High-fill reinforced slope monitoring and early warning method based on artificial intelligence

Abstract

The invention discloses a high-fill reinforced slope monitoring and early warning method based on artificial intelligence. The method comprises the steps of high-fill slope region division, perceptual geogrid arrangement, slope internal data monitoring, physical field inversion, spatio-temporal data set construction, slope abnormal state recognition, risk assessment and risk early warning. Sensing type geogrids integrated with distributed optical fiber sensors are arranged in a layered mode in the construction process, strain, displacement and water pressure data in the slope are collected in real time, a continuous stress field, a displacement field and a water pressure field are obtained through inversion, the abnormal state in the slope is intelligently recognized through a deep learning model, the risk is quantitatively evaluated, and multi-stage early warning is started. According to the invention, full-field, real-time and three-dimensional perception and intelligent prospective early warning of the internal state of the slope are realized, and the monitoring accuracy and the engineering safety are obviously improved.

Description

Technical Field

[0001] This invention belongs to the field of slope engineering technology and relates to a monitoring and early warning method for high embankment reinforced slopes based on artificial intelligence. Background Technology

[0002] Reinforced high embankment slopes are common geotechnical structures in major engineering projects such as highways, railways, airports, and water conservancy projects. Their stability directly affects the safety and long-term operational performance of the project. Geogrids, as reinforcement materials, effectively improve the overall stability and bearing capacity of the slope through their interaction with the soil. However, because the internal structure of reinforced slopes is invisible, traditional monitoring methods struggle to achieve comprehensive and continuous perception of their internal state, resulting in limited ability to identify localized damage such as reinforcement fracture, soil arching failure, and shear band formation. Once slope instability occurs, it often causes severe economic losses and safety hazards. Therefore, developing a method capable of real-time, accurate monitoring and intelligent early warning is of great significance.

[0003] Currently, monitoring of reinforced high-fill slopes mainly relies on point-based monitoring equipment such as inclinometers, settlement gauges, stress gauges, and pore water pressure gauges. These methods have the following significant limitations: First, the monitoring range is limited, only acquiring data from discrete points, making it difficult to comprehensively reflect the continuous distribution characteristics of the internal physical field of the slope, and easily missing the emergence and development of local anomalies; Second, there is a lack of deep fusion and coupling analysis between multi-physics field data, making it impossible to construct a three-dimensional visualization model that truly reflects the mechanical state of the slope, thus limiting the accuracy of overall stability assessment; Third, existing early warning mechanisms are mostly based on empirical thresholds for single parameters, making it difficult to identify failure modes with spatial distribution characteristics such as continuous shear deformation zones and reinforcement creep, leading to frequent false alarms and missed alarms, significant early warning lag, and a lack of foresight.

[0004] In recent years, fiber optic sensing technology has been gradually introduced into the field of slope engineering monitoring due to its advantages such as resistance to electromagnetic interference, high durability, and the ability to achieve distributed measurement. Existing research has attempted to attach fiber optic sensors to geogrids to monitor the strain distribution of the reinforced structure. However, current methods mostly remain at the level of simple processing of sensor data and threshold alarms, failing to effectively utilize massive distributed data to construct a three-dimensional solid model of the slope's internal state, nor fully integrate artificial intelligence technology to achieve proactive identification and risk assessment of abnormal patterns.

[0005] Therefore, there is an urgent need to develop a monitoring method that can integrate multi-source sensor information, construct a real physical field model inside the slope, and integrate intelligent algorithms to achieve accurate identification and graded early warning, so as to improve the safety monitoring capability and risk response level of high-fill reinforced slopes. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention proposes a monitoring and early warning method for reinforced high-fill slopes based on artificial intelligence. By embedding distributed optical fibers in the geogrid of the reinforced high-fill slope, multi-parameter data of the slope interior is acquired, and artificial intelligence technology is used to automatically identify anomalies within the slope, thereby enabling the monitoring and early warning of instability risks of reinforced high-fill slopes.

[0007] A monitoring and early warning method for reinforced high-fill slopes based on artificial intelligence includes the following steps: S1. Regional division of high embankment slopes: Based on the engineering geological and hydrogeological conditions of the reinforced high fill slope, a three-dimensional model of the slope was established. The safety factor of the slope was calculated using the finite element strength reduction method, and the hydraulic gradient distribution inside the slope was determined based on seepage finite element analysis. According to the calculation results, the high fill slope was divided into potential sliding zone, seepage sensitive zone, and low-risk zone: the potential sliding zone was defined as the zone with a safety factor lower than 1.25, the seepage sensitive zone was defined as the zone with a hydraulic gradient greater than the critical hydraulic gradient, and the remaining zones were classified as low-risk zones. S2. Layout of sensing geogrid: During the construction of high embankment slopes, reinforcement layers are laid in layers according to the design elevation to form a high embankment reinforced slope. For different risk areas divided in step S1, a differentiated sensing geogrid deployment strategy is adopted: in potential sliding areas and seepage-sensitive areas, the sensing geogrid is deployed in each reinforcement layer; in low-risk areas, a sensing geogrid is deployed every 2 to 4 reinforcement layers. The sensing geogrid, while possessing the mechanical properties of conventional geogrids, can also distributedly monitor strain, displacement, and groundwater pressure data at different locations inside the slope. S3. Monitoring of internal slope data: The multi-physics data inside the slope are collected in real time using the layered geogrid deployed in step S2; the data includes strain at different locations and depths along the laying path of the geogrid. e ( x , y , z , t ), displacement d ( x , y , z , t ) and water pressure p ( x , y , z , t ),in x , y , z Represents spatial coordinates,t Represents a time variable; S4, Inversion of Physical Fields: Based on the discrete monitoring data of the slope interior collected in step S3, i.e. strain e ( x , y , z , t ), displacement d ( x , y , z , t ) and water pressure p ( x , y , z , t The continuous stress field was obtained by inversion using spatial interpolation methods. E 0 Displacement field D 0 and water pressure field P 0 In the physical field inversion process, soil mechanics and seepage theory are introduced as physical constraints. Among them, stress field inversion needs to be combined with the Mohr-Coulomb constitutive model of slope soil to ensure that the principal stress value at any spatial point does not exceed the shear strength of the soil; water pressure field inversion needs to satisfy Darcy's law to maintain the rationality of the seepage field; by iteratively correcting the interpolation results, the inversion distortion caused by discrete data noise or sparsity is eliminated, and finally a continuous physical field distribution that conforms to the laws of mechanics and seepage is output. S5. Construction of spatiotemporal dataset: The stress field, displacement field, and water pressure field obtained from the inversion in step S4, which conform to physical laws, are preprocessed by normalization to uniformly scale the values ​​of each physical field to the [0,1] interval; the normalization process is performed according to the following expression:

[0008] in, X The physical field data to be normalized X max and X min Data respectively X The global maximum and global minimum values; The three-dimensional model of the slope is discretized into a regular three-dimensional mesh system, and then the normalized stress field is... E Displacement field D and water pressure field P The data is resampled to this grid node to ensure that each spatiotemporal point ( x , y , z ,t Each of these corresponds to a set of standardized physical field values. e , d , p This allows for the construction of standardized spatiotemporal data sequences. x , y , z , t , e , d , p This provides structured input data for subsequent deep learning models; S6. Slope anomaly identification: Based on the spatiotemporal data sequence constructed in step S5, a specially trained deep learning model is used to intelligently identify the abnormal states inside the slope. The deep learning model is a neural network adapted to the spatiotemporal data structure, which integrates three-dimensional convolution operations and attention mechanisms, and can effectively capture the characteristics of the abnormal states of the multi-physics fields inside the slope in spatial distribution and temporal evolution. The abnormal states include: continuous banded shear deformation zones appearing in the displacement field; signs of soil arching failure reflected by the stress field; abrupt changes in strain or displacement at the reinforcement location, indicating the risk of increased creep or fracture of the reinforcement; and potential slip surfaces shown by the coupling effect of the water pressure field and the displacement field. S7. Risk Assessment: Based on the slope internal anomaly patterns and their characteristic parameters identified in step 6, a real-time risk index is used. R A quantitative assessment of the slope safety status is conducted; the real-time risk index... R Calculated using the following multi-factor weighted model;

[0009] in, I a The abnormal region size factor is the ratio of the projected area of ​​the continuous band shear deformation zone identified in step S6 in three-dimensional space to the area of ​​the potential sliding region divided in step S1. I s The stress ratio factor is the maximum shear stress value obtained from the inversion in step S4 within the anomalous region. F s Combined shear strength provided by the soil and the sensing geogrid F t The ratio, i.e. I s = F s / F t ; I pThe water pressure influence factor is the average hydraulic gradient within the seepage-sensitive region. i ave Critical hydraulic gradient of soil i cr The ratio, i.e. I p = i ave / i cr ; I v The abnormal development rate factor is the average displacement increment along the direction of maximum deformation in the abnormal region during the current monitoring period. d m Monitoring time interval t The ratio; k 1 , k 2 and k 3 Let be the weight coefficients of each risk factor, and satisfy . k 1 + k 2 + k 3 =1, the specific value of the weight coefficient is determined by the analytic hierarchy process and combined with historical engineering case data, in order to objectively characterize the relative contribution of different risk factors to the overall stability of the slope; S8. Risk Warning: The real-time risk index calculated based on step S7 R The system activates a corresponding early warning response mechanism based on preset multi-level early warning thresholds; these thresholds are set according to the engineering safety level; when... R When the slope velocity is ≥0.7, a red alert is generated and issued, indicating that the slope faces high risk and has a high probability of instability; when 0.4 ≤ R When the slope velocity is less than 0.7, a yellow alert is generated and issued, indicating that the slope is in a medium-risk state and requires attention and enhanced monitoring; when... R When the value is less than 0.4, it is considered a safe state, and a blue safety alert can be issued or no warning can be generated.

[0010] Preferably, the sensing geogrid comprises a geogrid substrate, distributed optical fiber sensors, and a protective layer; the geogrid substrate is a bidirectional high-strength geogrid made of ultra-high molecular weight polyethylene after directional stretching; the distributed optical fiber sensors include strain sensors, displacement sensors, and water pressure sensors, which are encapsulated inside the geogrid substrate using a microgroove embedding process, enabling synchronous and distributed monitoring of strain, displacement, and groundwater pressure data at different locations within the slope; the protective layer is an anti-corrosion and anti-wear coating applied to the surface of the geogrid substrate and the area where the distributed optical fiber sensors are embedded.

[0011] Preferably, in step S3, the data acquisition frequency is dynamically adjusted according to external meteorological conditions: during the flood season, the sampling frequency is no less than once per hour; during the non-flood season, the sampling frequency is no less than four times per day; the external meteorological condition information is provided by the accessed local meteorological platform to ensure that the monitoring strategy can adapt to changes in the external environment and improve the timeliness and risk response capability of the data.

[0012] Preferably, in step S4, the physical field inversion process includes the following steps: S401. Initial Interpolation: Using the Kriging spatial interpolation algorithm, a continuous initial stress field and displacement field are generated based on the spatial location of the discrete sensor and its monitored values. D 0 and the initial water pressure field; S402. Constraint Check: Substitute the initial stress field into the Mohr-Coulomb strength criterion and determine whether the shear stress exceeds the shear strength for each element; substitute the initial water pressure field into Darcy's law and determine its rationality for each element. S403. Data Correction: For stress elements that do not meet the strength criteria, correct their stress values ​​to be within the yield surface; for water pressure elements that do not meet Darcy's law, make corresponding corrections and record the correction amount for each element. S404, Re-interpolation: Based on the corrected data points, re-perform the interpolation calculation and update the stress field. E 0 With water pressure field P 0 This yields physically consistent field distribution results.

[0013] Preferably, the structure of the neural network model includes, in sequence, an input layer, a feature extraction backbone network, an attention mechanism module, and an output layer; The input layer is used to receive standardized spatiotemporal data sequences; The feature extraction backbone network uses multi-layer three-dimensional convolutional modules to extract local features in the spatial dimension, and combines recurrent neural networks or temporal convolutional layers to capture dynamic dependencies in the time series. The attention mechanism module includes a channel attention module and a spatial attention module embedded in the backbone network; wherein, the channel attention module is used to adaptively weight the importance of different physical field features in the recognition task; and the spatial attention module is used to highlight abnormally active areas in space and enhance the model's ability to perceive key parts. The output layer adopts either a classification or segmentation form depending on the identification task. For classification tasks, the output layer uses a fully connected layer to output a category probability vector representing the overall abnormal state of the slope. For segmentation tasks, the output layer uses a deconvolution structure to output a three-dimensional grid with the same dimension as the input space, and each grid point outputs the probability of belonging to a specific abnormal category. The deep learning model is trained using supervised learning, and the training data is generated using the finite element numerical simulation method. Specifically, using the three-dimensional geomechanical model of the slope established in step S1, the mechanical response of the slope under different load conditions, seepage conditions, and preset abnormal conditions is simulated, and the corresponding stress field, displacement field, and water pressure field data are calculated as input samples, and the preset abnormal conditions are used as sample labels.

[0014] In summary, compared with existing technologies, the beneficial effects of this invention are as follows: To achieve intelligent early warning of instability risks in high-fill reinforced slopes, an artificial intelligence-based monitoring and early warning method for high-fill reinforced slopes is proposed. This monitoring and early warning method includes regional division of high-fill slopes, deployment of perceptual geogrids, monitoring of internal slope data, physical field inversion, construction of spatiotemporal datasets, identification of slope anomalies, risk assessment, and risk early warning, achieving the following breakthrough improvements: 1) Achieved three-dimensional perception of the internal condition of the slope: By deploying a sensing geogrid with integrated distributed optical fiber sensors in layers and zones in the high fill reinforced slope, the shortcomings of traditional point monitoring methods, such as limited coverage and blind spots, are overcome. It can synchronously and continuously acquire multi-physical field data such as strain, displacement and water pressure at different locations and depths inside the slope, providing a comprehensive and real-time data foundation for accurately assessing the health status of the slope. 2) Improved accuracy of physical field reconstruction: The inversion method combining spatial interpolation with soil mechanics and seepage theory physical constraints transforms discrete sensor data into continuous and complete stress field, displacement field and water pressure field. Through iterative correction, it ensures that the inversion results conform to physical laws such as the Mohr-Coulomb strength criterion and Darcy's law, effectively eliminating inversion distortion caused by data noise or sparsity, and generating a more realistic three-dimensional physical field model that reflects the internal mechanical state of the slope. 3) Intelligent identification and early warning of slope anomalies: By constructing standardized spatiotemporal data sequences and using a deep learning model that integrates three-dimensional convolution and attention mechanisms, key anomaly patterns such as continuous banded shear deformation zones, soil arching failure, reinforcement creep or fracture, and potential slip surfaces are automatically and intelligently identified. This effectively captures the subtle features of multi-physics fields in spatiotemporal evolution, significantly improves the ability to identify potential instability precursors, and changes the passive situation of traditional methods that rely on a single threshold and have delayed early warnings. 4) A quantitative risk assessment and multi-level early warning system has been established: The real-time risk index proposed in this invention integrates multiple key factors such as the scale of abnormal areas, stress ratio, water pressure influence, and abnormal development rate, and scientifically determines the weights through the analytic hierarchy process, realizing a dynamic and quantitative assessment of the slope safety status; the multi-level early warning thresholds set based on this index can activate differentiated response mechanisms according to the risk level, and the early warning information is detailed and clear, providing clear decision support for engineering management personnel and greatly improving the scientific nature of risk management and the efficiency of emergency response. Attached Figure Description

[0015] Figure 1 This is a flowchart of the monitoring and early warning method for reinforced high embankment slopes based on artificial intelligence, as described in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of the sensing geogrid according to an embodiment of the present invention; Among them, 1-geogrid substrate, 2-distributed optical fiber sensor. Detailed Implementation

[0016] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments given herein are for illustration and explanation only and are not intended to limit the present invention.

[0017] This application discloses, as follows: Figure 1-2 The method for monitoring and early warning of reinforced high-fill slopes based on artificial intelligence is shown below. A monitoring and early warning method for reinforced high-fill slopes based on artificial intelligence includes the following steps: S1. Regional Division of High Fill Slopes: Based on the engineering geological and hydrogeological conditions of the reinforced high fill slope, a three-dimensional model of the slope is established; the safety factor of the slope is calculated using the finite element strength reduction method, and the hydraulic gradient distribution inside the slope is determined based on seepage finite element analysis; according to the calculation results, the high fill slope is divided into potential sliding zone, seepage sensitive zone, and low-risk zone. Among them, the potential sliding zone is defined as the area with a safety factor lower than 1.25, the seepage sensitive zone is defined as the area with a hydraulic gradient greater than the critical hydraulic gradient, and the remaining areas are classified as low-risk zones.

[0018] S2. Layout of Sensing Geogrid: During the construction of high embankment slopes, reinforcement layers are laid in layers according to the design elevation to form a high embankment reinforced slope. Differentiated sensing geogrid layout strategies are adopted for different risk areas defined in step S1: In potential sliding areas and seepage-sensitive areas, the sensing geogrid is laid in each reinforcement layer; in low-risk areas, a layer of sensing geogrid is laid every 2 to 4 reinforcement layers. The sensing geogrid, while possessing the mechanical properties of conventional geogrids, can also distributedly monitor strain, displacement, and groundwater pressure data at different locations within the slope. In specific implementation, the sensing geogrid consists of a geogrid substrate 1, distributed optical fiber sensors 2, and a protective layer. The geogrid substrate 1 is a bidirectional high-strength geogrid made of ultra-high molecular weight polyethylene after directional stretching to meet the mechanical performance requirements of high-fill reinforced slopes. The distributed optical fiber sensors 2 include strain sensors, displacement sensors, and water pressure sensors, which are encapsulated inside the geogrid substrate 1 using a microgroove embedding process, enabling synchronous and distributed monitoring of strain, displacement, and groundwater pressure data at different locations inside the slope. The protective layer is an anti-corrosion and anti-wear coating applied to the surface of the geogrid substrate 1 and the embedded area of ​​the distributed optical fiber sensors 2. It is made of ethylene-vinyl acetate copolymer material, whose elastic modulus matches that of the geogrid substrate 1, to avoid additional shear stress on the distributed optical fiber sensors 2 when the soil deforms, ensuring the reliability of the monitoring data and the long-term durability of the sensors.

[0019] S3. Monitoring of internal slope data: Real-time acquisition of multi-physics data within the slope is achieved using the layered geogrid deployed in step S2. This data includes strain at different locations and depths along the geogrid's installation path. e ( x , y , z , t ), displacement d ( x , y , z , t ) and water pressure p ( x , y , z , t ),in x , y , z Represents spatial coordinates, tThe time variable is represented; in specific implementation, the data collection frequency is dynamically adjusted according to external meteorological conditions: during the flood season, the sampling frequency is no less than once per hour; during the non-flood season, the sampling frequency is no less than four times per day; the external meteorological condition information is provided by the local meteorological platform to ensure that the monitoring strategy can adapt to changes in the external environment and improve the timeliness and risk response capability of the data.

[0020] S4. Inversion of the physical field: Based on the discrete monitoring data of the slope interior collected in step S3, i.e., strain... e ( x , y , z , t ), displacement d ( x , y , z , t ) and water pressure p ( x , y , z , t The continuous stress field was obtained by inversion using spatial interpolation methods. E 0 Displacement field D 0 and water pressure field P 0 In the physical field inversion process, soil mechanics and seepage theory are introduced as physical constraints. Specifically, stress field inversion requires the use of the Mohr-Coulomb constitutive model of the slope soil to ensure that the principal stress value at any spatial point does not exceed the soil shear strength. Water pressure field inversion must satisfy Darcy's law to maintain the rationality of the seepage field. Through iterative correction of the interpolation results, inversion distortions caused by discrete data noise or sparsity are eliminated, ultimately outputting a continuous physical field distribution that conforms to the laws of mechanics and seepage. In specific implementation, the physical field inversion process includes the following steps: S401. Initial Interpolation: Using the Kriging spatial interpolation algorithm, a continuous initial stress field and displacement field are generated based on the spatial location of the discrete sensor and its monitored values. D 0 and the initial water pressure field; S402. Constraint Check: Substitute the initial stress field into the Mohr-Coulomb strength criterion and determine whether the shear stress exceeds the shear strength for each element; substitute the initial water pressure field into Darcy's law and determine its rationality for each element. S403. Data Correction: For stress elements that do not meet the strength criteria, correct their stress values ​​to be within the yield surface; for water pressure elements that do not meet Darcy's law, make corresponding corrections and record the correction amount for each element. S404, Re-interpolation: Based on the corrected data points, re-perform the interpolation calculation and update the stress field. E 0 With water pressure field P 0 This yields physically consistent field distribution results.

[0021] S5. Construction of the spatiotemporal dataset: The stress field, displacement field, and water pressure field obtained from the inversion in step S4, which conform to physical laws, are preprocessed by normalization to uniformly scale the values ​​of each physical field to the [0,1] interval; the normalization process is performed according to the following expression: (1) in, X The physical field data to be normalized X max and X min Data respectively X The global maximum and global minimum values; The three-dimensional model of the slope is discretized into a regular three-dimensional mesh system, and then the normalized stress field is... E Displacement field D and water pressure field P The data is resampled to this grid node to ensure that each spatiotemporal point ( x , y , z , t Each of these corresponds to a set of standardized physical field values. e , d , p This allows for the construction of standardized spatiotemporal data sequences. x , y , z , t , e , d , p This provides structured input data for subsequent deep learning models.

[0022] S6. Slope anomaly identification: Based on the spatiotemporal data sequence constructed in step S5 ( x , y , z , t , e , d , pThe method employs a specially trained deep learning model to intelligently identify abnormal states within the slope. This deep learning model is a neural network adapted to spatiotemporal data structures, integrating three-dimensional convolution operations and attention mechanisms. It can effectively capture the characteristics of abnormal states in the spatial distribution and temporal evolution of multiple physical fields within the slope. These abnormal states include: continuous banded shear deformation zones appearing in the displacement field; signs of soil arching failure reflected in the stress field; abrupt changes in strain or displacement at the reinforcement location, indicating accelerated creep or fracture risk of the reinforcement; and potential slip surfaces revealed by the coupling effect of the water pressure field and the displacement field.

[0023] In specific implementation, the structure of the neural network model includes, in sequence, an input layer, a feature extraction backbone network, an attention mechanism module, and an output layer; The input layer is used to receive standardized spatiotemporal data sequences; The feature extraction backbone network uses multi-layer three-dimensional convolutional modules to extract local features in the spatial dimension, and combines recurrent neural networks or temporal convolutional layers to capture dynamic dependencies in the time series. The attention mechanism module includes a channel attention module and a spatial attention module embedded in the backbone network; wherein, the channel attention module is used to adaptively weight the importance of different physical field features in the recognition task; and the spatial attention module is used to highlight abnormally active areas in space and enhance the model's ability to perceive key parts. The output layer adopts either a classification or segmentation form depending on the identification task. For classification tasks, the output layer uses a fully connected layer to output a category probability vector representing the overall abnormal state of the slope. For segmentation tasks, the output layer uses a deconvolution structure to output a three-dimensional grid with the same dimension as the input space, and each grid point outputs the probability of belonging to a specific abnormal category. The deep learning model is trained using supervised learning, and the training data is generated using the finite element numerical simulation method. Specifically, using the three-dimensional geomechanical model of the slope established in step S1, the mechanical response of the slope under different load conditions, seepage conditions, and preset abnormal conditions is simulated, and the corresponding stress field, displacement field, and water pressure field data are calculated as input samples, and the preset abnormal conditions are used as sample labels.

[0024] S7. Risk Assessment: Based on the anomaly patterns and characteristic parameters of the slope identified in Step 6, a real-time risk index is used. R A quantitative assessment of the slope safety status is conducted; the real-time risk index... R Calculated using the following multi-factor weighted model; (2) in, I aThe abnormal region size factor is the ratio of the projected area of ​​the continuous band shear deformation zone identified in step S6 in three-dimensional space to the area of ​​the potential sliding region divided in step S1. I s The stress ratio factor is the maximum shear stress value obtained from the inversion in step S4 within the anomalous region. F s Combined shear strength provided by the soil and the sensing geogrid F t The ratio, i.e. I s = F s / F t ; I p The water pressure influence factor is the average hydraulic gradient within the seepage-sensitive region. i ave Critical hydraulic gradient of soil i cr The ratio, i.e. I p = i ave / i cr ; I v The abnormal development rate factor is the average displacement increment along the direction of maximum deformation in the abnormal region during the current monitoring period. d m Monitoring time interval t The ratio; k 1 , k 2 and k 3 Let be the weight coefficients of each risk factor, and satisfy . k 1 + k 2 + k 3 =1, the specific value of the weight coefficient is determined by the analytic hierarchy process and combined with historical engineering case data, in order to objectively characterize the relative contribution of different risk factors to the overall stability of the slope.

[0025] S8. Risk Warning: Real-time risk index calculated based on step S7. R The system activates a corresponding early warning response mechanism based on preset multi-level early warning thresholds; these thresholds are set according to the engineering safety level; when... R When the slope velocity is ≥0.7, a red alert is generated and issued, indicating that the slope faces high risk and has a high probability of instability; when 0.4 ≤R When the slope velocity is less than 0.7, a yellow alert is generated and issued, indicating that the slope is in a medium-risk state and requires attention and enhanced monitoring; when... R When the risk level is less than 0.4, the situation is considered safe, and a blue safety alert may be issued or no warning may be generated. Warning information is sent to engineering management personnel in real time through various means, including audible and visual alarms, SMS, email, and push notifications from the monitoring center platform. The warning information includes the current warning level and the real-time risk index. R The numerical values, spatial coordinates of the main abnormal areas, identified abnormal pattern types, and recommendations for targeted handling measures.

[0026] The above describes one or more embodiments of the present invention in a relatively specific and detailed manner, but it should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. An artificial intelligence-based high fill reinforced slope monitoring and early warning method, characterized in that, The method comprises the following steps: S1, region division of high fill slope: based on the engineering geological conditions and hydrogeological conditions of the high fill reinforced slope, a three-dimensional model of the slope is established; The safety factor of the slope is calculated by using the finite element strength reduction method, and the hydraulic gradient distribution inside the slope is determined based on the seepage finite element analysis; according to the calculation results, the high fill slope is divided into a potential sliding area, a seepage sensitive area and a low risk area: the potential sliding area is defined as an area with a safety factor less than 1.25, the seepage sensitive area is defined as an area with a hydraulic gradient greater than the critical hydraulic gradient, and the remaining area is divided into a low risk area; S2, layout of the sensing geogrid: during the construction process of the high fill slope, the reinforced layer is layered according to the design elevation to form the high fill reinforced slope; for different risk areas divided in step S1, a differentiated sensing geogrid layout strategy is adopted: in the potential sliding area and the seepage sensitive area, the sensing geogrid is uniformly laid in each reinforced layer; in the low risk area, a layer of sensing geogrid is laid every 2 to 4 layers of reinforced layer; the sensing geogrid has the mechanical properties of a conventional geogrid and can monitor strain, displacement and groundwater pressure data at different positions inside the slope in a distributed manner; S3, monitoring of internal data of the slope: through the perception type geogrid laid in step S2, the multi-physical field data inside the slope is collected in real time; the data includes strain e ( x , y , z , t ) of different positions and different depths along the laying path of the perception type geogrid d ( x , y , z , t ) and water pressure p ( x , y , z , t ) of different positions and different depths along the laying path of the perception type geogrid, wherein x , y , z represents a spatial coordinate, t represents a time variable; S4, inversion of physical fields: the continuous stress field E 0 , displacement field D 0 and water pressure field P 0 are respectively obtained by spatial interpolation method from the discrete monitoring data of internal stress e ( x , y , z , t ), displacement d ( x , y , z , t ) and water pressure p ( x , y , z , t ) collected in step S3; in the process of inversion of physical fields, the soil mechanics and seepage theory are introduced as physical constraints, among which the stress field inversion needs to be combined with the Mohr-Coulomb constitutive model of the slope soil to ensure that the principal stress value of any spatial point does not exceed the shear strength of the soil; the water pressure field inversion needs to satisfy Darcy's law to maintain the reasonableness of the seepage field; the interpolation results are iteratively corrected to eliminate the inversion distortion caused by the noise or sparsity of discrete data; S5, construction of spatiotemporal data set: the stress field, displacement field and water pressure field obtained by inversion in step S4 are normalized and pretreated to unify the numerical values of each physical field to the interval [0, 1]; the normalization processing is carried out according to the following expression: wherein, X represents the physical field data to be normalized, X max and X min are the global maximum and the global minimum of the data X respectively. The three-dimensional model space of the slope is discretized into a regular three-dimensional grid system, and then the data of the normalized stress field E , displacement field D and water pressure field P are resampled to the grid nodes, so that each space-time point x , y , z , t corresponds to a set of standard physical field values e , d , p , thereby forming a standardized space-time data sequence x , y , z , t , e , d , p , which provides structured input data for subsequent deep learning models; S6, slope abnormal state recognition: according to the spatiotemporal data sequence constructed in step S5, a deep learning model specially trained is used to intelligently recognize the abnormal state inside the slope; the deep learning model is a neural network adapted to the spatiotemporal data structure, which combines three-dimensional convolution operation and attention mechanism, and can effectively capture the features of the abnormal state of the spatial distribution and time evolution of the multi-physical field inside the slope; the abnormal state includes: a continuous band-shaped shear deformation zone appearing in the displacement field; failure signs of soil arching effect reflected in the stress field; a sudden change zone of strain or displacement at the position of the reinforcement material, indicating the risk of intensified creep or fracture of the reinforcement material; and a potential sliding surface displayed by the coupling of the water pressure field and the displacement field; S7, risk assessment: according to the abnormal pattern inside the slope and its characteristic parameters identified in step 6, a real-time risk index is adopted R to quantitatively evaluate the safety state of the slope; the real-time risk index R is calculated by the following multi-factor weighted model; wherein, I a is an abnormal region scale factor, whose value is the ratio of the projected area of the continuous band-like shear deformation zone identified in step S6 in three-dimensional space to the area of the potential sliding region divided in step S1; I s is a stress ratio factor, whose value is the ratio of the maximum shear stress value in the abnormal region obtained from the inversion in step S4 F s to the comprehensive shear strength provided by the soil and the perceived geogrid at this location F t , i.e. I s = F s / F t ; I p is a water pressure influence factor, whose value is the ratio of the average hydraulic gradient in the seepage-sensitive region i ave to the critical hydraulic gradient of the soil i cr , i.e. I p = i ave / i cr ; I v is an abnormal development rate factor, whose value is the ratio of the average displacement increment of the abnormal region in the maximum deformation direction in the current monitoring period d m to the monitoring time interval t ; k 1 、 k 2 and k 3 are weight coefficients of the respective risk factors, and satisfy k 1 + k 2 + k 3 =1, the specific values of the weight coefficients are determined by the analytic hierarchy process and combined with historical engineering case data to objectively represent the relative contribution of different risk factors to the overall stability of the slope; S8, risk early warning: based on the real-time risk index calculated in step S7 R , according to the preset multi-level early warning threshold, the corresponding early warning response mechanism is started; the early warning threshold is set according to the engineering safety level; when R ≥ 0.7, a red early warning is generated and released, indicating that the slope is at high risk and has high possibility of instability; when 0.4 ≤ R < 0.7, yellow early warning information is generated and released, indicating that the slope is in a medium risk state, which needs to be paid attention to and the monitoring needs to be strengthened; when R < 0.4, it is determined as a safe state, and a blue safety prompt can be released or no early warning is generated.

2. The high fill reinforced slope monitoring and early warning method based on artificial intelligence according to claim 1, characterized in that, The sensing geogrid is composed of a geogrid base material, a distributed optical fiber sensor and a protective layer; the geogrid base material is a bidirectional high-strength geogrid made of ultrahigh molecular weight polyethylene material after directional stretching; the distributed optical fiber sensor includes a strain sensor, a displacement sensor and a water pressure sensor, which are packaged inside the geogrid base material through a micro-slot embedding process, and can synchronously and distributively monitor strain, displacement and groundwater pressure data at different positions inside the slope; the protective layer is an anti-corrosion and anti-wear coating coated on the surface of the geogrid base material and the embedded area of the distributed optical fiber sensor. 3.The high fill reinforced slope monitoring and early warning method based on artificial intelligence according to claim 1, characterized in that, In step S3, the data collection frequency is dynamically adjusted according to external meteorological conditions: in the flood season, the sampling frequency is not less than 1 time per hour; in the non-flood season, the sampling frequency is not less than 4 times per day; the external meteorological condition information is provided by the accessed local meteorological platform, so as to ensure that the monitoring strategy can adapt to the external environmental changes and improve the timeliness and risk response ability of the data. 4.The method of claim 1, wherein, In step S4, the physical field inversion process includes the following steps: S401、initial interpolation: using the Kriging spatial interpolation algorithm, according to the spatial position of the discrete sensor and its monitoring value, the continuous initial stress field, displacement field and initial water pressure field are generated D 0 and initial water pressure field; S402, constraint check: the initial stress field is substituted into the Mohr-Coulomb strength criterion to judge whether the shear stress exceeds the shear strength on a unit-by-unit basis; the initial water pressure field is substituted into Darcy's law to judge its rationality on a unit-by-unit basis; S403, data correction: for stress units that do not meet the strength criterion, the stress value is corrected to be within the yield surface; for water pressure units that do not meet Darcy's law, appropriate correction is made, and the correction amount of each unit is recorded; S404, re-interpolation: based on the corrected data points, re-perform the interpolation calculation and update the stress field E 0 with the water pressure field P 0 , and obtain a physically consistent field distribution result. 5.The high fill reinforced slope monitoring and early warning method based on artificial intelligence according to claim 1, characterized in that, The structure of the neural network model includes an input layer, a feature extraction backbone network, an attention mechanism module, and an output layer in sequence; the input layer is used to receive the standardized spatio-temporal data sequence; the feature extraction backbone network uses a multi-layer three-dimensional convolution module to extract local features in the spatial dimension, and combines a recurrent neural network or a time convolution layer to capture dynamic dependencies in the time series; the attention mechanism module includes a channel attention module and a spatial attention module embedded in the backbone network; the channel attention module is used to adaptively weight the importance of different physical field features in the identification task; the spatial attention module is used to highlight the abnormally active areas in the space and enhance the perception ability of the model to key parts; the output layer takes a classification form or a segmentation form according to different identification tasks; for the classification task, the output layer uses a fully connected layer to output a class probability vector representing the overall abnormal state of the slope; for the segmentation task, the output layer uses a deconvolution structure to output a three-dimensional grid consistent with the input spatial dimension, and each grid point outputs the probability that it belongs to a specific abnormal class; the deep learning model is trained through a supervised learning method, and the training data is generated by a finite element numerical simulation method.

Images