step and the hidden state of the previous time step to control the degree of forgetting of historical information; the update gate vector is calculated based on the same input to control the degree of updating of the hidden state with new information; candidate hidden states are calculated based on the output of the gating mechanism, and the final hidden state of the current time step is obtained by combining the result of the update gate; high-order temporal features are extracted layer by layer through the multilayer GRU network, and the hidden state of the last time step is input into a fully connected layer to map and obtain the predicted values ​​of key slope stability indicators for future periods, forming advanced early warning data.

Images