An ecological slope protection state estimation method and system based on adaptive data assimilation

Abstract

The present application belongs to the technical field of geotechnical engineering, ecological restoration and intelligent monitoring, and specifically discloses an ecological slope protection state estimation method and system based on adaptive data assimilation; the ecological slope protection state estimation method comprises the following steps: S1: constructing a state space model of the slope protection system; S2: fusing observation data from a sensor network in the slope protection area to estimate the state of the slope protection system in real time; S3: in the updating process of the ensemble Kalman filter, an adaptive covariance inflation technique is used to estimate the covariance inflation factor online, and the error covariance matrix is inflated to suppress filter divergence; S4: outputting the estimation result. The estimation system comprises a sensor network, a data processing unit and an output unit. The ecological slope protection state estimation method and system provided by the present application have adaptive anti-divergence capability and can adapt to sparse observation conditions, so as to realize accurate, stable and real-time perception of the state of the slope protection system and provide a scientific basis for safety warning and ecological management.

Description

Technical Field

[0001] This invention belongs to the fields of geotechnical engineering, ecological restoration and intelligent monitoring technology, and in particular relates to an ecological slope protection state estimation method and system based on adaptive data assimilation. Background Technology

[0002] Ecological slope protection is a core technical means for the management and ecological restoration of rivers, lakes, and reservoirs. It organically combines vegetation restoration with engineering structural measures (such as ecological concrete and reinforced cell structures) to improve the slope's resistance to erosion and overall stability, while also taking into account ecological functions such as soil and water conservation, landscape beautification, and restoration of biological habitats. Accurate and timely understanding of the internal state of the ecological slope protection system, including but not limited to displacement field, stress field, seepage field (pore water pressure and water content), and vegetation physiological and ecological indicators, is a key prerequisite for achieving engineering safety early warning, scientific maintenance management, and quantitative assessment of ecological benefits.

[0003] However, current practices in ecological slope protection status monitoring and assessment generally face the technical dilemma of a long-standing disconnect between physical models and field monitoring data, resulting in inconsistent data and inaccurate models; specifically manifested in:

[0004] First, the reliability of traditional slope stability analysis methods, such as the limit equilibrium method and finite element / finite difference numerical simulation, depends heavily on the accuracy of input material parameters (such as soil strength parameters and permeability coefficients) and boundary conditions (such as water level and load). In actual engineering, soil and rock materials exhibit significant spatial variability and temporal evolution (such as wet-dry cycles, freeze-thaw cycles, and changes in soil properties caused by root development), and ecological slope protection involves complex interactions between water, soil, vegetation, and structure, which traditional models struggle to accurately characterize. This results in a significant deviation between the pure model calculation results and the actual engineering conditions, limiting their predictive capabilities.

[0005] Secondly, with the development of sensing technology, modern slope protection projects can deploy various sensors (such as strain sensors, inclinometers, soil moisture sensors, pore water pressure gauges, etc.) and conduct periodic remote sensing observations (such as UAVs acquiring the Normalized Difference Vegetation Index (NDVI)), generating massive amounts of multi-source heterogeneous monitoring data. However, currently, these data are mostly used for simple threshold alarms or post-event analysis, and have not been effectively and in real-time integrated into the dynamic model describing the physical laws of the system for state correction, thus failing to release the potential value of the data.

[0006] Data assimilation technology provides a strong theoretical framework for bridging the gap between models and data. Originating from numerical weather prediction, the core idea of ​​this technology is to integrate noisy observation data into a dynamic model in real time through algorithms (such as ensemble Kalman filtering), continuously correcting the model's state variables, and making the model trajectory approximate the actual system evolution trajectory as closely as possible. In recent years, this technology has begun to be attempted to be introduced into the field of geotechnical engineering.

[0007] For example, patent CN118968734A discloses a method for predicting landslide disasters in complex terrain based on the dynamic coupling of meteorological and hydrological landslides. It proposes using ensemble Kalman filtering (EnKF) to couple the WRF-Hydro hydrological model with the TRIGRS landslide physical model, correcting model parameters by assimilating observational data, and then conducting short- to medium-term landslide disaster forecasts. This scheme demonstrates the application potential of data assimilation in slope disaster early warning. However, its focus is on parameter adjustment and disaster forecasting, rather than directly addressing the more fundamental and ongoing need for real-time, high-precision estimation of the multi-physics state within the slope protection system. More importantly, this scheme does not address the common filtering divergence problem and its solution when applying EnKF in nonlinear, high-dimensional systems. Filtering divergence refers to the continuous underestimation of the covariance of the state estimation error by the filtering algorithm due to model errors, finite set sampling errors, etc., leading to the failure of the Kalman gain, the gradual neglect of observational data during the assimilation process, and ultimately causing the estimation results to deviate from the true state. This problem severely restricts the long-term and stable application of data assimilation technology in nonlinear geotechnical systems.

[0008] Furthermore, patent CN120299173A discloses a dynamic landslide early warning method and device based on landslide physical models and data assimilation. It uses data assimilation technology combined with physical models and machine learning models for dynamic early warning. However, it is vague about the specific implementation algorithm of data assimilation (e.g., whether to use EnKF, what assimilation strategy to use), and does not mention how to ensure the numerical stability and robustness of the assimilation process. In addition, this solution also fails to solve the challenge that the observation space is much smaller than the state space due to the sparse distribution of monitoring points in ecological slope protection projects. Sparse observations are prone to spurious correlations in data assimilation, leading to incorrect correction of state variables far from the observation points.

[0009] Furthermore, existing attempts have focused on the hydro-mechanical coupling early warning of landslide disasters, failing to fully consider vegetation as a living element in ecological slope protection. Vegetation affects the seepage field through transpiration and influences the mechanical field through root reinforcement and anchoring. Its health status is also an important indicator for engineering evaluation. Therefore, there is an urgent need for a collaborative state estimation method that can integrate hydro-mechanical-ecological multi-field processes and overcome filtering divergence and adapt to sparse observations.

[0010] In summary, there is an urgent need for an ecological slope protection state estimation method and system that can deeply integrate multi-source monitoring data and multi-field coupled physical models, while possessing adaptive anti-divergence capabilities and adapting to sparse observation conditions. This would enable accurate, stable, and real-time perception of the slope protection system's state and provide a scientific basis for safety early warning and ecological management. Summary of the Invention

[0011] In view of this, the purpose of this invention is to provide an ecological slope protection state estimation method and system that can deeply integrate multi-source monitoring data and multi-field coupled physical models, while possessing adaptive anti-divergence capabilities and adapting to sparse observation conditions, so as to achieve accurate, stable, and real-time perception of the slope protection system state and provide a scientific basis for safety early warning and ecological management.

[0012] To achieve the above objectives, on the one hand, the present invention provides an ecological slope protection state estimation method based on adaptive data assimilation, comprising the following steps:

[0013] Step S1: Construct a state-space model of the slope protection system. First, establish a dynamic equation describing the evolution of the slope protection system. The dynamic equation is coupled with at least two of the following: a hydrological process model, a geotechnical mechanics model, and a vegetation growth model. Then, based on the dynamic equation, construct the state vector X, the observation vector Y, and the observation operator H.

[0014] Step S2: Based on the state-space model, the ensemble Kalman filter algorithm is used to fuse observation data from the sensor network in the slope protection area to estimate the state of the slope protection system in real time;

[0015] Step S3: In the update process of the ensemble Kalman filter, the covariance inflation factor is estimated online using the adaptive covariance inflation technique, and the error covariance matrix is ​​inflated to suppress filter divergence.

[0016] Step S4: Output the state estimation results of the slope protection system after the divergence suppression process in step S3.

[0017] Furthermore, in step S1, the state vector X includes at least two state variables from the displacement field, pore water pressure field, volumetric water content field, and vegetation biomass field; the observation vector Y corresponds to the observation data collected by sensors deployed in the slope protection area, and includes at least two observation variables from strain, soil moisture, pore water pressure, and normalized vegetation index; the observation operator H describes the mapping relationship from the state vector X to the observation vector Y, i.e., Y=H(X)+ε, where ε is the observation error vector.

[0018] Furthermore, in step S1, the kinetic equations include at least two of the following: the Richards equation describing soil moisture movement, the equilibrium equation describing stress-strain relationship, and the kinetic equation describing vegetation growth.

[0019] Furthermore, the specific sub-steps for state estimation using ensemble Kalman filtering in step S2 include:

[0020] S21: Generate N set members ,i=1,2,...,N, where each set member represents a possible state realization;

[0021] S22: Based on the dynamic equation, predict each set member to obtain the prior state set at the current moment;

[0022] S23: Calculate the mean of the prior state set. and the prior error covariance matrix P f ;

[0023] S24: Obtain the actual observation data Y at the current moment. t and the observation error covariance matrix R;

[0024] S25: Calculate the Kalman gain matrix K and update the members of each set to obtain the analysis state set { }

[0025] Furthermore, in step S3, the adaptive covariance inflation technique estimates the covariance inflation factor λ online using the following formula:

[0026]

[0027] in, Let R be the observation increment vector, representing the difference between the actual observations and prior estimates in the observation space; R is the observation error covariance matrix; H is the observation operator; P f The prior error covariance matrix; Represents the trace of a matrix.

[0028] Furthermore, between steps S23 and S25, a step of spatial localization processing of the prior error covariance matrix Pf is included, wherein the processing is performed using a localization function. With matrix P f Implementing the Schur product operation: Where d is the spatial distance between the state variable and the observation point, the function The value of d decreases as d increases.

[0029] Furthermore, following step S4, step S5 is also included:

[0030] Using the state estimation results as initial conditions, the slope protection state at future times is predicted based on the dynamic equations. The prediction results include at least one of displacement field evolution, safety factor change, and vegetation coverage change.

[0031] On the other hand, the present invention also provides an ecological slope protection state estimation system based on adaptive data assimilation, for implementing the aforementioned method, comprising:

[0032] Sensor networks, deployed in the slope protection area, are used to collect multi-source heterogeneous observation data;

[0033] The data processing unit includes a processor and a memory, wherein the memory stores a computer program, which, when executed by the processor, is used to implement the aforementioned method;

[0034] The output unit is used to output the state estimation results, prediction results, and early warning information.

[0035] Compared with existing technologies, the ecological slope protection state estimation method and system based on adaptive data assimilation provided by this invention has at least the following beneficial technical effects:

[0036] First, this invention systematically applies Kalman filtering data assimilation technology to the complex scenario of ecological slope protection, constructing a unified framework that integrates a multi-field coupled physical model of water-mechanical-ecological fields with multi-source monitoring data. Through the assimilation algorithm, real-time observation data is continuously used to correct the model state, enabling the model trajectory to closely track the real system, thus solving the problems of inaccurate models and unusable data.

[0037] Second, the online adaptive covariance inflation technique based on the statistical characteristics of observation increments proposed in this invention can dynamically adjust the inflation factor λ according to the real-time consistency between model predictions and observations, without relying on repeated adjustments based on human experience. This enhances the robustness and long-term stability of the assimilation system, making the long-term and reliable application of data assimilation technology in practical engineering possible.

[0038] Third, this invention integrates spatial localization technology into the data assimilation process. By using a distance-decaying function, it suppresses spurious statistical correlations between distant points in the state space, ensuring that the observation information mainly corrects the state variables of the spatially adjacent region.

[0039] Fourth, the observation operator and state vector of this invention can be compatible with observation data with different physical meanings and dimensions, such as strain, water content, pore water pressure, and vegetation index (NDVI), to achieve the synergistic assimilation of multi-source information from mechanics, hydraulics, and ecology. This enables the simultaneous output of synergistic estimation results of displacement field, seepage field, and vegetation ecological field, providing comprehensive data support for the structural safety assessment and ecological benefit evaluation of slope protection projects.

[0040] Fifth, this invention can predict displacement trends, safety factor changes, and saturation zone expansion in future time periods, and can identify potential instability risks several hours to tens of hours in advance, providing a valuable decision-making window for emergency response and scientific management of engineering projects.

[0041] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0042] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute a limitation thereof.

[0043] Figure 1 This is a flowchart of an ecological slope protection state estimation method based on adaptive data assimilation proposed in this invention;

[0044] Figure 2 This is a diagram illustrating the composition of an ecological slope protection state estimation system based on adaptive data assimilation proposed in this invention.

[0045] Figure 3 This is a schematic diagram of the ecological slope protection sensor network layout according to an embodiment of the present invention;

[0046] Figure 4 This is a schematic diagram of the data assimilation process according to an embodiment of the present invention;

[0047] Figure 5 This is a comparison chart of the adaptive covariance inflation effect in embodiments of the present invention. Detailed Implementation

[0048] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0049] Example 1

[0050] This embodiment uses a river ecological slope protection project as an application scenario; the slope is approximately 500m long with a slope of approximately 1:2.5, employing a vegetation-type ecological concrete slope protection structure, with the surface covered with soil and mixed with vegetation such as Bermuda grass and Amorpha fruticosa. This embodiment aims to achieve real-time and accurate estimation of the displacement field, pore water pressure field, and volumetric water content field of the slope protection system using the method of this invention.

[0051] 1. Construct a state-space model of the slope protection system.

[0052] 1.1 Construction of dynamic equations

[0053] In this embodiment, a hydrological process model and a geotechnical mechanics model are coupled to construct a dynamic equation describing the hydro-mechanical coupling evolution of the slope protection system.

[0054] (1) Hydrological process model: The Richards equation, which describes the movement of unsaturated soil moisture, is adopted.

[0055]

[0056] Where θ is the soil volumetric water content; t is time; z is the vertical spatial coordinate (positive upwards); h is the pressure head; K(h) is the unsaturated permeability coefficient function, which is a function of h; and S(z,t) is the root water absorption term, which reflects the consumption of soil moisture by vegetation transpiration.

[0057] (2) Geotechnical model: The stress-strain-seepage coupling process of saturated / unsaturated soil is described using the Biot consolidation theory framework.

[0058] Equilibrium equations:

[0059] Where σ is the total stress tensor; ρ is the soil density; and g is the gravitational acceleration vector. In this embodiment, the constitutive relation of the soil adopts the modified Cambridge model that can take into account the influence of suction.

[0060] The two governing equations mentioned above are strongly coupled through the principles of pore water pressure (or suction) and effective stress, as well as the relationship between volume change and water content change, forming a closed set of equations that can be solved numerically using the finite element method.

[0061] 1.2 Definitions of State Vector, Observation Vector, and Observation Operator

[0062] (1) State vector X: The slope protection area is discretized into a finite element mesh containing approximately 2000 nodes; the state vector consists of the state variables of each node. In this embodiment, two horizontal displacements (u) of each node are selected. x ,u y The state vector consists of pore water pressure p and volumetric water content θ; therefore, the total dimension of the state vector is 2000 × 4 = 8000 dimensions. That is:

[0063]

[0064] (2) Observation vector Y and sensor network: such as Figure 3 As shown, a sensor network is deployed at key sections of the slope protection (such as the top, middle, and toe of the slope).

[0065] FBG fiber optic strain sensor: Deployed within the ecological concrete layer on the slope surface, with a total of 10 measuring points, to measure surface strain ε.

[0066] Soil multi-parameter sensor: buried in soil at different depths, with a total of 8 measuring points, to measure volumetric water content θ. obs And matrix suction (which can be converted into pore water pressure p) obs ).

[0067] Pore ​​water pressure gauges: Five gauges are deployed near the possible groundwater level at deeper depths to directly measure pore water pressure p. obs .

[0068] UAV-borne hyperspectral imager: Conduct aerial surveys once a month to obtain the spatial distribution map of the Normalized Difference Vegetation Index (NDVI) of the entire slope protection area.

[0069] Therefore, at a certain moment, the observation vector Y consists of data collected in real time (or near real time) by all the aforementioned sensors, and its dimension is equal to the number of all valid observation data.

[0070] (3) Observation operator H: Establishes a mapping from the high-dimensional state space (8000 dimensions) to the observation space.

[0071] For strain ε observation: observation operator H ε Using finite element shape functions and geometric equations, the nodal displacements (u) are expressed. x ,u y The strain value at the measuring point is obtained by interpolation and differentiation.

[0072] For water content θ obs and pore water pressure p obs Observation: Observation operator H θ and H p It is a simple selection operator that directly extracts the θ and p values ​​of the corresponding node positions from the state vector X.

[0073] For NDVI observations: Observation operator H NDVI Based on an empirical or semi-empirical vegetation growth model, the vegetation biomass information (or vegetation growth state related to water content and stress) implicit in the state vector is mapped to NDVI value; for example, a linear relationship of NDVI=a*B+b can be established, where B is the estimated biomass and a and b are empirical coefficients.

[0074] Ultimately, the overall observation operator H is the set of the aforementioned observation operators.

[0075] 2. State estimation is performed using ensemble Kalman filtering.

[0076] 2.1 Set Generation

[0077] Set the number of members in the set to N=100; at the initial time (t=0), generate 100 different initial state sets by using the key uncertain parameters of the perturbation model (such as the elastic modulus of the soil, the saturated permeability coefficient, and the water absorption intensity of the vegetation roots). The magnitude of the disturbance is based on the range of parameter uncertainties determined by field tests, laboratory tests, or engineering experience.

[0078] 2.2 Forecast Steps

[0079] Within the interval from time t−1 to time t, each set member As initial conditions, these are substituted into the hydrodynamic coupling equations (finite element model) constructed in Section 1.1 for numerical integration, with a time step Δt = 1 hour. After integration, the prior state set at time t is obtained. .

[0080] Calculate the mean and error covariance matrix of the prior state set:

[0081] Set mean:

[0082] Prior error covariance matrix:

[0083] 2.3 Analysis Step

[0084] Obtain the actual observation data Y at time t t Constructing the observation error covariance matrix R, it is usually assumed to be a diagonal matrix, with the elements on the diagonal representing the variance of the measurement errors of each sensor (such as the strain sensor error). Soil moisture sensor error wait).

[0085] Since the observation operator H may be nonlinear (especially for strain and NDVI), a set approximation is used to calculate the statistics of the observation space:

[0086] Mean of the observation set:

[0087] Prior error covariance of the observation space:

[0088] Cross-covariance between state and observation:

[0089] Calculate the Kalman gain matrix:

[0090] Update each set member (analysis step):

[0091] Among them, v iIt is a perturbation vector randomly sampled from a multidimensional normal distribution with a mean of 0 and a covariance matrix of R, used to preserve the statistical properties of the set.

[0092] 3. Adaptive covariance dilation technique is used to suppress filter divergence.

[0093] This is one of the core innovations of this method; after the analysis step, the observation increment is calculated:

[0094] The observation increment reflects the difference between actual observations and the model ensemble average forecast.

[0095] Online estimation of adaptive covariance inflation factor λ t :

[0096] in, The trace of a matrix (the sum of its diagonal elements); numerator It measures the portion of the observation increment that exceeds the observation error, and can be regarded as the manifestation of the model prediction error in the observation space; the denominator λ represents the uncertainty of the forecast ensemble in the observation space. The formula means that when the observation increment is significantly greater than the observation error and also greater than the dispersion of the forecast ensemble itself, it indicates that the model forecast may have a systematic bias or that the ensemble dispersion is underestimated. In this case, λ... t A positive value indicates that expansion is needed.

[0097] Application bloat:

[0098] The expansion factor is applied to the perturbation of the updated state set after the analysis step, thereby expanding the covariance of the analysis error. An effective approach is:

[0099] in, This is the updated mean of the set. The expanded set { The dispersion of the prediction step used for the next time step (t+1) is increased by 1+λ. t This multiplies the uncertainty in the model forecast in the next cycle, making the filter more confident in the observations and effectively suppressing divergence.

[0100] 4. Introducing space localization technology

[0101] To address the sparse observation problem in this embodiment, where the number of observation points (23 sensor points + monthly NDVI areal data) is far less than the number of state variables (8000 dimensions), spatial localization is introduced; in calculating the Kalman gain matrix K... t Previously, regarding the prior error covariance matrix... Localized processing is performed.

[0102] The Gaspari-Cohn function is used as the localization function ρ(d), where d is the spatial distance between the state variable grid point and the observation point; a localization matrix ρ is defined, whose elements ρ ij =ρ(d ij Then, the perturbations in the forecast set are locally corrected (equivalent to performing a Schur product on the covariance matrix):

[0103]

[0104] Where ∘ represents element-wise multiplication (Schur product); the Gaspari-Cohn function approaches 1 when the distance d is less than the localized scale L, and decays to 0 when d is greater than 2L. In this embodiment, based on the sensor deployment density and the slope protection size, L = 50m is chosen; this ensures that the information from an observation point mainly affects the state variables within a range of approximately 100m around it, suppressing spurious correlations over long distances.

[0105] 5. Output and Prediction

[0106] (1) State estimation output: Calculate the mean of the analysis state set after dilation. This serves as the optimal estimate of the slope protection system state at the current time t; the result includes the displacement field (u) of the entire slope protection area (2000 nodes) at the current time. x ,u y The complete spatial distribution of pore water pressure field p and volumetric water content field θ.

[0107] (2) Future state prediction: Estimating the optimal state at the current moment Using the initial conditions, the coupled hydraulic finite element model is run again, and the boundary conditions (such as rainfall forecast and water level changes) for a future period (e.g., 24 hours) are input to predict the slope protection status at future times. The prediction results can generate displacement cloud map evolution animation, potential slip surface location, safety factor time history curve, etc. When the predicted safety factor is lower than the preset threshold (e.g., 1.25), the system automatically triggers an early warning.

[0108] Figure 4 This embodiment demonstrates the complete closed-loop process of adaptive data assimilation and state estimation.

[0109] Example 2

[0110] This embodiment provides a specific implementation and deployment method for an ecological slope protection state estimation system based on adaptive data assimilation, corresponding to the method described in Embodiment 1.

[0111] like Figure 2 and Figure 3As shown, the system is physically composed of three main parts: a sensor network 101 deployed at the slope protection site, a data processing unit 102 located in the monitoring center or cloud, and a user-facing output unit 103. Data interaction is achieved through wired / wireless communication networks, forming a complete perception-computation-decision closed loop.

[0112] 1. Sensor Networks 101

[0113] Sensor network 101 is responsible for real-time or periodic acquisition of multi-source heterogeneous data, serving as the data foundation for state estimation. In this embodiment, the network adopts a hierarchical heterogeneous architecture:

[0114] Field layer (data acquisition):

[0115] Fiber Bragg grating (FBG) strain sensing subsystem: A distributed FBG demodulator (such as the MOI sm125 series) is selected as the main unit. A series of FBG strain sensors (such as the OS3100 series) are pre-embedded or pasted in the ecological concrete layer on the slope surface and connected in series to form a sensing optical cable. The sensors are deployed at key sections such as potential slip surfaces and stress concentration areas, with a spacing of about 20-30 meters. The demodulator calculates the micro-strain values ​​of each measuring point in real time at a frequency of 1Hz and outputs them through RS485 or Ethernet interface.

[0116] Soil multi-parameter sensing subsystem: Industrial-grade multi-parameter sensors supporting the Modbus protocol (such as Decagon5 or similar products) are selected to measure volumetric water content, temperature, and conductivity; the sensors are connected to the field data acquisition unit (RTU) via waterproof cables; eight sensors are buried at different depths (such as 0.5m, 1.0m, and 1.5m) at the top, middle, and bottom of the slope to obtain the vertical distribution of water content.

[0117] Pore ​​water pressure monitoring subsystem: Employs vibrating wire pore water pressure gauges (such as the Kicon 4500 series), which have stable signals and are suitable for long-term monitoring; Five measuring holes are arranged in the deep potential water level fluctuation zone and the top of the impermeable layer, and the sensors are connected to the same RTU via cables; The RTU has a built-in 4G DTU module, which periodically (e.g., every 10 minutes) converts the collected frequency mode values ​​into pressure values ​​and packages them for uploading to the cloud server.

[0118] Power supply and communication: Field sensors and RTUs are powered by solar panels combined with batteries; data is transmitted back to the monitoring center via LoRa, 4G or fiber optic networks.

[0119] Empty base layer (periodic isometric observation):

[0120] The UAV hyperspectral remote sensing subsystem: A monthly aerial survey is conducted using a multi-rotor UAV equipped with a hyperspectral imager (such as Headwall Nano-Hyperspec); ground control points are set up before the flight, and flight routes are planned to ensure that the image overlap is greater than 70%; after the flight, orthorectification and radiometric calibration are performed using software such as Pix4D to generate an NDVI distribution map with a spatial resolution better than 5cm; this NDVI raster data is used as areal observation data and imported into the data processing unit 102 through a dedicated interface.

[0121] 2. Data Processing Unit 102

[0122] The data processing unit 102 is the brain of the system, responsible for running the physical model and executing the assimilation algorithm; functionally, it mainly includes a processor 1021 and a memory 1022; this embodiment adopts a hybrid architecture of cloud server + edge computing to balance computing load and real-time requirements.

[0123] Hardware and basic software:

[0124] The core servers are cloud servers equipped with high-performance multi-core CPUs (such as Intel Xeon Gold series), large-capacity memory (≥128GB), and GPU accelerator cards (such as NVIDIA Tesla series) to run computationally intensive finite element models and ensemble predictions.

[0125] Deploy time-series databases (such as InfluxDB) to store massive amounts of high-frequency sensor time-series data, and relational databases (such as PostgreSQL) to store model parameters, assimilation configurations, and result snapshots.

[0126] The operating system uses a Linux distribution (such as Ubuntu Server), and the various functional microservices are deployed using Docker containerization technology, which facilitates management and expansion.

[0127] Core software modules:

[0128] Data access and preprocessing module: runs as a microservice, continuously monitoring data streams from 4G DTU and API interface; parses raw data, aligns timestamps, unifies units, detects and removes outliers, and writes it to the time series database; performs raster-to-vector processing on UAV NDVI data and spatially correlates it with the model mesh.

[0129] Multi-field coupled finite element model module: built based on COMSOL Multiphysics or the self-developed FEM kernel (written in C++); the model is coupled with the Biot theory according to the Richards equations described in Example 1 and driven by the API for parameterization; this module is encapsulated, can accept initial state, boundary conditions and parameter inputs, and return the predicted state field.

[0130] The adaptive ensemble Kalman filter assimilation engine is the core algorithm module of the system, developed using Python (NumPy, SciPy) or Julia. Its main submodules include: an ensemble manager responsible for generating, storing, and evolving the state vectors of 100 ensemble members; a forecast scheduler that, upon receiving new observation data, schedules the finite element model module to perform parallel integral forecasts for each ensemble member (accelerated using GPUs or clusters); an assimilation analyzer that performs online calculation of the adaptive covariance inflation factor λ and spatial localization (using the Gaspari-Cohn function). This module reads the forecast ensemble and observation data, calculates the Kalman gain, and updates and inflates the state ensemble; and a state updater that writes the analyzed optimal state estimate to the database and uses it as the initial condition for the next loop.

[0131] Workflow: The system is configured with scheduled tasks (e.g., triggered hourly); upon triggering, the data preprocessing module provides the latest observations, and the assimilation engine automatically executes the complete process of forecasting -> analysis -> adaptive dilation, and estimates the optimal state. The data is stored; subsequently, the prediction module uses the aforementioned estimate as the initial field and calls the model to predict the evolution over the next 24 hours.

[0132] 3. Output Unit 103

[0133] The output unit 103 is responsible for presenting the data processing results to the user in an intuitive and operable form.

[0134] Web visualization platform: A B / S architecture monitoring platform developed based on the Vue.js + ECharts framework. The platform's main functional interfaces may include:

[0135] Panoramic cockpit: Based on WebGIS (such as Cesium), it displays the 3D model of the slope protection and overlays real-time displacement cloud map, pore water pressure contour lines, and NDVI distribution map to achieve panoramic and dynamic visualization of the state field.

[0136] Safety Status Panel: Displays the displacement time history of key sections and the overall safety factor change curves in the form of curves and dashboards. When the predicted safety factor is lower than the threshold, the panel highlights an alarm.

[0137] Prediction and Early Warning Center: Displays animations of displacement evolution and risk zoning maps for the next 24 hours. Early warning information is released through pop-ups, audio, and banners on the platform's homepage.

[0138] Data management backend: Provides sensor management, model parameter configuration, assimilation strategy settings, and historical data query and export functions.

[0139] Mobile Apps and Push Notifications: Develop a supporting WeChat mini-program or lightweight app for inspection personnel. Simultaneously, integrate SMS, email, and DingTalk / WeChat Work robot APIs to automatically push information such as alert level, location, and time to pre-defined responsible parties when the system triggers an alert.

[0140] Data service interface: Provides an API interface that conforms to the RESTful specification, allowing state estimation results and prediction data to be securely connected to the superior smart water conservancy or smart construction site management platform to achieve data sharing and business collaboration.

[0141] The system described in this embodiment, by closely integrating specific sensor selection, communication scheme, computing architecture, software modules and the core method, forms a complete and implementable solution from physical world perception to information world decision-making, fully demonstrating the practicality of this system.

[0142] Example 3

[0143] This embodiment, based on Embodiment 1, specifically compares the effectiveness of the adaptive covariance expansion technique; it selects 30 consecutive days of monitoring data from the slope protection project and uses three different schemes for state estimation:

[0144] Option A (No Inflation): Use standard set Kalman filtering without any covariance inflation (λ≡0).

[0145] Option B (fixed dilation): Use standard set Kalman filtering and employ an empirical fixed dilation factor (λ≡0.1).

[0146] Solution C (the present invention): The adaptive covariance expansion technique described in Example 1 is adopted.

[0147] Based on the measured values ​​of a high-precision fiber optic strain sensor, the root mean square error (RMSE) of the three schemes for estimating the strain on the slope protection surface was calculated, and the results are as follows. Figure 5 As shown.

[0148] Scheme A (no inflation) performed reasonably well in the first few days, but starting from around day 5, the RMSE rose sharply, indicating that the filter diverged and the estimation results deviated completely from the true state.

[0149] Although scheme B (fixed expansion) avoids divergence and keeps the RMSE stable, the average error is large (the 30-day average RMSE is about 12.3 microstrain), indicating that the fixed expansion factor cannot adapt to the dynamic changes of model error and observation information, and the estimation accuracy is limited.

[0150] Scheme C (this method, adaptive expansion) remained stable throughout the 30 days and had the lowest RMSE (30-day average RMSE of approximately 7.8 microstrains). Compared to the fixed expansion scheme, the estimation accuracy was improved by approximately 36.6%. This demonstrates that the adaptive expansion technique can effectively suppress divergence and dynamically optimize the assimilation process according to the actual working conditions, thereby obtaining a more accurate and stable state estimate.

[0151] Example 4

[0152] The method and system described in Example 1 were applied to the long-term monitoring of the reservoir slope in a reservoir expansion and dredging project. The slope was equipped with 24 sensors, including surface displacement gauges, deep inclinometers, pore water pressure gauges, and soil moisture sensors. The method described in Example 1 was used to perform continuous state estimation and assimilation updates for a period of 6 months.

[0153] The results show that:

[0154] The estimation accuracy was significantly improved: through data assimilation, the average correlation coefficient between the calculated slope displacement and all inclinometer measured values ​​increased from 0.76 before assimilation to 0.94.

[0155] Successful early warning: The system successfully issued early warnings for two small, shallow landslides within six months, with advance warning times of 8 hours and 15 hours respectively, which bought valuable time for on-site personnel evacuation and emergency response.

[0156] Ecological status estimation: When a simplified vegetation growth model was coupled into the model and monthly UAV NDVI data was assimilated, the prediction error for vegetation cover was reduced from 22% before assimilation to 9%.

[0157] Example 5

[0158] To verify the applicability of the present invention in low-cost, low-density scenarios, this embodiment deliberately reduced the number of sensors to 5 (far lower than the conventional density) on a test slope section; applied the method described in Embodiment 1, and enabled spatial localization technology (localization scale L=30m).

[0159] The overall field state estimation results obtained by the method described in Example 1 under sparse observation (5 points) were compared with the reference field obtained by direct interpolation using a high-density sensor network 101 (20 points). The results show that at key sections, the estimation error of the displacement field is approximately 12.3%, and the estimation error of the volumetric water content field is approximately 15.1%. Although this accuracy is slightly lower than the result of high-density observation assimilation, it fully meets the daily needs of engineering monitoring. This proves that the method can still obtain a relatively reasonable overall field state estimate under sparse observation conditions through physical model constraints and localized assimilation, significantly reducing the hardware cost and deployment difficulty of the monitoring system.

[0160] In summary, the present invention provides an ecological slope protection state estimation method and system based on adaptive data assimilation. By deeply fusing multi-field coupled models and multi-source observation data, and innovatively introducing adaptive covariance expansion and spatial localization techniques, it effectively solves the problems of inaccurate models, easy data assimilation divergence, and difficulty in utilizing sparse observations in traditional methods. It achieves high-precision, high-robustness, real-time collaborative perception and prediction of ecological slope protection state, and has important theoretical value and broad engineering application prospects.

[0161] Finally, it should be noted that this article uses specific examples to illustrate the principles and implementation methods of the present invention. The above description of the embodiments is only for the purpose of helping to understand the core ideas of the present invention. Without departing from the principles of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the protection scope of the present invention.

Claims

1. A method for estimating the state of ecological slope protection based on adaptive data assimilation, characterized in that, Includes the following steps: Step S1: Construct a state-space model of the slope protection system. First, establish a dynamic equation describing the evolution of the slope protection system. The dynamic equation is coupled with at least two of the following: a hydrological process model, a geotechnical mechanics model, and a vegetation growth model. Then, based on the dynamic equation, construct the state vector X, the observation vector Y, and the observation operator H. Step S2: Based on the state-space model, the ensemble Kalman filter algorithm is used to fuse observation data from the sensor network in the slope protection area to estimate the state of the slope protection system in real time; Step S3: In the update process of the ensemble Kalman filter, the covariance inflation factor is estimated online using the adaptive covariance inflation technique, and the error covariance matrix is ​​inflated to suppress filter divergence. Step S4: Output the state estimation results of the slope protection system after the divergence suppression process in step S3.

2. The method for estimating the state of ecological slope protection based on adaptive data assimilation according to claim 1, characterized in that: In step S1, the state vector X includes at least two state variables from the displacement field, pore water pressure field, volumetric water content field, and vegetation biomass field; the observation vector Y corresponds to the observation data collected by the sensors deployed in the slope protection area, and includes at least two observation variables from strain, soil moisture, pore water pressure, and normalized vegetation index; the observation operator H describes the mapping relationship from the state vector X to the observation vector Y, i.e., Y=H(X)+ε, where ε is the observation error vector.

3. The method for estimating the state of ecological slope protection based on adaptive data assimilation according to claim 1 or 2, characterized in that: In step S1, the kinetic equations include at least two of the following: the Richards equation describing soil moisture movement, the equilibrium equation describing stress-strain relationship, and the kinetic equation describing vegetation growth.

4. The method for estimating the state of ecological slope protection based on adaptive data assimilation according to claim 1, characterized in that, The specific sub-steps for state estimation using ensemble Kalman filtering in step S2 include: S21: Generate N set members ,i=1,2,...,N, where each set member represents a possible state realization; S22: Based on the dynamic equation, predict each set member to obtain the prior state set at the current moment; S23: Calculate the mean of the prior state set. and the prior error covariance matrix P f ; S24: Obtain the actual observation data Y at the current moment. t and the observation error covariance matrix R; S25: Calculate the Kalman gain matrix K and update the members of each set to obtain the analysis state set { } 5. The method for estimating the state of ecological slope protection based on adaptive data assimilation according to claim 4, characterized in that, In step S3, the adaptive covariance inflation technique estimates the covariance inflation factor λ online using the following formula: in, Let R be the observation increment vector, representing the difference between the actual observations and prior estimates in the observation space; R is the observation error covariance matrix; H is the observation operator; P f The prior error covariance matrix; Represents the trace of a matrix.

6. The method for estimating the state of ecological slope protection based on adaptive data assimilation according to claim 4 or 5, characterized in that: Between steps S23 and S25, a step of spatial localization processing of the prior error covariance matrix Pf is also included. This processing is performed using a localization function. With matrix P f Implementing the Schur product operation: Where d is the spatial distance between the state variable and the observation point, the function The value of d decreases as d increases.

7. The method for estimating the state of ecological slope protection based on adaptive data assimilation according to claim 1, characterized in that, Following step S4, step S5 is also included: Using the state estimation results as initial conditions, the slope protection state at future times is predicted based on the dynamic equations. The prediction results include at least one of displacement field evolution, safety factor change, and vegetation coverage change.

8. An ecological slope protection state estimation system based on adaptive data assimilation, used to implement the method according to any one of claims 1-7, characterized in that, include: Sensor networks, deployed in the slope protection area, are used to collect multi-source heterogeneous observation data; A data processing unit includes a processor and a memory, the memory storing a computer program, which, when executed by the processor, is used to implement the method as described in any one of claims 1-7; The output unit is used to output the state estimation results, prediction results, and early warning information.

Images