Slope non-destructive monitoring and early warning system based on micro-motion sensing

Abstract

This invention relates to the field of non-destructive slope monitoring technology, specifically to a non-destructive slope monitoring and early warning system based on micro-motion sensing. This system precisely couples a dynamic model of the slope geological body, a simulation of the micro-motion sensing system, and a dynamic model of environmental disturbances. Through a closed loop of "interference correction-signal feedback," it incorporates actual disturbances such as wind vibration and traffic vibration, constructing a full-condition associated simulation scenario covering "daily stability - mild risk - extreme risk." This solves the problem of early warning deviations caused by traditional monitoring being detached from the actual environment and having a single scenario. The state reward function improves the comprehensiveness and reliability of monitoring accuracy assessment. The scheme decision-making mechanism based on the improved NSGA-II algorithm ensures the scientific nature and engineering feasibility of the monitoring scheme.

Description

Technical Field

[0001] This invention relates to the field of non-destructive slope monitoring technology, and more specifically, to a non-destructive slope monitoring and early warning system based on micro-motion sensing. Background Technology

[0002] In the field of non-destructive slope monitoring, traditional monitoring systems generally suffer from a core problem: the simulation scenario is disconnected from the actual environment. These systems often struggle to effectively coordinate and couple the dynamic evolution of the slope geological body, the full-process monitoring behavior of micro-motion sensors, and complex on-site disturbances (such as natural wind vibration, traffic vibration, temperature fluctuations, and inherent noise of the sensor system). They are mostly based on ideal, interference-free environments or single, fixed operating conditions to construct simulation models. This singular scenario design cannot comprehensively cover the full range of slope conditions, from normal stability to extreme risks, nor can it realistically simulate the impact of disturbance factors on slope deformation and monitoring signals. Ultimately, this leads to discrepancies between early warning results and actual engineering conditions, making it difficult to meet the needs of accurate monitoring under complex geological conditions.

[0003] Traditional methods for assessing the accuracy of slope monitoring have significant limitations. They often rely on single error indicators to quantify monitoring effectiveness, failing to match the actual pattern of "multi-feature coordinated abrupt changes" before slope instability. These methods typically focus only on the estimation error of the overall slope state (such as displacement rate and safety factor), neglecting the local error analysis of micro-motion characteristics (such as time-domain kurtosis, frequency-domain dominant frequency, and time-frequency domain energy distribution) that play a crucial indicative role in instability early warning. This single-dimensional accuracy assessment cannot prioritize the accuracy of highly sensitive micro-motion characteristics, nor can it effectively determine the "accuracy level" through gradient differentiation. The one-sidedness of the indicators easily leads to missed or false alarms of instability precursors, significantly reducing the early warning reliability of the monitoring system.

[0004] Based on the above, this invention proposes a non-destructive monitoring and early warning system for slopes based on micro-motion sensing. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a non-destructive monitoring and early warning system for slopes based on micro-motion sensing.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A non-destructive slope monitoring and early warning system based on micro-motion sensing includes,

[0008] The slope non-destructive monitoring coupling module is used to coordinate and couple the dynamic model unit of the slope geological body, the simulation unit of the micro-motion sensing system, and the dynamic model unit of environmental disturbance to construct a slope non-destructive monitoring associated simulation scenario.

[0009] The slope micro-motion sensing and conversion module is used to define the slope state space and micro-motion feature mapping space based on the slope non-destructive monitoring associated simulation scenario.

[0010] The slope non-destructive monitoring enhancement decision design module is used to define the action space of slope non-destructive monitoring and design the state reward function. Constrained reward function Further determine the two-dimensional dynamic reward function ;

[0011] The slope non-destructive monitoring scheme decision module generates non-destructive monitoring early warning decision schemes based on the Pareto optimal solution set of the improved NSGA-II algorithm.

[0012] Furthermore, the dynamic model unit for slope geological bodies is used to construct the dynamic model of the slope geological body. The specific construction process is as follows:

[0013] Step 1: Collect basic geological data of the slope and test the mechanical parameters of the soil and rock mass of the slope. Then, fuse the multi-source data obtained from the collection and testing.

[0014] Step 2: Construct a three-dimensional geometric model of the slope based on the fused multi-source data, and then mesh the three-dimensional geometric model of the slope.

[0015] Step 3: Select the appropriate constitutive model according to the lithological units in the three-dimensional geometric model of the slope, and assign the mechanical parameters to the constitutive model of each lithological unit in space.

[0016] Step 4: For the three-dimensional geometric model of the slope endowed with mechanical properties, first set the initial conditions and boundary conditions, and then apply dynamic loads;

[0017] Step 5: Based on the three-dimensional geometric model of the slope in Step 2, the constitutive parameters in Step 3, and the boundary conditions and dynamic loads in Step 4, multi-physics coupling processing is carried out. The multi-field response of the slope under dynamic load is calculated through numerical simulation, and the simulation accuracy is verified by combining field monitoring data. Finally, a dynamic model of the slope geological body that conforms to the actual evolution law is constructed.

[0018] Furthermore, the micro-motion sensing system simulation unit is used to simulate the full-process monitoring behavior of micro-motion sensors on slopes;

[0019] The environmental disturbance dynamic model unit is used to construct the environmental disturbance dynamic model.

[0020] Furthermore, the slope state space includes a geomechanical foundational subspace. Multiphysics Dynamic Response Subspace Stability level subspace Environmental interference coupling subspace .

[0021] Furthermore, the micro-motion feature mapping space includes temporal features. Frequency domain characteristics Time-frequency domain characteristics Spatial features .

[0022] Furthermore, the action space for non-destructive slope monitoring includes the following action dimensions: sensor deployment and adjustment. Sampling strategy adjustment , Transmission parameter adjustment Signal preprocessing adjustment .

[0023] Furthermore, the state reward function ,in, For feature error weights; This is the error attenuation coefficient; The slope condition is inferred from the micro-motion characteristics; This represents the actual state of the slope; The mean square error of the state estimation; This is an estimate of highly sensitive micro-motion characteristics. This represents the true value of highly sensitive micro-motion characteristics.

[0024] Furthermore, constrain the reward function ;in, These are lossless weights; The soil and rock disturbance index; The threshold value is non-destructive. The total cost of the monitoring system; This is the upper limit of the budget.

[0025] Furthermore, the two-dimensional dynamic reward function ;in, The dynamic weighting factor is determined by the stability index SI, and the formula is: .

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] This invention's system precisely couples a dynamic model of the slope geological body, a simulation of a micro-motion sensing system, and a dynamic model of environmental disturbances. Through a closed-loop "interference correction-signal feedback" mechanism, it incorporates actual disturbances such as wind vibration and traffic vibration, constructing a full-condition associated simulation scenario covering "daily stability - mild risk - extreme risk." This solves the problem of early warning bias caused by traditional monitoring being detached from the actual environment and having a single scenario. The state reward function improves the comprehensiveness and reliability of monitoring accuracy assessment. This function overcomes the limitations of a single error index, co-quantifying the overall deviation of slope state estimation with the local errors of highly sensitive micro-motion features. It prioritizes the accuracy of key highly sensitive features for instability early warning through feature error weights, while using an error attenuation coefficient to achieve gradient differentiation of accuracy, effectively avoiding early warning bias caused by misjudgment of a single index. This makes the monitoring accuracy assessment more closely aligned with the actual law of "multi-feature coordinated abrupt change" before slope instability. The constrained reward function achieves non-destructive and economical compliant management. Its core lies in transforming the qualitative requirement of "non-destructive slope" into a quantifiable soil and rock disturbance index, which is then incorporated into the constraint framework along with the cost of the monitoring system. By prioritizing the integrity of the soil and rock mass through non-destructive weighting, it avoids slope structural damage caused by excessive cost control, effectively selecting compliant solutions with "low disturbance and low cost," balancing the safety and feasibility of the monitoring solutions, and ensuring the scientific nature and engineering feasibility of the monitoring solutions based on the improved NSGA-II algorithm decision-making mechanism. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of a non-destructive monitoring and early warning system for slopes based on micro-motion sensing.

[0029] Figure 2 A flowchart illustrating the construction process of a dynamic model of a slope geological body;

[0030] Figure 3 This is a schematic diagram of the simulation unit for the micro-motion sensing system.

[0031] Figure 4 A flowchart for constructing a dynamic model of environmental disturbances.

[0032] Reference Figures 1 to 4 A non-destructive slope monitoring and early warning system based on micro-motion sensing includes,

[0033] The slope non-destructive monitoring coupling module is used to coordinate and couple the dynamic model unit of the slope geological body, the simulation unit of the micro-motion sensing system, and the dynamic model unit of environmental disturbance to construct a related simulation scenario for slope non-destructive monitoring.

[0034] The dynamic model element for slope geological bodies is used to construct a dynamic model of slope geological bodies. The specific construction process is as follows:

[0035] Step 1: Collect basic geological data of the slope and test the mechanical parameters of the soil and rock mass of the slope. Then, fuse the multi-source data obtained from the collection and testing.

[0036] Basic geological data includes topographic data, geological structure data, and hydrogeological data. Topographic data: Three-dimensional point clouds of slopes are acquired using UAV oblique photography to generate digital elevation models and orthophotos, identifying macroscopic morphological parameters such as slope ratio, slope height, and platform width. For complex areas (such as steep cliffs and gullies), ground-based LiDAR scanning is supplemented to improve local topographic accuracy. Geological structure data: Joints, faults, lithological interfaces, and other discontinuities are marked through surface geological mapping, recording their strike, dip angle, spacing, and infill material properties. Combined with borehole core logging (core taken every 50cm) and ground-penetrating radar detection, lithological stratification and structural distribution within a depth of 20-50m are determined, and vertical profiles are drawn. Hydrogeological data: Groundwater monitoring wells are deployed (well depth ≥ potential slip surface depth of the slope) to monitor dynamic changes in groundwater levels. The permeability coefficient of soil and rock is determined through pressure water tests (distinguishing between vertical and horizontal permeability coefficients). Regional rainfall data is collected to determine hydrological load boundaries.

[0037] Mechanical parameter tests were conducted on the soil and rock mass of the slope, specifically as follows: representative rock / soil samples were collected from the slope, and triaxial compression tests (to determine cohesion c, internal friction angle φ, and elastic modulus E), direct shear tests (to simulate the shear characteristics of the sliding surface), and disintegration tests (to evaluate the stability of weathered rock when exposed to water) were performed; compaction tests were used on the loose soil to determine the degree of compaction and optimum moisture content, providing a basis for parameter assignment.

[0038] The multi-source data obtained from collection and testing are fused and processed, including:

[0039] Data standardization and unification: First, multi-source data are incorporated into a unified benchmark framework—a coordinate system. Using tools such as ArcGIS, UAV point clouds, ground-penetrating radar data (local coordinate system), and borehole data (construction coordinate system) are uniformly converted to the CGCS2000 geodetic coordinate system to ensure spatial alignment. In terms of numerical units, mechanical parameters, hydrological parameters, etc., are compressed to the [0,1] range using the "Min-Max normalization method" to avoid the impact of unit differences on subsequent analysis. In terms of data format, unstructured data such as point clouds and images are converted to LAS and TIFF formats, while mechanical test curves and water level time series are converted to Excel and CSV structured formats for easy unified access.

[0040] Data cleaning and completion: To address data redundancy and missing values, statistical and spatial methods were employed. For outlier removal, Grubbs' test was used to identify strength anomalies caused by sample disturbance in triaxial tests and sudden water level jumps due to equipment malfunctions in hydrological monitoring, which were then replaced with the mean of the same batch of data. For missing value completion, spatial data (such as missing lithological stratification between boreholes) were processed using Kriging interpolation (based on lithological continuity constraints) to generate continuous lithological distribution surfaces; time-series data (such as missing rainfall data) were supplemented using linear interpolation; and mechanical parameters (such as unmeasured compaction of some soil samples) were estimated by establishing a "lithology-compaction degree" regression model to ensure data integrity.

[0041] Spatial Correlation and Mapping: Establishing a spatial correspondence between "topography-geology-parameters"—using a digital elevation model generated by UAVs as the base map, overlaying underground structural surfaces (such as faults and joints) detected by ground-penetrating radar with vertical profiles drawn from boreholes to form a three-dimensional geological framework; associating the geotechnical mechanical parameters (such as c, φ, and E values ​​at different depths) and hydrological parameters (such as the permeability coefficient and groundwater level of each monitoring well) obtained from boreholes according to "depth-spatial location" and assigning them to the corresponding grid cells of the three-dimensional geological framework, realizing a one-to-one mapping of "spatial coordinates → lithology type → mechanical parameters → hydrological properties"; for example, the cohesion of 25 kPa and the permeability coefficient of 1e-6 m / s measured by boreholes at the toe of the slope are accurately matched to the grid at the corresponding depth of the toe of the slope in the three-dimensional model, ensuring that the spatial distribution of parameters is consistent with the actual geological conditions.

[0042] Step 2: Construct a three-dimensional geometric model of the slope based on the fused multi-source data, and then mesh the three-dimensional geometric model of the slope.

[0043] A three-dimensional geometric model of the slope is constructed based on fused multi-source data, as detailed below:

[0044] Using the digital elevation model (DEM) in the CGCS2000 coordinate system generated in step one as the base map, a three-dimensional basic framework is built in professional geological modeling software: Determine the spatial range of the model: in the horizontal direction, it extends 2-3 times the slope height beyond the top / toe of the slope (to avoid boundary effects), and in the vertical direction, it extends 5-10m below the potential slip surface (to ensure that it is covered by stable rock strata); Import and merge data: import the lithological distribution surface, fault / joint coordinates, borehole vertical profiles and other data from step one into the software as the constraint boundary for geometric modeling, forming a preliminary three-dimensional framework of "surface topography-subsurface structure". Following the logic of "macroscopic lithology first, then fine structure," geological elements are transformed into geometric entities: Lithological unit modeling: Based on the continuous lithological distribution surface generated by kriging interpolation in step one, different lithological entity blocks (such as moderately weathered sandstone, strongly weathered mudstone, and residual soil) are divided within a three-dimensional framework to ensure that the spatial boundaries of each lithological unit are consistent with the actual lithological interfaces revealed by the borehole; Structural surface modeling: Input the joint and fault parameters (strike, dip, and spacing) recorded in step one to generate a three-dimensional structural interface; For weak interlayers with a thickness ≥0.5m, thin-layer entity units are constructed separately (to avoid merging with surrounding lithology and causing accuracy loss) to accurately restore their spatial distribution. Combining the actual geological conditions of the slope and the requirements of numerical simulation, the boundary morphology of the three-dimensional model is clarified.

[0045] The 3D geometric model of the slope is meshed as follows: Based on the complexity of the geometric model and the simulation accuracy requirements, meshes are designed and generated for different regions: Mesh type selection: Hexahedral structured meshes are used in homogeneous lithological areas (high computational efficiency and accurate stress transfer); tetrahedral unstructured meshes are used in areas with complex joints / faults (adapting to irregular structural surfaces); pyramidal meshes are used to connect the two mesh types in transitional areas to avoid distortion; Mesh density control: The mesh size is refined to 0.5-2m in critical areas (potential slip surfaces, stress concentration areas at the toe of the slope, and weak interlayers) (to ensure the capture of stress / displacement abrupt changes), and the mesh size is set to 5-10m in non-critical areas (stable rock areas inside the slope) (to reduce computational load); Automatic mesh generation: The initial mesh is generated based on the above scheme using the software mesh generation function, and the lithological units from step one are associated with it—mesh units within the same lithological unit are automatically grouped together for easy subsequent parameter assignment.

[0046] Step 3: Select the appropriate constitutive model according to the lithological units in the three-dimensional geometric model of the slope, and assign the mechanical parameters to the constitutive model of each lithological unit in space.

[0047] Select the appropriate constitutive model based on the lithological elements in the 3D geometric model of the slope, as follows:

[0048] Based on the lithological zoning of the three-dimensional geometric model in step two, and combined with the geological exploration data from step one, the lithological units are further classified and their deformation characteristics are clarified, providing a basis for subsequent constitutive model selection: Classification basis: Based on the "lithology-depth" distribution data generated in step one, the lithological units in the three-dimensional geometric model are divided into four core types: intact hard rock (e.g., moderately weathered granite), jointed rock (e.g., fractured sandstone), weathered soft rock / residual soil (e.g., weathered mudstone layer, residual clay), and weak interlayers (e.g., argillaceous interlayers, fault gouge). Characteristic analysis: For each type of unit, the deformation characteristics are analyzed based on the experimental data (disintegration test, triaxial test curves) from step one—for example, intact hard rock only undergoes small-scale elastic deformation without obvious plastic yielding; jointed rock is controlled by joint surfaces and is prone to shear slip; weathered soft rock / residual soil exhibits dilatation and pore water pressure sensitivity; and weak interlayers exhibit large deformation and long-term creep characteristics. Based on the deformation characteristics of different lithological units, a constitutive model capable of accurately simulating their mechanical behavior is selected, and the core function and applicable boundaries of the model are clearly defined. When the lithological unit type is intact hard rock, a linear elastic model is selected as the appropriate constitutive model. The selection criteria and core function are that deformation is mainly elastic with no plastic yielding. It can simulate stress distribution and vibration response, and the input parameters are elastic modulus (E) and Poisson's ratio (ν). When the lithological unit type is jointed rock, a Mohr-Coulomb elastoplastic model is selected as the appropriate constitutive model. The selection criteria and core function are that there is obvious shear yielding and joint surface control instability. It can simulate plastic zone expansion and joint slip, and the input parameter is cohesion. (c) Internal friction angle (φ), E, ​​ν; When the lithological unit type is weathered soft rock / residual soil, the modified Cambridge model is selected as the adaptive constitutive model. The selection criteria and core function are that it is greatly affected by pore water pressure and has dilatancy. It can simulate consolidation deformation and pore water pressure changes. The input parameters are pre-consolidation pressure (p_c), compression index (λ), rebound index (κ), c, φ; When the lithological unit type is weak interlayer, the double yield surface model is selected as the adaptive constitutive model. The selection criteria and core function are that it is prone to large deformation and creep. It can simulate post-yield plastic flow and long-term deformation. The input parameters are yield stress ratio, E, ν, creep coefficient (such as η).

[0049] The constitutive models of each lithological unit are spatially assigned mechanical parameters as follows: Mechanical parameters required for the corresponding constitutive model are extracted from the experimental data in Step 1, and consistency verification and standardization are performed to ensure parameter validity: Parameter selection: Core parameters are extracted according to the requirements of the constitutive model—for example, for the elastic model, E and ν from the triaxial tests in Step 1 are selected; for the Mohr-Coulomb model, c and φ are extracted (prioritizing field direct shear test data; indoor triaxial data should be multiplied by a reduction factor of 0.8~0.9); for the modified Cambridge model, p_c, λ, and κ from the compaction tests in Step 1 are extracted. Consistency verification: The parameter results of different tests for the same lithological unit are compared (e.g., indoor direct shear c value and field pressuremeter test-estimated c value for the same soil layer). If the deviation is >15%, a weighted average method is used for correction (field test weight 0.6, indoor test weight 0.4) to eliminate data inconsistencies. Standardization: Following the "Min-Max normalization method" from step one, all parameters are compressed to the [0,1] range to avoid affecting the convergence of the numerical simulation due to differences in parameter magnitudes (e.g., E = 1e4kPa, c = 20kPa). Based on the spatial distribution characteristics of lithological units, three scenarios are categorized: "homogeneous units," "heterogeneous units," and "structural surfaces." The preprocessed parameters are accurately mapped to the mesh units of the 3D geometric model. Homogeneous lithological unit assignment: For units with a thickness ≥ 5m and uniform lithology (e.g., thick, intact granite), the average mechanical parameters of the unit are directly assigned in batches to the corresponding meshes—for example, E = 30GPa and ν = 0.25 for the intact granite unit at the top of the slope are uniformly assigned to all hexahedral meshes within that unit. Assignment of heterogeneous lithological units: For units with strong spatial variability of parameters, such as weathering transition zones and residual soils, a "random field model" is used to achieve heterogeneous assignment: Based on the borehole data (parameter values ​​at different depths) from step one, the statistical characteristics of the unit parameters (mean μ, standard deviation σ, coefficient of variation CV) are calculated; a parameter spatial distribution field matching the grid size is generated using a Gaussian random function (ensuring that the parameters change continuously in space, such as the weathering layer gradually changing from 5 GPa to 15 GPa from the slope surface to the depth); the parameter field is mapped one-to-one with the three-dimensional grid units to achieve "each grid unit corresponds to a unique parameter value". Structural surface parameter assignment: For structural surfaces such as joints and faults, define separate mechanical parameters (distinguishing them from surrounding lithological units): Shear strength parameters (c_j, φ_j): Take 0.3~0.5 times the corresponding lithological unit parameters (e.g., c_j=8kPa for jointed sandstone, which is 0.4 times that of sandstone c=20kPa); Stiffness parameters: Normal stiffness K_n=1e5kPa / m, tangential stiffness K_s=5e4kPa / m (based on the adjustment of structural surface roughness from the ground-penetrating radar detection in step one, K_n and K_s for rough surfaces can be increased by 20%~30%); Assign the parameters to the grid interface where the structural surface is located to simulate the shear slip characteristics of the structural surface.

[0050] Step 4: For the three-dimensional geometric model of the slope endowed with mechanical properties, first set the initial conditions and boundary conditions (initial conditions: initial stress field: calculate the self-weight stress based on the density of the soil and rock mass, and superimpose the regional tectonic stress; initial seepage field: set the pore water pressure according to the groundwater level, and calculate the steady-state seepage using the permeability coefficient. Boundary conditions: displacement: fixed at the bottom, rolling on the sides, and free at the top; mechanical: free stress at the bottom / side, and 101.3 kPa atmospheric pressure applied at the top; seepage: permeable at the top, impermeable at the bottom, and constant water head on the sides), and then apply dynamic loads (including but not limited to rainfall loads, seismic loads, and long-term creep loads).

[0051] Step 5: Based on the three-dimensional geometric model of the slope in Step 2, the constitutive parameters in Step 3, and the boundary conditions and dynamic loads in Step 4, multi-physics field coupling processing such as seepage-stress and heat-water-force is carried out. The multi-field response of the slope under dynamic load (such as stress evolution, displacement change, and dynamic distribution of pore water pressure) is calculated through numerical simulation. The simulation accuracy is verified by combining field monitoring data. Finally, a dynamic model of the slope geological body that conforms to the actual evolution law is constructed.

[0052] The micro-motion sensing system simulation unit is used to simulate the full-process monitoring behavior of micro-motion sensors on slopes. Specifically, it imports two types of core input data to provide basic scene parameters for the simulation: Slope dynamic deformation data: Key information is obtained from the "Slope Geological Dynamic Model," including three-dimensional displacement time histories and vibration acceleration signals (such as soil creep caused by rainfall and vibration generated by seismic wave propagation) at different depths and locations on the slope. The data sampling interval is set to 10ms (matching the conventional response rate of MEMS sensors); Environmental interference data: Wind vibration, traffic vibration, and temperature noise are loaded from the "Environmental Interference Dynamic Model," and the interference signal strength is adjusted according to the statistical characteristics of field measurement data (such as amplitude distribution and duration). The two types of data are then time-synchronized (unified timestamps) and normalized (converting physical quantities such as displacement and acceleration into electrical signal quantities that the sensor can respond to), forming the simulation input dataset. Based on the actual working principle of MEMS (Micro-Electro-Mechanical Systems) sensors, the physical deformation of the slope is converted into an electrical signal output by the sensor. The core is to simulate the conversion process of "mechanical quantity → electrical signal": Sensing principle mapping: If a capacitive MEMS accelerometer is used, the slope acceleration is converted into an initial voltage signal using the formula V_out=K·a (where K is the sensor sensitivity coefficient and a is the slope vibration acceleration); Noise characteristic superposition: Simulating the inherent noise of the MEMS sensor (thermal noise, 1 / f noise) and environmental noise, a noise signal is generated according to the noise density in the sensor datasheet and superimposed with the initial voltage signal to obtain the noisy original signal. The wireless transmission loss between the micro-motion sensor and the base station is simulated to restore the signal transmission characteristics in complex field environments. The signal preprocessing stage in an actual monitoring system is simulated to output "quasi-measured data" that can be directly used for feature extraction. The final output consists of two core types of data to support subsequent reinforcement learning training and feature analysis: Time-series dataset: the original digital signal time history of each sensor (including timestamp, sensor ID, and location coordinates) and the preprocessed clean signal time history, in CSV or MAT format, which can be directly used for time-domain / frequency-domain feature extraction; Simulation parameter report: records key parameters such as sensor deployment parameters (location, depth, sampling frequency), environmental interference intensity, transmission loss coefficient, and energy consumption status, providing a basis for subsequent optimization of sensor deployment schemes and adjustment of sampling strategies.

[0053] The environmental disturbance dynamic model unit is used to construct the environmental disturbance dynamic model. The specific construction process is as follows: Typical disturbance sources at the slope site are identified and divided into three categories: Dynamic vibration disturbances: wind vibration (wind speed, wind direction), traffic vibration (vehicle flow, vehicle type, distance), and construction machinery vibration (equipment type, operating frequency); Environmental physical disturbances: temperature changes (daily temperature difference, seasonal fluctuations), humidity / rainfall (raindrop impact, changes in soil moisture content); and inherent disturbances from sensing systems: sensor noise (thermal noise, circuit noise) and wireless transmission interference (electromagnetic signals). For each type of disturbance, core influencing parameters are determined (e.g., wind speed corresponds to wind vibration intensity, and traffic flow corresponds to vibration amplitude), and the parameter value range and temporal variation patterns are clarified (e.g., traffic vibration exhibits morning / evening peak time sequence characteristics). Data was acquired using a combination of field measurements and historical data fusion: Field data collection: During the stable period when the slope showed no significant deformation, a MEMS sensor array was deployed to simultaneously collect interference signals and environmental parameters, continuously collecting data for 30 days to construct the original dataset; Historical data supplementation: Historical data from meteorological stations in the slope area and traffic flow statistics from transportation departments were retrieved to supplement interference characteristics under extreme conditions; Feature extraction: Time-domain and frequency-domain features were extracted from the collected interference signals to establish a database of "interference parameters - signal features". For each type of core interference source, a dedicated mathematical model was established to quantify its signal output: Wind vibration model: The Von Karman wind spectrum was used to describe the power spectral density of pulsating wind, combined with the slope topography to correct the spectral function, outputting wind vibration time-domain signals at different wind speeds; Traffic vibration model: Based on random pulse sequences, with traffic flow as input, each vehicle corresponds to a half-sine pulse (amplitude positively correlated with vehicle type, duration negatively correlated with vehicle speed), superimposed to generate a time-series vibration signal; Temperature interference model: A linear model of temperature-signal drift was established, fitting the relationship between temperature change and sensor output voltage drift using measured data. Considering the interaction between interference sources (such as rainfall increasing soil damping and weakening vibration propagation), a comprehensive model is constructed using "weighted superposition + coupling correction": Basic superposition: The interference signals output by each single-source model are weighted and superimposed according to the proportion of interference intensity measured in the field (such as wind vibration accounting for 30% and traffic vibration accounting for 50% during the stable period) to obtain the initial composite interference signal; Coupling correction: A coupling coefficient between interferences is introduced, such as the increased propagation attenuation coefficient of traffic vibration during rainfall and the increased sensor noise amplitude when the temperature rises. The coupling coefficient is optimized through field verification data, and the final output is a dynamic interference signal consistent with the actual scenario.To adapt to the temporal changes in on-site disturbances (such as seasonal changes leading to changes in wind vibration frequency, and road expansion leading to increased traffic volume), a dynamic model update process is established: Real-time data access: Environmental parameters (real-time wind speed, traffic flow) and disturbance signal data from the slope monitoring system are accessed into the model at hourly frequencies; Online parameter correction: A sliding window (window size of 7 days) is used to calculate the statistical quantities of disturbance characteristics, and key parameters in the model (such as the average pulse amplitude of traffic vibration and the power spectral coefficient of wind vibration) are dynamically updated to ensure that the model output always matches the current on-site disturbance state.

[0054] The following is a detailed simulation scenario for non-destructive monitoring of slopes:

[0055] Using the CGCS2000 geodetic coordinate system of the dynamic model of the slope geological body as the benchmark, the sensor layout coordinates of the micro-motion sensing system simulation unit and the spatial locations of interference sources (such as roads and wind turbines) in the environmental disturbance dynamic model are uniformly transformed to this coordinate system to ensure accurate spatial alignment of the three. Simultaneously, the data format is unified: deformation data from the geological body model, interference signals from the disturbance model, and parameter files from the sensing system are all converted to an interactive JSON format for easy cross-unit access. A unified timestamp is set to align the dynamic deformation time history output by the dynamic model of the slope geological body, the interference signal time history output by the environmental disturbance dynamic model, and the sampling trigger signal of the sensing system. If the data sampling frequency of a certain unit is inconsistent, linear interpolation is used to complete the interval to 10ms to avoid coupling deviations caused by time differences. The physical quantities (such as displacement mm, acceleration m / s²) output from the dynamic model of the slope geological body are converted into input quantities that the sensing system can recognize. For example, the vibration acceleration a = 0.1 m / s² at the toe of the geological body is converted into an initial voltage input of 10 mV by combining the sensitivity K = 100 mV / (m / s²) of the MEMS sensor in the simulation unit of the sensing system with the formula V_in = a × K. At the same time, the interference signals output from the dynamic model of environmental interference (such as the acceleration of 0.02 m / s² corresponding to wind vibration) are also converted into voltage quantities according to the same rules to ensure that they are consistent with the magnitude of the deformation signal of the geological body and to meet the conversion logic of "mechanical quantity → electrical signal" of the sensing system. The real-time disturbance parameters of the environmental disturbance dynamic model are input into the slope geological body dynamic model to correct its mechanical properties and load conditions, simulating slope deformation under real-world conditions. If the disturbance is "rainfall", the rainfall output from the environmental model is input into the geological body model to correct the permeability coefficient of the soil and rock, thereby updating the seepage field and pore water pressure, and finally adjusting the displacement time history output by the geological body model. If the disturbance is "strong wind vibration", the wind load corresponding to the wind speed output from the environmental model is applied to the slope top / slope surface of the geological body model to supplement the dynamic load and correct the slope vibration acceleration time history. Through this coupling, the "slope deformation data" output by the geological body model is no longer an ideal, disturbance-free state, but a real deformation signal that incorporates the influence of environmental disturbance.The "slope deformation data with interference effects" obtained from the first coupling is co-input into the micro-motion sensing system simulation unit with the "inherent interference of the sensing system" (such as sensor thermal noise and electromagnetic interference) output by the environmental interference model, simulating the complete monitoring signal generation process: Step 1: Input "deformation signal + environmental interference signal" - superimpose the slope acceleration output by the geological model with the traffic vibration acceleration output by the environmental model to obtain the total input acceleration a_total = 0.12 m / s²; Step 2: Sensing signal conversion and noise superposition - convert a_total into an initial voltage signal of 12mV according to the MEMS sensor principle, and then superimpose the sensor inherent noise and wireless transmission interference output by the environmental model to obtain the noisy original voltage signal; Step 3: Simulate transmission loss and preprocessing - correct the signal amplitude based on the wireless transmission loss coefficient of the slope area output by the environmental model; then output a clean simulated monitoring signal according to the conventional preprocessing process of the sensing system; through this coupling, end-to-end simulation of "real slope deformation → monitoring signal affected by environmental interference" is realized, restoring the signal characteristics of field monitoring. To avoid the interference signal output by the environmental interference model becoming disconnected from the actual monitoring scenario, the "simulated monitoring signal" output by the sensor system simulation unit is fed back to the environmental interference model to dynamically correct the interference parameters: the interference characteristics of the simulated monitoring signal are extracted, and the deviation between its amplitude and the traffic vibration amplitude initially output by the environmental model is calculated; if the deviation exceeds a preset threshold, the pulse amplitude parameter of traffic vibration in the environmental model is adjusted and the interference signal is re-output until the deviation of the feedback interference characteristics is ≤15%; this coupling forms a closed loop of "interference output → signal feedback → interference correction", ensuring that the output of the environmental interference model always matches the actual interference scenario of the slope monitoring. Based on different risk conditions, the interference types of the environmental interference model are combined, and the above coupling process is repeated to generate multiple sets of associated simulation scenarios: for example, "rainstorm condition": rainfall of 80mm / h + wind speed of 10m / s + traffic volume of 200 vehicles / h, outputting the slope deformation time history, sensor simulated monitoring signal, and interference parameter change curve under this condition; forming a full-condition simulation scenario library covering "daily stability - mild risk - extreme risk", supporting subsequent reinforcement learning training for different scenarios.

[0056] The final output is a spatiotemporally integrated slope non-destructive monitoring correlation simulation scenario package, containing three core deliverables: Dynamic simulation dataset: Simulated monitoring signal time histories of each MEMS sensor (distinguished by deployment location) under different working conditions, accompanied by corresponding real slope deformation data and environmental interference parameters; 3D visualization scenario: Sensor spatial distribution, real-time deformation cloud map, and interference source location and intensity markings (e.g., using different colors to represent the impact range of wind vibration / traffic vibration) are overlaid in the 3D mesh of the geological model, intuitively presenting the spatial correlation between "slope deformation - sensing monitoring - environmental interference"; Coupling parameter configuration file: Records the coupling parameters of the three units (e.g., time step 10ms, feedback threshold 15%, sensor sensitivity 100mV / (m / s²)), which can be quickly called and modified when adjusting working conditions or optimizing sensor schemes.

[0057] The slope micro-motion sensing and conversion module is used to define the slope state space and micro-motion feature mapping space based on the slope non-destructive monitoring associated simulation scenario.

[0058] The slope state space includes the geomechanical foundation subspace. Multiphysics Dynamic Response Subspace Stability level subspace Environmental interference coupling subspace ;

[0059] Geomechanics Fundamental Subspace Specifically, it includes the following core variables: 1. Elastic modulus E of soil and rock mass (assigned to corresponding grid cells); 2. Cohesion c (corrected value from direct shear test); 3. Angle of internal friction φ (reduced value from triaxial test); 4. Permeability coefficient k (value from water pressure test).

[0060] Multiphysics Dynamic Response Subspace Specifically, it includes the following core variables: 1. Vertical displacement rate (time-series output of the dynamic model of the slope geological body); 2. Horizontal displacement rate (time-series output of the dynamic model of the slope geological body); 3. Shear stress τ (stress value of potential slip surface element); 4. Pore water pressure u (output of seepage field); 5. Plastic zone area ratio A_p (number of plastic elements / total number of meshes);

[0061] Stability level subspace Specifically, it includes the following core variables: 1. Safety factor FS (calculated based on the limit equilibrium method, FS≥1.2 indicates stability, FS<1.0 indicates instability); 2. Stability index SI (FS normalized to [0,1], SI=0 indicates instability, SI=1 indicates stability); 3. Instability risk probability P (based on the output of a probability model trained on historical working conditions; the process of building a probability model trained on historical working conditions is as follows: data source and preprocessing: extract “slope state variables (such as displacement rate, safety factor) + micro-motion characteristics (such as dominant frequency, kurtosis) + working condition labels” from the historical working condition library of related simulation scenarios (such as daily stability, rainstorm, earthquake, etc., 20+ types of working conditions). The sample set was set as follows: "(Stable = 0 / Critical = 1 / Instability = 2)". Data was standardized (Min-Max normalization), and samples were balanced (using SMOTE to address the scarcity of instability samples). The data was then divided into a training set (70%), a validation set (20%), and a test set (10%). Model selection: A hybrid LSTM-XGBoost model was chosen (balancing temporal characteristics and feature interactions): the LSTM layer captures the temporal evolution of slope state / micro-motion characteristics (e.g., the increasing trend of displacement rate), the XGBoost layer mines the nonlinear correlations between features (e.g., the coupling relationship between pore water pressure and instability), and the output layer uses the Softmax activation function. The model directly outputs the probabilities of three working conditions: "stable / critical / instability," with the "instability probability" taken as P. Training and validation: Model parameters are optimized using the "cross-entropy loss function," and validated using "accuracy, AUC value, and reliability plot." The requirements are: test set AUC ≥ 0.95 (ability to distinguish working conditions), probability calibration error ≤ 5% (ensuring the reliability of P). For example, if the model outputs P = 85% under heavy rain conditions, the actual instability probability needs to be close to 85%. 4. Critical deformation threshold ratio R (current cumulative displacement / critical instability displacement; current cumulative displacement: obtained from the dynamic model of the slope geological body in the slope non-destructive monitoring associated simulation scenario—the model is set to 1...). The 0ms time step outputs the time-series displacement data of key risk units (such as potential slip surfaces and slope toes) of the slope. The cumulative displacement of the unit is obtained by summing the displacement values ​​at each time step within a certain monitoring period. The critical displacement for instability is obtained in two ways: First, by simulating slope instability conditions in a related simulation scenario (such as continuously loading rainfall / seismic loads until the safety factor FS < 1.0), the cumulative displacement of the key risk unit when the slope just begins to instability is recorded as the critical displacement for instability at the simulation level; second, by combining field instability case data from similar slope projects and historical monitoring records of critical instability, the simulation values ​​are corrected to obtain a more realistic critical displacement for instability.

[0062] Environmental interference coupling subspace Specifically, it includes the following core variables: 1. Hourly rainfall R_h (output of the environmental disturbance dynamic model); 2. Groundwater level depth h_w (data from hydrological monitoring wells); 3. Wind load intensity F_w (calculated using Von Karman wind spectrum); 4. Traffic vibration amplitude a_tr (output of the traffic vibration model); 5. Temperature influence coefficient K_t (correlation value between temperature and soil stiffness, [0,1], the calculation principle of the correlation value between temperature and soil stiffness: Data acquisition: Extract the elastic modulus E of soil at different temperatures from the geological model of the associated simulation scenario to form a dataset corresponding to "temperature T - elastic modulus E"; Benchmark setting: Take the elastic modulus E0 at room temperature as the benchmark stiffness, at this time K_t=1; Correlation calculation: Calculate according to the formula K_t=E / E0 (E is the elastic modulus at the current temperature). Since E will decrease when the temperature deviates from room temperature (such as high temperature softening, low temperature frost heave affecting stiffness), E≤E0, K_t falls in the [0,1] interval); 6. Comprehensive disturbance intensity I (determined) Core Interference Items and Normalization: From the environmental interference model of the associated simulation scenario, select the interference with the greatest impact on monitoring (such as rainfall, wind load, traffic vibration, and temperature), and normalize the actual values ​​of each interference to [0,1] using the Min-Max method to obtain each interference component (such as rainfall component R_norm and wind load component F_norm); Assign interference weights: Set weights according to the degree of impact of the interference on slope monitoring (such as rainfall having the greatest impact, set 0.4, traffic vibration 0.3, wind load 0.2, and temperature 0.1, with a weight sum of 1); Calculate the weighted sum I: Use the formula I = (R_norm × 0.4) + (F_norm × 0.2) + (traffic vibration component × 0.3) + (temperature component × 0.1), the result is the comprehensive intensity of interference falling within [0,1]).

[0063] The micro-motion feature mapping space includes time-domain features Frequency domain characteristics Time-frequency domain characteristics Spatial features ;

[0064] Temporal characteristics Specifically, it includes the following core features: 1. Peak value; 2. Root mean square (RMS); 3. Kurtosis (reflects the impulsiveness of the signal, which is prone to sudden changes before instability); 4. Waveform factor (waveform factor = RMS / average amplitude, the average amplitude is the arithmetic mean of the amplitude (absolute value) of the micro-motion sensing signal within a specified analysis period); 5. Peak factor (peak factor = peak value / RMS); 6. Temporal entropy;

[0065] Frequency domain characteristics Specifically, it includes the following core features: 1. Dominant frequency (the frequency corresponding to the peak power spectrum, the dominant frequency of slope deformation); 2. Spectral entropy; 3. Frequency band energy proportion (2-5Hz / 5-10Hz, differences in energy distribution at different deformation stages); 4. Peak power spectral density (reflecting the energy intensity in the frequency domain); 5. Harmonic noise ratio (dominant frequency energy / total energy of other frequencies).

[0066] Time-frequency domain features Specifically, it includes the following core features: 1. Wavelet packet decomposition energy (db4 wavelet, 3-level decomposition, energy values ​​of 8 frequency bands, reflecting the time-frequency distribution of the signal); 2. HHT instantaneous frequency (instantaneous frequency of each IMF component after EMD decomposition, capturing non-stationary deformation); 3. HHT marginal spectrum peak (energy concentration point in the time-frequency domain, corresponding to the moment of deformation burst); 4. Wavelet entropy (uncertainty of the signal after wavelet decomposition).

[0067] Spatial features Specifically, it includes the following core features: 1. Sensor array correlation coefficient (Pearson correlation coefficient between any two sensor signals, reflecting deformation consistency); 2. Signal propagation speed (calculated by the time difference between adjacent sensor signals, propagation speed increases before instability); 3. Spatial entropy (distribution entropy of sensor array eigenvalues, reflecting deformation spatial uniformity); 4. Eigenvalue variance (variance of the sensor array's dominant frequency, small variance when stable); 5. Array energy concentration (variance coefficient of RMS of all sensors, high concentration when unstable).

[0068] The slope non-destructive monitoring enhancement decision design module is used to define the action space of slope non-destructive monitoring and design the state reward function. Constrained reward function Further determine the two-dimensional dynamic reward function , ;

[0069] The action space for non-destructive slope monitoring includes the following dimensions: sensor deployment and adjustment. Sampling strategy adjustment , Transmission parameter adjustment Signal preprocessing adjustment ;

[0070] Sensor deployment adjustment Action options and parameter ranges: 1. Position fine-tuning: Within the three-dimensional coordinate system of the slope non-destructive monitoring associated simulation scenario, the sensor switches between the preset areas of "slope top / slope toe / potential slip surface"; 2. Quantity increase / decrease: Within budget constraints, the number of sensors is ±1~2; 3. Depth adjustment: Under the premise of non-destructive deployment, the sensor burial depth is ±0.5~1m.

[0071] Sampling strategy adjustment Action options and parameter range: 1. Sampling frequency: 10Hz / 20Hz / 50Hz; 2. Sampling mode: continuous sampling / intermittent sampling (when intermittent, the sampling interval is 5 / 10 / 30min).

[0072] Transmission parameter adjustment Action options and parameter range: Transmission power: 10dBm / 20dBm / 30dBm (balance transmission distance and energy consumption, power ≤30dBm to avoid excessive energy consumption).

[0073] Signal preprocessing adjustment Action options and parameter range: 1. Filtering method: Low-pass filter / wavelet filter (low-pass is suitable for stable signals, wavelet is suitable for non-stationary micro-motion signals); 2. Filter cutoff frequency: 5Hz / 10Hz / 20Hz (matches the main frequency range in the frequency domain characteristics to avoid filtering out effective signals).

[0074] State reward function ,in, The feature error weight (the feature error weight can be set to 0.6 to prioritize the accuracy of highly sensitive features, as these features contribute more to instability warning). The error attenuation coefficient is determined comprehensively based on the engineering accuracy requirements of non-destructive slope monitoring, the gradient differentiation requirements of the state reward function, and the historical data calibration of the associated simulation scenario. You can take 5). For slope conditions inferred from micro-motion characteristics (such as displacement rate) Safety factor ), the output of the mapping model from the slope micro-motion sensing and conversion module; Numerical simulation results (such as time-series outputs) from dynamic model elements of the slope geological body represent the actual state of the slope. Calculated using the limit equilibrium method ); The mean square error of the state estimate quantifies the overall state deviation (e.g., 'n' represents the number of core variables in the state space, and the values ​​are the safety factor FS, stability index SI, and vertical displacement rate of the multiphysics subspace. Horizontal displacement rate There are a total of 4 core variables in the state space. Let be the estimated value of the i-th state-space core variable. (where i is the true value of the i-th state space core variable). These are estimates of highly sensitive micro-motion features (such as time-domain kurtosis, frequency-domain dominant frequency, and time-frequency-domain wavelet packet energy, derived from sensitivity verification results in the micro-motion feature mapping space). The true value of the highly sensitive micro-motion characteristics (derived from the clean signal extraction result of the micro-motion sensing system simulation unit); ; The total number of highly sensitive micro-motion features. The absolute error of the m-th highly sensitive micro-motion feature (the absolute difference between the estimated value and the true value of the highly sensitive micro-motion feature);

[0075] Constrained reward function ;in, This is the non-destructive weight (to prioritize ensuring the slope remains undamaged and avoid soil and rock damage due to cost control, a value of 0.6 can be used). This represents the soil and rock disturbance index. , The sensor burial depth adjustment amount comes from the "depth adjustment" dimension of the motion space; This is the maximum allowable value for burial depth adjustment. To add more borehole sensors (from the "Quantity Increase / Decrease" action in the "Sensor Deployment Adjustment" dimension of the action space). This represents the maximum allowed number of new boreholes. Since the loosening effect of the hole is greater than the loosening effect of the burial depth adjustment, the weighting coefficient is taken as 0.4 / 0.6 in the formula. The non-destructive threshold is set to 1, which comes from the "non-destructive monitoring" requirement of the slope geological body dynamic model. D≤1 means that non-destructive monitoring is satisfied. To monitor the total cost of the system, Sensor quantity cost: Calculated by multiplying the sensor quantity increase or decrease in the motion space by (the sensor unit price in the micro-motion sensing simulation report + engineering installation cost); Transmission power cost: Calculated by multiplying the transmission power energy consumption of the environmental interference dynamic model by the annual monitoring duration and electricity price, plus the fixed base station maintenance cost; Sampling strategy cost: Calculated by multiplying the sampling parameter energy consumption of the micro-motion sensing simulation report by the annual duration and electricity price, plus the depreciation of sampling equipment and data storage costs; Through cost accounting for "20 typical working conditions" in the associated simulation scenario, the average proportion of the three types of costs is calculated as follows: Sensor quantity cost accounts for an average of 52%, close to 0.5; Transmission power cost accounts for an average of 28%, close to 0.3; Sampling strategy cost accounts for an average of 20%, perfectly matching 0.2; Therefore... The weighting coefficients in the formula are allocated as follows; This represents the upper limit of the budget (derived from the actual budget constraints of the project).

[0076] Two-dimensional dynamic reward function ;in, The dynamic weighting factor is determined by the stability index SI and is adjusted using a "piecewise exponential" method. The formula is as follows: ;

[0077] The slope non-destructive monitoring scheme decision module generates non-destructive monitoring early warning decision schemes based on the Pareto optimal solution set of the improved NSGA-II algorithm, specifically as follows: The traditional NSGA-II algorithm is improved in the following three aspects:

[0078] Initial population generation guided by reinforcement learning: Random initialization is abandoned. An initial population (size 100-150) is generated using reinforcement learning agents trained with the "Slope Non-destructive Monitoring Reinforcement Decision Design Module." Each individual in the population corresponds to a set of action combinations: "sensor deployment + sampling strategy + transmission parameters + preprocessing adjustment," and must satisfy the constraint condition (D≤1, ...). This ensures that over 70% of the initial population consists of high-quality solutions that balance accuracy and constraints.

[0079] Constraint-aware crossover and mutation operators: During crossover, priority is given to pairing individuals with significant differences in objective functions (state rewards, constraint rewards) (e.g., high-precision, low-economic solution × low-precision, high-economic solution), preserving effective combinations in the action space; during mutation, "engineering constraint verification" is added, and if the mutated solution fails to meet the lossless (D>1) or economic requirements (…), the verification will be performed accordingly. Constraints automatically revert to the pre-mutation state to avoid generating invalid individuals;

[0080] Reward function fusion selection strategy: "Two-dimensional dynamic reward function" "As a basis for evaluating individual fitness, in rapid non-dominated ranking, priority is given to retaining..." Individuals (i.e., the "accuracy-compliant + constraint-compliant" solution) improve the efficiency of population evolution.

[0081] Pareto optimal solution set generation process:

[0082] With the three optimization objectives of "maximizing monitoring accuracy (state reward), minimizing soil and rock disturbance (D), and minimizing cost (C)," the solution set is generated according to the following steps:

[0083] 1. Population coding and objective function definition

[0084] Individual coding: A "mixed integer + real number coding" is adopted. Each individual (monitoring scheme) corresponds to a 12-bit code, for example: [01 (slope foot position), 02 (2 sensors), 01 (0.5m burial depth), 02 (20Hz sampling), 01 (intermittent 10min), 02 (20dBm transmission), 02 (wavelet filtering), 02 (10Hz cutoff frequency)]. The coding value corresponds to the specific options in the action space.

[0085] Objective function:

[0086] (Minimize negative state rewards, i.e. maximize monitoring accuracy).

[0087] (Minimize soil and rock disturbance);

[0088] (Minimize the normalization cost, ensuring it is ≤1).

[0089] 2. Algorithm Iteration Process (50-80 iterations)

[0090] Initial population generation: Generate 100 valid individuals according to "Improvement point 1", and calculate the three objective function values ​​for each individual;

[0091] Fast non-dominated sorting: The population is stratified according to the objective function. Individuals that are not dominated by any individual are assigned to the first frontier (potential optimum), and then successively assigned to the second and third frontiers. The individuals in the first three frontiers (about 60) are retained.

[0092] Crowding calculation: For individuals at the same frontier, calculate their crowding distance in the three objective function dimensions (the larger the distance, the better the individual diversity), retain the top 80% of individuals in terms of crowding distance, and avoid convergence of solution sets;

[0093] Selection-Crossover-Mutation: The tournament selection method is used to select parent individuals, with a crossover probability of 0.8 (constrained-aware crossover) and a mutation probability of 0.1 (constrained-verified mutation), to generate a progeny population (100 individuals).

[0094] Population merging and iteration: Merge the parent and offspring populations (200 individuals), repeat steps 2-3, retain 100 high-quality individuals to enter the next generation, until the iteration converges (the first leading individual shows no significant change for 5 consecutive generations).

[0095] 3. Determining the Pareto optimal solution set

[0096] After iterative convergence, the individuals at the first front constitute a Pareto optimal solution set (size 20-30 sets), and each solution set corresponds to a complete non-destructive monitoring scheme, including:

[0097] Hardware parameters: sensor deployment location / quantity / burial depth, transmission power;

[0098] Operating parameters: sampling frequency / mode, filtering method / cutoff frequency;

[0099] Objective function value: State reward function Disturbance index (D), normalized cost ( (This information is provided for reference in subsequent decision-making.)

[0100] Based on the Pareto optimal solution set and considering the priority requirements of different risk conditions of the slope (normal stability, critical risk, and extreme risk), the final decision is made using the Fuzzy Hierarchical Analysis (FAHP) method, with the following steps:

[0101] 1. Determine the appropriate matching of decision evaluation indicators and weights.

[0102] To ensure the practicality of the project, four core evaluation indicators were selected, with their weights dynamically adjusted according to the operating conditions: Evaluation indicator: Monitoring accuracy (Indicator definition: State reward function) Weighting: Daily stability: 0.3; Critical risk: 0.4; Extreme risk: 0.5); Evaluation metric: Losslessness (metric definition: Disturbance index D (lower is better, ≤0.8 is excellent), weighting: Daily stability: 0.3; Critical risk: 0.3; Extreme risk: 0.2); Evaluation metric: Economy (metric definition: Normalized cost) The lower the better, ≤0.8 is preferred. Weighting: Daily stability: 0.3; Critical risk: 0.2; Extreme risk: 0.2); Evaluation index: Implementation difficulty (Indicator definition: Equipment / construction complexity required for the plan (e.g., sensor burial depth ≤1m is easy, recorded as 1; >1m is difficult, recorded as 0.5), Weighting: Daily stability: 0.1; Critical risk: 0.1; Extreme risk: 0.1).

[0103] 2. Scheme scoring and ranking

[0104] Indicator Quantification: For each Pareto optimal solution, a quantitative score is given according to the indicator definition (e.g., ...). (85 points, D=0.7, 90 points)

[0105] Weighted score calculation: Calculate the total score of the solution according to the weight of the current working condition (e.g., under extreme risk, a certain solution has 85 points for accuracy × 0.5 + 80 points for non-destructive performance × 0.2 + 75 points for economy × 0.2 + 90 points for difficulty × 0.1 = 82 points).

[0106] Scheme ranking: Sort by total score in descending order, and select the top 3 as candidate schemes.

[0107] 3. Final Solution Determination and Verification

[0108] Working condition adaptation verification: For candidate schemes, verify their actual monitoring effect under the corresponding working conditions of the "slope non-destructive monitoring associated simulation scenario" (e.g., extreme risk takes the "rainstorm + earthquake" scenario) - requiring early warning response time ≤10s and false alarm rate ≤5%;

[0109] Expert review: Based on the experience of engineering experts, select one solution from the candidate solutions that has the highest score and has passed verification as the final implementation plan;

[0110] Output Engineering Guide: Generates a detailed implementation guide for the solution, including CAD drawings of sensor deployment, parameter configuration tables, and early warning thresholds.

[0111] The above formulas are all dimensionless calculations, and the preset parameters in the formulas should be set by those skilled in the art according to the actual situation.

[0112] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0113] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0114] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0115] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0116] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A non-destructive monitoring and early warning system for slopes based on micro-motion sensing, characterized in that, include, The slope non-destructive monitoring coupling module is used to coordinate and couple the dynamic model unit of the slope geological body, the simulation unit of the micro-motion sensing system, and the dynamic model unit of environmental disturbance to construct a slope non-destructive monitoring associated simulation scenario. The spatiotemporal integrated slope non-destructive monitoring correlation simulation scenario package includes three core deliverables: dynamic simulation dataset: the time history of simulated monitoring signals from each MEMS sensor under different working conditions, accompanied by the actual slope deformation data and environmental interference parameters at the corresponding time; 3D visualization scenario: the spatial distribution of sensors, real-time deformation cloud map, and the location and intensity of interference sources are superimposed on the 3D mesh of the geological model, intuitively presenting the spatial correlation between slope deformation, sensing monitoring, and environmental interference; Coupling parameter configuration file: Records the coupling parameters of the three units, which can be quickly called up and modified when adjusting the working conditions or optimizing the sensor scheme; The dynamic model element for slope geological bodies is used to construct a dynamic model of slope geological bodies. The specific construction process is as follows: Step 1: Collect basic geological data of the slope and test the mechanical parameters of the soil and rock mass of the slope. Then, fuse the multi-source data obtained from the collection and testing. Step 2: Construct a three-dimensional geometric model of the slope based on the fused multi-source data, and then mesh the three-dimensional geometric model of the slope. Step 3: Select the appropriate constitutive model according to the lithological units in the three-dimensional geometric model of the slope, and assign the mechanical parameters to the constitutive model of each lithological unit in space. Step 4: For the three-dimensional geometric model of the slope endowed with mechanical properties, first set the initial conditions and boundary conditions, and then apply dynamic loads; Step 5: Based on the three-dimensional geometric model of the slope in Step 2, the constitutive parameters in Step 3, and the boundary conditions and dynamic loads in Step 4, multi-physics coupling processing is carried out. The multi-field response of the slope under dynamic load is calculated through numerical simulation, and the simulation accuracy is verified by combining field monitoring data. Finally, a dynamic model of the slope geological body that conforms to the actual evolution law is constructed. The micro-motion sensing system simulation unit is used to simulate the full-process monitoring behavior of micro-motion sensors on slopes; The environmental disturbance dynamic model unit is used to construct the environmental disturbance dynamic model. The slope micro-motion sensing and conversion module is used to define the slope state space and micro-motion feature mapping space based on the slope non-destructive monitoring associated simulation scenario. The slope state space includes the geomechanical foundation subspace. Multiphysics Dynamic Response Subspace Stability level subspace Environmental interference coupling subspace ; The micro-motion feature mapping space includes time-domain features Frequency domain characteristics Time-frequency domain characteristics Spatial features ; The slope non-destructive monitoring enhancement decision design module is used to define the action space of slope non-destructive monitoring and design the state reward function. Constrained reward function Further determine the two-dimensional dynamic reward function ; The action space for non-destructive slope monitoring includes the following dimensions: sensor deployment and adjustment. Sampling strategy adjustment , Transmission parameter adjustment Signal preprocessing adjustment ; The slope non-destructive monitoring scheme decision module generates non-destructive monitoring early warning decision schemes based on the Pareto optimal solution set of the improved NSGA-II algorithm.

2. The slope non-destructive monitoring and early warning system based on micro-motion sensing according to claim 1, characterized in that, State reward function in, For feature error weights; This is the error attenuation coefficient; The slope condition is inferred from the micro-motion characteristics; This represents the actual condition of the slope; The mean square error of the state estimation; This is an estimate of highly sensitive micro-motion characteristics. This represents the true value of highly sensitive micro-motion characteristics.

3. The slope non-destructive monitoring and early warning system based on micro-motion sensing according to claim 1, characterized in that, Constrained reward function in, These are lossless weights; The soil and rock disturbance index; The threshold value is non-destructive. The total cost of the monitoring system; This is the budget ceiling.

4. The slope non-destructive monitoring and early warning system based on micro-motion sensing according to claim 1, characterized in that, Two-dimensional dynamic reward function in, The dynamic weighting factor is determined by the stability index SI, and the formula is: .

Images