Slope multi-physical field fusion early warning decision system based on digital twinning

Abstract

The application relates to the technical field of intelligent early warning of digital twinning, and particularly discloses a slope multi-physical field fusion early warning decision system based on digital twinning, physical monitoring data representing a macro state of a slope and micro physical response signals reflecting internal damage evolution are synchronously collected by a multi-modal data perception module; a damage intrinsic state extraction module is used for space-time alignment, standardization and cross-modal fusion analysis, and a unique intrinsic damage variable quantitatively representing the real-time deterioration degree of material strength is interpreted; a twinning body self-evolution module takes the variable as a core observation quantity, uses a data assimilation method to drive the collaborative dynamic update of slope mechanical model parameters and states, and realizes high-fidelity tracking of a physical reality by a digital model; and a forward-looking early warning decision module deduces a future space-time evolution path of the material strength parameter based on the calibrated model, and quantifies a risk probability of instability by combining Monte Carlo simulation, so that grading early warning and intelligent decision support are realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent early warning technology based on digital twins, specifically to a multi-physics field fusion early warning and decision-making system for slopes based on digital twins. Background Technology

[0002] The safety and stability of slope engineering is a key issue of long-term concern in civil engineering, water conservancy, transportation, and other fields. With the development of monitoring technology and numerical simulation, slope safety monitoring systems based on digital twins have become an important research direction. These systems construct virtual models of slopes in a spatial data environment and integrate monitoring data on physical fields such as surface displacement, deep deformation, and groundwater levels to achieve real-time visualization of slope conditions and a certain degree of safety assessment. Existing technologies typically rely on preset, fixed geotechnical parameters to establish numerical models and perform feedback analysis by comparing the differences between the model's calculated values ​​and the monitoring data.

[0003] The existing technology has the following shortcomings:

[0004] The core constitutive relations of the model are based on static or quasi-static material parameter assumptions, which cannot truly reflect the real-time degradation effect of material strength parameters (such as cohesion and internal friction angle) caused by the accumulation of damage such as the generation and propagation of micro-cracks within the soil and rock mass under external loads (such as rainfall and excavation). Because on-site monitoring methods cannot directly and in-situ measure the dynamic evolution values ​​of these key parameters during deformation, the internal state of the digital twin gradually becomes disconnected from physical reality. This disconnect makes it difficult for the model to accurately capture the critical damage state before slope instability, greatly limiting the foresight and reliability of the early warning system and preventing true disaster prediction. Summary of the Invention

[0005] The purpose of this invention is to provide a digital twin-based multiphysics fusion early warning and decision-making system for slopes to solve the problems mentioned above.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A digital twin-based multiphysics fusion early warning and decision-making system for slopes includes:

[0008] The multimodal data sensing module is used to synchronously acquire spatiotemporal continuous monitoring data of the slope, including: physical monitoring data characterizing macroscopic field variables and physical response signals characterizing internal damage evolution;

[0009] The damage eigenstate extraction module is used to perform cross-modal fusion analysis of physical response signals and physical monitoring data, including: spatiotemporal alignment and standardization of physical response signals and physical monitoring data; by mining their inherent spatiotemporal correlation patterns, coupling and mapping the nonlinear change sequence of microscopic response signals with the evolution trend of macroscopic physical field variables, and interpreting the unique intrinsic damage variables that quantitatively characterize the real-time deterioration degree of strength of soil and rock materials.

[0010] The twin self-evolution module uses intrinsic damage variables as core observations to drive the parameters and state of the slope mechanics model to evolve in a coordinated and dynamic manner, so that the material strength parameters inherent in the slope mechanics model can track the physical reality represented by the intrinsic damage variables in real time.

[0011] The forward-looking early warning decision module is based on the evolved slope mechanics model, which deduces the future spatiotemporal evolution path of material strength parameters, and provides predictive early warning and decision support for slope stability based on the instability risk probability of the future spatiotemporal evolution path.

[0012] As a further aspect of the present invention: the spatiotemporal alignment and standardization specifically include:

[0013] Establish a unified spatiotemporal framework based on the slope geological coordinate system; interpolate physical response signals and physical monitoring data from different sampling frequencies and spatial locations onto equally spaced spatiotemporal grid nodes under the unified spatiotemporal framework;

[0014] The interpolated grid node data is dedimensionalized to eliminate the influence of different physical dimensions on the fusion analysis.

[0015] The dimensionless processing converts all data into dimensionless values ​​with a mean of zero and a standard deviation of one.

[0016] As a further aspect of the present invention: the mining of its inherent spatiotemporal correlation patterns specifically includes:

[0017] Construct a dynamic spatiotemporal graph, where the nodes of the graph represent spatiotemporal grid nodes, the node attributes are the dimensionless data of the corresponding location, and the edges of the graph are determined by the spatial distance between nodes and the data correlation.

[0018] Learn the low-dimensional vector representation of nodes in dynamic spatiotemporal graphs. Low-dimensional vectors condense the correlation features of multimodal data in the spatiotemporal domain.

[0019] By analyzing the evolution and separation trends of low-dimensional vector clusters, we can identify macroscopic patterns that characterize the overall damage state of slopes, as well as local anomaly patterns that indicate the formation of potential sliding surfaces.

[0020] As a further aspect of the present invention: the process for obtaining the intrinsic damage variable is as follows:

[0021] Macroscopic patterns and local anomaly patterns are used as inputs and fed into a deep sequence-to-sequence prediction network;

[0022] The contribution of the prediction network to the evolution trend of macroscopic physical field variables through the nonlinear variation sequence of dynamically weighted microscopic response signals;

[0023] Within the prediction network, through multi-layer nonlinear transformation, the input spatiotemporal pattern is mapped to a low-dimensional, continuous intrinsic damage variable space, where each point in the space uniquely corresponds to a strength state of the slope soil and rock material.

[0024] Through mapping relationships, the real-time input data sequence is interpreted into a time series of unique intrinsic damage variables.

[0025] As a further aspect of the present invention: the parameters and states of the driving slope mechanical model undergo coordinated dynamic evolution, specifically including:

[0026] The element material strength parameters after discretization of the slope mechanics model and the physical state variables of the nodes are combined to form a dynamic state vector.

[0027] The intrinsic damage variable is used as an observation vector and applied to the spatial location corresponding to the dynamic state vector.

[0028] The observed information of intrinsic damage variables is fused with the theoretical predictions of the slope mechanics model;

[0029] Within each assimilation step, the material strength parameters and physical state variables in the dynamic state vector are simultaneously optimally estimated, thereby achieving the coordinated dynamic evolution of the two driven by the same damage observation information.

[0030] As a further aspect of the present invention: the construction process of the slope mechanics model is as follows:

[0031] The inputs to the slope mechanics model include: the geometric and topological information of the slope, the initial physical properties of the soil and rock mass, the boundary conditions, and the spatiotemporal field composed of intrinsic damage variables.

[0032] The internal structure of the slope mechanics model is a nonlinear finite element calculation framework that can reflect the water-mechanical coupling effect.

[0033] The output of the slope mechanics model is an updated dynamic state vector, which simultaneously includes the optimized unit material strength parameter field and the physical state fields of nodal displacement, stress, and pore water pressure.

[0034] The slope mechanics model enables the direct embedding of damage observations into the numerical computation kernel and outputs the complete mechanical state after co-evolution.

[0035] As a further aspect of the present invention: the real-time tracking of the physical reality represented by the intrinsic damage variable specifically includes:

[0036] A parametric evolution equation is defined for the material strength parameters in the dynamic state vector. The parametric evolution equation is driven primarily by the current value and rate of change of the intrinsic damage variable.

[0037] The predicted values ​​of the parameter evolution equation are continuously revised based on the latest observed intrinsic damage variables.

[0038] Through iterative correction, the calculated distribution of material strength parameters is made to maintain dynamic consistency with the intrinsic damage variable field interpreted from actual monitoring data.

[0039] As a further aspect of the present invention: the predicted spatiotemporal evolution path of the material strength parameters specifically includes:

[0040] Using the current slope mechanics model optimized by data assimilation as the initial state, load the boundary condition change sequence within a preset period in the future. The boundary conditions include rainfall, water level changes and seismic loads.

[0041] A fast solution is provided for the response of the slope mechanics model under future boundary conditions;

[0042] By solving quickly, we can obtain the spatial distribution prediction data of the material strength parameter field in the slope at different future times, thus forming a spatiotemporal evolution path of material strength parameters extending from the current time to a future preset time.

[0043] The spatiotemporal evolution path characterizes the dynamic decay process of slope material strength under future loads in the form of a parametric field sequence.

[0044] As a further aspect of the present invention: the predictive early warning and decision support for slope stability based on the probability of instability risk in future spatiotemporal evolution paths specifically includes:

[0045] Stability calculations were performed on the spatiotemporal evolution path of material strength parameters to obtain the variation curves of slope stability coefficients at future times.

[0046] Based on the uncertainty of future boundary conditions, a large number of possible evolution path samples are generated;

[0047] The proportion of samples with statistical stability coefficients below the critical threshold is quantified as the probability of instability risk at different future time periods.

[0048] Based on the lead time and probability value of the instability risk exceeding the preset alarm threshold, graded early warning information is issued, and corresponding engineering intervention measures are matched for different early warning levels.

[0049] The beneficial effects of this invention are:

[0050] (1) By constructing a slope mechanics model under a unified spatiotemporal framework, macroscopic field variables and microscopic response signals are spatiotemporally aligned and correlated. This effectively overcomes the limitations of traditional methods that isolate multi-source data analysis and make it difficult to reveal internal damage mechanisms. It interprets the intrinsic damage variables that can uniquely characterize the degree of strength deterioration of soil and rock materials, and significantly improves the accuracy of understanding the internal damage evolution process of slopes.

[0051] (2) Relying on digital twin technology, the intrinsic damage variables driven by measured data are dynamically integrated into the mechanical model parameter update process to ensure that the virtual model continuously approximates the physical reality; on this basis, the future evolution path is deduced and the probabilistic instability risk assessment is carried out to achieve the leap from qualitative judgment to quantitative prediction, and scientific decision support is provided by combining graded early warning thresholds and preset disposal strategies, thereby improving the timeliness, accuracy and operability of slope disaster early warning. Attached Figure Description

[0052] The invention will now be further described with reference to the accompanying drawings.

[0053] Figure 1 This is a flowchart of the system of the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Please see Figure 1 As shown, this invention is a digital twin-based multiphysics fusion early warning and decision-making system for slopes, comprising:

[0056] The multimodal data sensing module is used to synchronously acquire spatiotemporal continuous monitoring data of the slope, including: physical monitoring data characterizing macroscopic field variables and physical response signals characterizing internal damage evolution;

[0057] The damage eigenstate extraction module is used to perform cross-modal fusion analysis of physical response signals and physical monitoring data, including: spatiotemporal alignment and standardization of physical response signals and physical monitoring data; by mining their inherent spatiotemporal correlation patterns, coupling and mapping the nonlinear change sequence of microscopic response signals with the evolution trend of macroscopic physical field variables, and interpreting the unique intrinsic damage variables that quantitatively characterize the real-time deterioration degree of strength of soil and rock materials.

[0058] The twin self-evolution module uses intrinsic damage variables as core observations to drive the parameters and state of the slope mechanics model to evolve in a coordinated and dynamic manner, so that the material strength parameters inherent in the slope mechanics model can track the physical reality represented by the intrinsic damage variables in real time.

[0059] The forward-looking early warning decision module is based on the evolved slope mechanics model, which deduces the future spatiotemporal evolution path of material strength parameters, and provides predictive early warning and decision support for slope stability based on the instability risk probability of the future spatiotemporal evolution path.

[0060] In the multimodal data sensing module, which serves as the fundamental interface for information interaction between the system and the physical slope environment, its core function is to synchronously acquire multi-dimensional and heterogeneous component data that comprehensively reflects the slope's condition. Through the system-integrated sensor network, two types of data streams are continuously collected in time and space: the first type is physical monitoring data characterizing the macroscopic field variables of the slope, and the second type is physical response signals characterizing the evolution of damage within the soil and rock mass. Macroscopic field variable data describes the overall or large-scale physical state changes of the slope, such as the displacement field between the surface and deep layers, and the seepage field formed by pore water pressure. These data directly reflect the macroscopic response of the slope under external loads. Physical response signals reveal the changes in the microscopic structure of materials invisible to the human eye, such as the elastic wave signals released by frictional sliding between soil and rock particles, the generation and expansion of microcracks, and temperature changes distributed along the sensing path. These two types of data, from different scales, together constitute a comprehensive perception of the slope's health condition.

[0061] The data acquisition process begins with the systematic deployment of sensor networks. Acquiring macroscopic physical monitoring data relies on sensor arrays spatially optimized according to the slope's geological structure and potential slip surface characteristics. For example, GPS receivers and surveying robots are deployed at key points on the slope surface to capture the three-dimensional displacement field. Deep displacement monitoring is achieved through fixed inclinometer arrays installed in boreholes, distributed at preset intervals along the borehole depth to monitor horizontal displacement at different depths. Pore water pressure monitoring is accomplished through piezometers buried near potential slip zones and in different aquifers, thereby constructing the spatial distribution of groundwater level or pore water pressure. The spatial deployment scheme of these sensors aims to control the key geometric morphology and hydrogeological units of the entire slope.

[0062] The acquisition of physical response signals relies on specialized sensors sensitive to specific physical phenomena. Acoustic emission or microseismic sensors are installed on the slope surface or inside boreholes, forming a three-dimensional monitoring network to capture weak elastic vibration signals generated within the soil or rock mass due to damage events such as crack propagation or particle friction. These sensors are typically arranged in arrays, and spatial location of damage events can be achieved by analyzing the time difference of signals arriving at different sensors. Distributed fiber optic sensing systems lay sensing optical cables along specific paths on the slope surface or bury them within the fill. The optical cable itself serves as both a transmission medium and a sensor, continuously measuring strain and temperature changes at every point along the cable path, thereby obtaining one-dimensional continuously distributed strain and temperature field data.

[0063] Raw data collected by all sensors is aggregated through an integrated data transmission network. This network employs a hybrid wired and wireless approach to ensure reliable, low-latency transmission of data to the central processing unit. Data transmission protocols guarantee data integrity and timing consistency. For analyses requiring high-precision time correlation, the system uses a unified high-precision clock source to accurately timestamp all sensor data streams, forming the basis for subsequent spatiotemporal alignment of multimodal data. Data streams are transmitted continuously or intermittently at high frequency, creating a continuous spatiotemporal data sequence.

[0064] The multimodal data sensing module performs preliminary data preprocessing and integration. This includes basic quality checks on the raw data, such as identifying outliers or missing data due to sensor malfunctions or communication interference, and performing appropriate labeling or simple interpolation. More importantly, this module converts data collected from different sensors in various formats into a unified standardized data format and adds its corresponding spatial coordinate information to each data record. This process establishes a mapping relationship between the raw data and the three-dimensional spatial location of the slope, laying the necessary foundation for subsequent multi-source data fusion analysis under a unified geographic coordinate system. The standardized data stream, after preprocessing and integration, is provided in real time to the downstream damage eigenstate extraction module for in-depth analysis.

[0065] In the damage intrinsic state extraction module, which is the core calculation and analysis unit of the entire system, the task is to transform the raw data acquired by the multimodal data sensing module into a core indicator that can directly and quantitatively characterize the internal strength degradation state of the soil and rock materials. This module processes the physical response signal and physical monitoring data through a systematic cross-modal fusion analysis process. This process includes three main stages: spatiotemporal alignment and standardization, spatiotemporal correlation pattern mining, and intrinsic damage variable interpretation. Its ultimate goal is to output a unique intrinsic damage variable that changes over time, reflecting the real-time changes in the strength of the slope soil and rock materials.

[0066] Spatiotemporal alignment and standardization are the primary steps in multi-source data fusion. This step begins by establishing a unified spatiotemporal framework based on the actual geological coordinate system of the slope. This framework defines three-dimensional spatial coordinates and a unified time axis. Subsequently, physical response signals and physical monitoring data from different sensors with varying sampling frequencies and spatial locations are all mapped into this unified framework. Specifically, various time-series data are interpolated along the time dimension, for example, using cubic spline interpolation to resample non-equidistant sampling data into data points with equal time intervals. Spatially, the entire slope area is discretized into a three-dimensional, equidistant spatial grid, with each grid node representing a virtual observation point. Using the Kriging spatial interpolation algorithm, data from the actual sensor locations are interpolated onto each grid node, generating multiple physical field data regularly distributed within the unified spatiotemporal framework. After interpolation, the various physical quantities at each grid node are dedimensionalized to eliminate the magnitude influence of different physical dimensions such as displacement, pressure, and strain on subsequent data analysis. Specifically, the Z-score standardization method is adopted. For any physical quantity data, the average value across all grid nodes during the entire analysis period is subtracted, and then divided by its standard deviation. This transforms all the processed data into a dimensionless numerical sequence with a mean of zero and a standard deviation of 1. This process lays the foundation for subsequent direct comparisons and correlation analyses between different physical quantities.

[0067] After data preprocessing, the next stage is spatiotemporal correlation pattern mining. This stage aims to discover deep structural information hidden in standardized data. The technical approach is to construct a dynamic spatiotemporal graph. In this graph, each node corresponds to a grid node within a unified spatiotemporal framework, and the attribute features of a node are vectors composed of all types of dimensionless standardized data at that location. Connections between nodes are established based on two relationships: spatial proximity, where nodes that are spatially closer are connected, and the weight of the edge decreases as the distance increases; and data correlation, where the Pearson correlation coefficient of different node attribute vectors is calculated at different time steps, and node pairs with correlation exceeding a certain threshold are also connected, with the weight proportional to the absolute value of the correlation coefficient. This constructed graph structure captures both the spatial dependence and statistical correlation of the data. Next, graph neural network technology, especially graph attention networks, is used to learn the low-dimensional vector representation of each node in this dynamic graph. Graph attention networks use a message passing mechanism to allow each node to aggregate information from its neighbors and dynamically weight the importance of different neighbors using attention coefficients. After propagation through multiple network layers, each node ultimately obtains a low-dimensional vector embedding. This vector encapsulates the spatiotemporal correlation characteristics of the node and its local neighborhood's multimodal data. Finally, clustering algorithms, such as Gaussian mixture models, are applied to the low-dimensional vector embeddings of all nodes to analyze the evolution of cluster centers over time and the separation trends of node vectors from different categories. Stable clusters formed by the aggregation of node vectors typically characterize a certain stable state of the slope as a whole, while the movement of cluster centers and the emergence of new clusters reveal the evolution of damage. The trend of node vectors separating from the main clusters and forming small clusters is considered a local anomalous pattern; these anomalous nodes often spatially correspond to the incubation area of ​​potential sliding surfaces.

[0068] The final stage is the interpretation of intrinsic damage variables, aiming to transform the extracted macroscopic and local patterns into a continuous scalar index. This stage employs a deep sequence-to-sequence model based on an encoder-decoder structure. The model's input is a spatiotemporal pattern sequence within a continuous time window, composed of macroscopic pattern feature vectors and local anomaly pattern feature vectors. The encoder, typically composed of a long short-term memory network, progressively reads the input pattern sequence and compresses the entire sequence's information into a fixed-dimensional context vector containing the key spatiotemporal dependencies within the sequence. Inside the encoder, an attention mechanism is introduced, dynamically calculating the importance weights of patterns at different time steps in the input sequence when encoding each time step, thus explicitly modeling the contribution of the microscopic response signal sequence to the trend of macroscopic field variables. The decoder, also composed of a long short-term memory network, starts with the context vector and progressively decodes and outputs predicted intrinsic damage variable values ​​for several future time steps. During the training phase, the model's optimization objective is to ensure that the output intrinsic damage variable sequence optimally matches the material strength reduction coefficient obtained through offline numerical simulation or historical case inversion. Within the trained model, through multi-layer nonlinear transformations of the encoder, a mapping relationship is learned from a complex input spatiotemporal pattern to a low-dimensional, continuous intrinsic damage variable space. In practical applications, the spatiotemporal pattern sequence acquired and processed in real time is input into the trained model, which can directly interpret the unique intrinsic damage variable value corresponding to the current moment, thereby achieving continuous and quantitative tracking of the real-time degradation degree of soil and rock material strength.

[0069] In the twin self-evolution module, the core computing engine in the digital twin system enables the synchronous evolution of the virtual model and the physical entity. The module's core function is to use the intrinsic damage variables output by the damage eigenstate extraction module as the most critical system observations, and through a specific data assimilation algorithm, drive the coordinated dynamic update of the internal parameters and physical state of the slope mechanics model. Its goal is to enable the material strength parameters in the numerical model to track and reflect the true damage state of the soil and rock mass revealed by actual monitoring data in real time, thereby overcoming the prediction bias caused by constant parameters in traditional models. The module's operating mechanism comprises three closely interconnected components: the coordinated dynamic evolution of parameters and state, the construction of the slope mechanics model, and the real-time tracking of model parameters against physical reality.

[0070] The co-evolution of parameters and states is achieved through an ensemble Kalman filter, a data assimilation algorithm. The first step is to define the system's dynamic state vector. After spatial discretization of the slope mechanics model, the material strength parameters of each finite element, such as cohesion and internal friction angle, are combined with the physical state variables of each grid node, including displacement, stress, and pore water pressure, to form a high-dimensional dynamic state vector. This vector fully describes the mechanical state and material properties of the slope at a specific moment. The second step is to use intrinsic damage variables as observation vectors. Since intrinsic damage variables are field variables obtained through intelligent interpretation of monitoring data, they also have spatial distribution. Therefore, the intrinsic damage variable value at each spatial location is used as the observed value of the state vector at that location. The third step is the fusion process. At the beginning of each assimilation step, the state vector of the previous moment is predicted based on the theoretical equations of the slope mechanics model, obtaining the predicted state vector value for the current moment. Simultaneously, the observed values ​​of the intrinsic damage variables at the current moment are acquired. The ensemble Kalman filter algorithm generates an optimal Kalman gain matrix by calculating the covariance matrix between the predicted state vector and the observed vector. Using this gain matrix, the predicted and observed values ​​of the state vector are weighted and fused to obtain a statistically optimal estimate. This estimation process simultaneously applies to the material strength parameters and physical state variables in the state vector, thereby achieving coordinated adjustment and evolution of both driven by the same damage observation information.

[0071] The construction of the slope mechanics model provides the physical background and computational framework for this assimilation process. The model's input comprises four main parts. The first part is the slope's geometric topology information, namely the three-dimensional topographic surface, stratigraphic interfaces, and fault structures obtained through geological exploration and surveying. The second part is the initial physical properties of the soil and rock mass, including the density, elastic modulus, Poisson's ratio, and permeability coefficient of each soil and rock layer; these parameters constitute the initial conditions for the model's calculations. The third part is the boundary conditions, including stress boundaries and displacement constraints on the slope surface, as well as hydraulic boundaries such as rainfall infiltration, reservoir water level changes, and groundwater flow boundaries. The fourth part is the spatiotemporal field composed of intrinsic damage variables, which serves as the external input driving the model parameter updates. The model's internal structure is a nonlinear finite element computational framework capable of reflecting the fully coupled hydraulic-hydraulic effect. This framework is based on Biot's consolidation theory or similar coupled theories, simultaneously solving the equilibrium equations and the seepage continuity equations, and employing nonlinear constitutive models such as the Mohr-Coulomb or Drucker-Prag to describe the plastic behavior of the soil and rock mass. The model's output is the dynamic state vector updated by the data assimilation algorithm. This vector includes not only the optimized and corrected spatially varying element material strength parameter field, but also the corresponding nodal displacement field, stress field, and pore water pressure field. Through this construction method, the model directly embeds damage observation into the kernel of numerical calculation and outputs a physically self-consistent and complete mechanical state that matches the observation data.

[0072] Real-time tracking of the physical reality represented by intrinsic damage variables is achieved by introducing a parametric evolution equation within the data assimilation framework and continuously refining it. Specifically, a parametric evolution equation is defined for the material strength parameter in the dynamic state vector. This equation is not a fixed physical law, but rather an empirical time-series model, with the current value of the intrinsic damage variable and its rate of change over time serving as the primary driving terms on the right-hand side. For example, the parametric evolution equation can be expressed as the material strength parameter at the next time step equals the parameter value at the current time step, plus a correction term consisting of a linear or nonlinear combination of the current intrinsic damage variable and its rate of change. This equation describes how the parameter naturally evolves according to its current damage state in the absence of new observational data. In the update step of the data assimilation process, after obtaining the latest intrinsic damage variable observations, the algorithm compares the parameter values ​​predicted by the parametric evolution equation with the parameter requirements implied by the observations (established through observation operators). Based on the difference, the algorithm either adjusts the coefficients in the parametric evolution equation in reverse or directly corrects the predicted parameter values, making the corrected parameter estimates more consistent with the latest observational evidence. Through this iterative cycle of "prediction-update-correction" in each assimilation period, the spatial distribution of material strength parameters calculated in the numerical model can continuously approximate the real damage state represented by the intrinsic damage variable sites interpreted from actual monitoring data, thereby realizing real-time and dynamic tracking of model parameters to physical reality and ensuring the high fidelity of the digital twin throughout its entire life cycle.

[0073] In the forward-looking early warning decision-making module, which is an advanced application module in the digital twin system for realizing advanced risk perception and scientific decision support, the core function of this module is to extrapolate and analyze future scenarios based on a high-fidelity slope mechanics model that has been calibrated in real time by the twin's self-evolution module. Its workflow consists of two main stages: first, extrapolating the future spatiotemporal evolution path of material strength parameters, i.e., predicting the possible future state of the slope; and second, quantitatively assessing the probability of instability risk based on this predicted path, and generating early warning information and decision-making methods accordingly. This module extends the system's capabilities from precise perception of the current state to quantitative prediction of future risks, achieving true forward-looking early warning.

[0074] Extrapolating the future spatiotemporal evolution path of material strength parameters is a forward simulation process based on a physical model. This process uses the current-moment slope mechanics model, optimized by a data assimilation algorithm, as the initial state. This initial state not only includes the current optimal physical field distribution (such as displacement, stress, and pore water pressure fields), but more importantly, it includes the spatial distribution of material strength parameters that highly matches the actual situation. Subsequently, a sequence of boundary condition changes over a predetermined future period is loaded into the model. These boundary conditions are external factors driving the future evolution of the slope state, mainly including rainfall sequences (based on the intensity-duration distribution of weather forecasts), planned or predicted changes in reservoir water levels, and potential seismic loads (based on probabilistic seismic motion time histories from seismic hazard analysis). Next, the computational kernel of the slope mechanics model is invoked to quickly solve for the model's response under the future boundary condition sequence. In practical applications, validated model reduction techniques or surrogate models are often used to achieve rapid solutions. For example, a high-precision neural network surrogate model can be pre-trained using the full model; this surrogate model can approximate the slope response under given boundary conditions at a much faster speed than the full model. This rapid solution method allows for the prediction of the spatial distribution of material strength parameter fields within the slope at different future time points (e.g., 24 hours, 48 ​​hours, 72 hours, etc.). Combining this series of time-sequential parameter field data forms a spatiotemporal evolution path of material strength parameters extending from the current moment to a predetermined future moment. This path, in the form of a parameter field sequence, intuitively characterizes the dynamic decay process of slope material strength under expected external loads, revealing the evolution trend of potentially weak areas.

[0075] After obtaining the spatiotemporal evolution path of future material strength parameters, the next stage is the instability risk probability assessment and early warning decision-making phase. This phase begins with stability calculations of the evolution path. Specifically, the material strength parameter field for each future time point is extracted from the future spatiotemporal evolution path and used as static input. This is then substituted into the slope limit equilibrium analysis method or the strength-reduced finite element method to calculate the overall stability of the slope at that moment. This process is repeated for all preset future time points to obtain a curve showing the slope stability changing over time at each future moment. Next, to quantify the risk brought about by the uncertainty of future boundary conditions, the Monte Carlo stochastic simulation method is used. This method first generates a large number (e.g., 1000 or 10000) of possible future boundary condition sequence samples. These samples are obtained by randomly sampling key uncertainty parameters (such as total rainfall, peak intensity, and water level change amplitude) within their probability distribution range. For each randomly generated boundary condition sample, the path deduction and stability coefficient calculation process is repeated to generate a future stability coefficient evolution curve. This yields a probability distribution of future stability. Then, a critical threshold for the stability coefficient is set (e.g., 1.0 or 1.05). For a specific time period in the future (e.g., the next 24 to 48 hours), the number of samples in all Monte Carlo simulations that have their stability coefficient first fall below the critical threshold during that time period is counted. The ratio of this number to the total number of samples is quantified as the probability of slope instability during that time period.

[0076] Finally, early warnings and decisions are made based on the calculated probability of instability risk. The system presets multiple risk probability alarm thresholds for different levels. For example, a risk probability below 50% can be defined as a concern level (blue warning), a probability between 50% and 20% as a warning level (yellow warning), a probability between 20% and 50% as a alert level (orange warning), and a probability above 50% as a red warning. The issuance of early warning information is based not only on the magnitude of the risk probability value but also on its time attribute, such as the time in advance when the warning event may occur. The system determines the urgency of the warning based on the earliest time point corresponding to the risk probability that triggers different levels of warnings. Simultaneously, for each warning level, the system matches corresponding engineering intervention measures from a preset decision knowledge base. For example, for a yellow warning, manual patrols are strengthened; for an orange warning, the high-frequency mode of the automated monitoring system is activated and emergency supplies are prepared; for a red warning, emergency plans are activated and personnel evacuation is organized. This probability- and time-based hierarchical early warning and decision-making mechanism provides managers with scientific, quantitative, and operable decision support, significantly improving the accuracy and proactivity of slope safety management.

[0077] The working principle of this invention is as follows: A multimodal data sensing module synchronously acquires physical monitoring data characterizing the macroscopic state of a slope and physical response signals reflecting the evolution of internal damage, achieving continuous spatiotemporal data acquisition and preprocessing. A damage eigenstate extraction module performs spatiotemporal alignment, standardization, and cross-modal fusion analysis on the multi-source data. Dynamic spatiotemporal graph modeling and graph neural networks are used to mine the coupled evolution characteristics of microscopic signals and macroscopic field variables. A deep learning model with an encoder-decoder structure decodes a unique intrinsic damage variable that quantitatively characterizes the real-time deterioration of the strength of soil and rock materials. This intrinsic damage variable serves as the core observation input twin self-evolution module, driving... The slope mechanics model, constructed based on the hydro-mechanical coupled finite element theory, employs ensemble Kalman filtering to achieve coordinated dynamic updates of material strength parameters and physical state variables, enabling the digital model to accurately track physical reality. The forward-looking early warning decision module, based on the calibrated high-fidelity model, performs forward extrapolation in conjunction with future boundary conditions (such as rainfall and water level changes) to generate the future spatiotemporal evolution path of material strength parameters. It also quantifies the probability of instability risk at different time points through Monte Carlo simulation, implements graded early warning based on preset thresholds, and outputs corresponding emergency responses in conjunction with the decision knowledge base. This constructs a closed-loop intelligent early warning system from perception, cognition, modeling to prediction and decision-making.

[0078] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A slope multi-physics fusion early warning and decision-making system based on digital twins, characterized in that, include: A multimodal data sensing module is used to synchronously acquire spatiotemporal continuous monitoring data of the slope, including: physical monitoring data characterizing macroscopic field variables and physical response signals characterizing internal damage evolution; The damage eigenstate extraction module is used to perform cross-modal fusion analysis on physical response signals and physical monitoring data, including: spatiotemporal alignment and standardization of physical response signals and physical monitoring data; by mining their inherent spatiotemporal correlation patterns, coupling and mapping the nonlinear change sequence of microscopic response signals with the evolution trend of macroscopic physical field variables, and interpreting the unique intrinsic damage variables that quantitatively characterize the real-time deterioration degree of strength of soil and rock materials. The process of obtaining the intrinsic damage variables is as follows: Macroscopic patterns and local anomaly patterns are used as inputs and fed into a deep sequence-to-sequence prediction network; The contribution of the prediction network to the evolution trend of macroscopic physical field variables through the nonlinear variation sequence of dynamically weighted microscopic response signals; Within the prediction network, through multi-layer nonlinear transformation, the input spatiotemporal pattern is mapped to a low-dimensional, continuous intrinsic damage variable space, where each point in the space uniquely corresponds to a strength state of the slope soil and rock material. Through mapping relationships, the real-time input data sequence is interpreted into a time series of unique intrinsic damage variables; The twin self-evolution module uses intrinsic damage variables as core observations to drive the parameters and state of the slope mechanics model to evolve in a coordinated and dynamic manner, so that the material strength parameters inherent in the slope mechanics model can track the physical reality represented by the intrinsic damage variables in real time. The forward-looking early warning decision module is based on the evolved slope mechanics model, which deduces the future spatiotemporal evolution path of material strength parameters, and provides predictive early warning and decision support for slope stability based on the instability risk probability of the future spatiotemporal evolution path.

2. The slope multiphysics fusion early warning and decision-making system based on digital twin as described in claim 1, characterized in that, The spatiotemporal alignment and standardization specifically include: Establish a unified spatiotemporal framework based on the slope geological coordinate system; interpolate physical response signals and physical monitoring data from different sampling frequencies and spatial locations onto equally spaced spatiotemporal grid nodes under the unified spatiotemporal framework; The interpolated grid node data is dedimensionalized to eliminate the influence of different physical dimensions on the fusion analysis. The dimensionless processing converts all data into dimensionless values ​​with a mean of zero and a standard deviation of one.

3. The slope multiphysics fusion early warning and decision-making system based on digital twin as described in claim 1, characterized in that, The mining of its inherent spatiotemporal correlation patterns specifically includes: Construct a dynamic spatiotemporal graph, where the nodes of the graph represent spatiotemporal grid nodes, the node attributes are the dimensionless data of the corresponding location, and the edges of the graph are determined by the spatial distance between nodes and the data correlation. Learn the low-dimensional vector representation of nodes in dynamic spatiotemporal graphs. Low-dimensional vectors condense the correlation features of multimodal data in the spatiotemporal domain. By analyzing the evolution and separation trends of low-dimensional vector clusters, we can identify macroscopic patterns that characterize the overall damage state of slopes, as well as local anomaly patterns that indicate the formation of potential sliding surfaces.

4. The slope multiphysics fusion early warning and decision-making system based on digital twin as described in claim 1, characterized in that, The parameters and states of the driving slope mechanical model undergo coordinated dynamic evolution, specifically including: The element material strength parameters after discretization of the slope mechanics model and the physical state variables of the nodes are combined to form a dynamic state vector. The intrinsic damage variable is used as an observation vector and applied to the spatial location corresponding to the dynamic state vector. The observed information of intrinsic damage variables is fused with the theoretical predictions of the slope mechanics model; Within each assimilation step, the material strength parameters and physical state variables in the dynamic state vector are simultaneously optimally estimated, thereby achieving the coordinated dynamic evolution of the two driven by the same damage observation information.

5. The slope multiphysics fusion early warning and decision-making system based on digital twin as described in claim 4, characterized in that, The process of constructing the slope mechanics model is as follows: The inputs to the slope mechanics model include: the slope's geometric topology information, the initial physical properties of the soil and rock mass, boundary conditions, and the spatiotemporal field composed of intrinsic damage variables; The internal structure of the slope mechanics model is a nonlinear finite element calculation framework that can reflect the water-mechanical coupling effect. The output of the slope mechanics model is an updated dynamic state vector, which simultaneously includes the optimized unit material strength parameter field and the physical state fields of nodal displacement, stress, and pore water pressure. The slope mechanics model enables the direct embedding of damage observation into the numerical calculation kernel and outputs the complete mechanical state after co-evolution.

6. The slope multiphysics fusion early warning and decision-making system based on digital twin as described in claim 1, characterized in that, The real-time tracking of the physical reality represented by the intrinsic damage variables specifically includes: A parameter evolution equation is defined for the material strength parameter in the dynamic state vector. The parameter evolution equation is driven primarily by the current value and rate of change of the intrinsic damage variable. The predicted values ​​of the parameter evolution equation are continuously revised based on the latest observed intrinsic damage variables. Through iterative correction, the calculated distribution of material strength parameters is made to maintain dynamic consistency with the intrinsic damage variable field interpreted from actual monitoring data.

7. The slope multiphysics fusion early warning and decision-making system based on digital twin as described in claim 1, characterized in that, The predicted future spatiotemporal evolution path of the material strength parameters specifically includes: Using the current slope mechanics model optimized by data assimilation as the initial state, load the boundary condition change sequence within a preset period in the future. The boundary conditions include rainfall, water level changes and seismic loads. A fast solution is provided for the response of the slope mechanics model under future boundary conditions; By solving quickly, we can obtain the spatial distribution prediction data of the material strength parameter field in the slope at different future times, thus forming a spatiotemporal evolution path of material strength parameters extending from the current time to a future preset time. The spatiotemporal evolution path characterizes the dynamic decay process of slope material strength under future loads in the form of a parametric field sequence.

Images