Intelligent monitoring system for dike disturbance based on river-crossing bridge

Abstract

The invention relates to an intelligent monitoring system for embankment disturbance based on a river-crossing bridge, in particular to the field of water conservancy projects, and realizes full-flow accurate perception and mechanism description of embankment disturbance caused by bridge construction through a multi-source data fusion and dynamic digital twinning technology. In combination with causal reasoning analysis, key disturbance factors and influence intensity thereof can be identified from complex working conditions, and internal causal laws between construction activities and levee responses are disclosed; and finally, depending on prospective deduction and networked risk entropy evaluation, the crossing from passive monitoring to active early warning is realized, the early recognition and pre-control capability of the safety risk of the dam structure is remarkably improved, and intelligent decision support is provided for construction safety and engineering protection.

Description

Technical Field

[0001] This invention relates to the field of water conservancy engineering, and more specifically, to a smart monitoring system for the disturbance of a levee to a cross-river bridge. Background Technology

[0002] During the foundation construction phase of cross-river bridges, especially in soft soil foundations or high-water river areas, construction activities pose a significant threat to the stability of the embankment structure. This scenario typically involves complex engineering geological and hydrological conditions, such as loose soil structure and high permeability, coupled with seasonal fluctuations or even short-term rapid changes in river water levels. Key construction procedures, including but not limited to pile driving for bridge foundations, large-scale excavation of abutment pits, and frequent movement of heavy construction machinery, apply continuous dynamic loads and vibration excitations to the embankment and foundation. Construction factors and external environmental factors, such as changes in groundwater levels caused by rainfall infiltration and the scouring effect of increased river flow on the embankment, are intertwined and coupled, which can easily disrupt the original stress balance of the embankment. Such disturbances may induce displacement of the embankment beyond the allowable range, uneven settlement, or even cause local instability risks such as the expansion of cracks in the embankment or piping leakage, directly threatening the safety and durability of the entire embankment project. Therefore, in this specific scenario, achieving real-time and accurate monitoring and safety early warning of construction disturbance behavior is the core link to ensure the smooth progress of the project and the safety of the surrounding environment. Currently, the industry's commonly used technologies for monitoring the stability of embankment structures during bridge construction primarily rely on traditional fixed-point monitoring equipment, such as mechanical settlement meters, fixed inclinometers, or periodic manual measurements based on total stations. These solutions typically have several significant drawbacks. First, the data acquisition frequency is low, often involving intermittent measurements on a daily or weekly basis, making it difficult to capture dynamic responses that change at the minute or even second level during construction, thus failing to meet real-time requirements. Second, the monitoring parameters are relatively limited, mostly focusing on macroscopic displacement or tilt angle measurements, lacking simultaneous acquisition and fusion analysis of key multi-physics parameters such as vibration acceleration, soil pore water pressure, and deep horizontal displacement, resulting in a lack of understanding of the disturbance mechanism. Thirdly, the existing technology system lacks comprehensive understanding. It mainly relies on preset fixed thresholds for alarms and lacks data-driven intelligent early warning models. It cannot identify potential instability risks in the early stage or predict trends. It can only issue an alarm after the deformation reaches a certain level, which is a post-event remedial measure and lacks foresight. Fourthly, the sensor deployment scheme often fails to achieve full coverage of key sections and potentially weak areas of the dike, resulting in monitoring blind spots. Therefore, the current technology level urgently needs a solution that can achieve high-frequency, multi-parameter, full-coverage real-time monitoring and intelligent early warning through advanced data analysis methods to make up for the shortcomings of the existing technology in terms of real-time performance, comprehensiveness, and intelligence. Summary of the Invention

[0003] This invention addresses the technical problems existing in the prior art by providing an intelligent monitoring system for the disturbance of a river-crossing bridge to a levee. The system utilizes a data fusion governance module, a digital twin construction module, a causal reasoning analysis module, and an early warning decision-making module to solve the problems mentioned in the background.

[0004] The technical solution of this invention to solve the above-mentioned technical problems is as follows: specifically including: a data fusion and governance module, a digital twin construction module, a causal reasoning and analysis module, and an early warning and decision-making module, wherein; Data fusion and governance module: It is used to synchronously collect dam response data, construction activity data and environmental data measured by sensors deployed at key sections of the dam and bridge construction points through the Internet of Things sensor network to form a multi-source heterogeneous data stream. It also uses a data fusion algorithm based on spatiotemporal dual registration to clean and align the multi-source heterogeneous data stream in real time, and generate a spatiotemporally synchronized multi-dimensional data cube. Digital twin building module: It is used to receive a multidimensional data cube and embed a simplified physical model with parameters that can be updated in real time. It uses data assimilation technology to compare and calibrate the monitoring data in the multidimensional data cube with the prediction results of the simplified physical model, and dynamically updates the parameters of the simplified physical model to build a dynamic digital twin that is consistent with the state of the physical dam. Causal Reasoning Analysis Module: Based on the structural causal model framework, this module analyzes the changes in the conditional probability distribution of dam state variables represented by dynamic digital twins before and after construction intervention, identifies the causal links between construction activities and dam response, and establishes a causal graph model that quantifies the effects of causal links. Early warning decision module: used to couple the causal graph model and the dynamic digital twin. For the construction activities corresponding to the causal links identified by the causal reasoning analysis module, it performs multi-condition forward-looking simulation in the dynamic digital twin, calculates the overall risk entropy based on complex network theory, and generates forward-looking early warning information based on the evolution trend of the overall risk entropy. In a preferred embodiment, the data fusion and governance module includes the dam response data, which includes vibration acceleration collected by vibration sensors, micro-deformation collected by tilt sensors, and pore water pressure collected by pore water pressure gauges; the construction activity data includes pile driving energy, excavation depth, and machinery location information obtained in real time through the construction management system; and the environmental data includes river water level and flow velocity information provided by hydrological stations. The raw readings of vibration acceleration, micro-deformation, pore water pressure, pile driving energy, excavation depth, machine position, river water level and flow velocity constitute the raw observations in the multi-source heterogeneous data stream.

[0005] In a preferred embodiment, the specific operation of using a spatiotemporal dual-registration-based data fusion algorithm to perform real-time cleaning and alignment of multi-source heterogeneous data streams is as follows: First, for each raw reading in the raw observations, the calibration function of the corresponding sensor is applied to convert the raw reading into an engineering value, thus completing the initial standardization of the raw observations. Subsequently, a dual registration of time and space is performed. Time registration involves attaching a time identifier based on a high-precision clock synchronization protocol to each engineering value after initial standardization, so that all data are unified on the same time reference axis. Spatial registration is the process of assigning each sensor node a unique three-dimensional location identifier within the global coordinate system of the embankment. Ultimately, a set of discrete data points with spatiotemporal labels is formed, defined by time markers, three-dimensional location markers, and corresponding engineering values.

[0006] In a preferred embodiment, the process of generating a spatiotemporally synchronized multidimensional data cube specifically involves: After obtaining a set of discrete data points with spatiotemporal labels, a data fusion method based on physical field reconstruction theory is adopted. This method operates on a continuous spatial distribution field representing the physical state of a dam to be reconstructed. The method includes constructing an optimization objective function and solving the function to obtain the optimal continuous spatial physical field. The objective function is defined as the sum of a data fitting term and a physical regularization term, whereby the data fitting term is calculated as follows: For each data point in the discrete data point set, calculate the square of the difference between the estimated value of the continuous spatial distribution field to be reconstructed at the three-dimensional location marker of the data point and the engineering value recorded at the data point. Then multiply this squared difference by a weighting coefficient pre-assigned to the corresponding sensor of the data point. Finally, sum the above calculation results for all data points. The physical regularization term is calculated as follows: For the continuous spatial distribution field to be reconstructed, find the integral of the square of the Laplace operator of the field over the entire defined dam space domain; The sum of the data fitting term and the physical regularization term is obtained by scaling the physical regularization term with a regularization parameter and then adding it to the unscaled data fitting term. By adjusting the regularization parameter and solving for the physical field spatial distribution that minimizes this weighted sum, the optimal continuous spatial physical field is obtained. For each processing time point, the corresponding continuous spatial physical field is reconstructed through the above process. Then, the continuous spatial physical fields obtained from different time points are discretized and their values ​​are taken at each node of a predefined regular three-dimensional spatial grid. The field values ​​from different physical states at the same time point and the same spatial grid node are combined into a multidimensional vector. Then, the multidimensional vectors on all grid nodes are arranged and organized in chronological order to generate a spatiotemporally synchronized multidimensional data cube.

[0007] In a preferred embodiment, the specific operation of embedding a simplified physical model with real-time updatable parameters and receiving a multidimensional data cube in the digital twin construction module is as follows: First, a simplified physical model is defined as a finite element mechanical model of a dam based on the linear elastic assumption. The dynamic behavior of this model is described by the mass matrix, damping matrix, and stiffness matrix, and is driven by the external load vector. The dynamic response of the dam finite element mechanical model is manifested as displacement vector, velocity vector, and acceleration vector. The basic relationship between them is: the sum of the product of the mass matrix and the acceleration vector, the product of the damping matrix and the velocity vector, and the product of the stiffness matrix and the displacement vector is equal to the external load vector. Next, a state vector is defined, which is a column vector composed of two parallel parts: the first part is the transpose of the key model parameter set to be dynamically updated in the simplified physical model, and the second part is the transpose of some key state variables in the finite element mechanical model of the dam; the key model parameter set includes the equivalent elastic modulus parameter reflecting the stiffness of the dam soil, and the key state variables include the stress vector and displacement vector of the key nodes of the dam. The multidimensional data cube provides initial boundary conditions, load conditions, and observation data sources for comparison to simplify the physical model; the process of dynamically updating the parameters of the simplified physical model is the process of dynamically updating the key model parameter set in the state vector.

[0008] In a preferred embodiment, the specific process of comparing and calibrating the monitoring data in the multidimensional data cube with the prediction results of the simplified physical model using data assimilation technology to construct a dynamic digital twin is as follows: The data assimilation technique is implemented using an ensemble Kalman filter algorithm, whose processing flow includes iterative prediction and correction steps. In the prediction step, the optimal state vector estimate obtained from the correction step is used as the initial condition for the current prediction. The initial condition is input into the model evolution operator, which represents the dynamic evolution law of the simplified physical model, to calculate a preliminary predicted value of the state vector at the current moment. This preliminary predicted value is then superimposed with a model process noise vector, which characterizes the inaccuracy of the simplified physical model itself, to finally obtain the predicted value of the state vector at the current moment. Based on this predicted value of the state vector, the predicted value of the observed physical quantity at the current moment is calculated through an observation operator. The predicted value of the observed physical quantity refers to the displacement and acceleration values ​​at the sensor deployment location predicted by the simplified physical model. In the calibration step, the current real monitoring data is extracted from the multidimensional data cube. The real monitoring data refers to the actual measured values ​​of the dam response data corresponding to the predicted values ​​of the observed physical quantities in the multidimensional data cube. The calibration step further includes the following sub-steps: A1. Generate a set containing multiple members based on the predicted values ​​of the state vector; A2. Calculate the Kalman gain matrix, which is determined in the following way: The covariance matrix of the predicted state is calculated based on set calculation, and is combined with the observation operator and its transpose, as well as an observation noise covariance matrix that characterizes the uncertainty of the observation data. A3. Perform state update, that is, use the Kalman gain matrix to correct the predicted state vector of each member in the set. The correction amount is proportional to the difference between the actual monitoring data and the predicted observation value corresponding to each member, so as to obtain the corrected state vector of each member. The mean of the corrected state vectors of all members is the optimal state vector estimate after correction at the current time. This process is repeated at each time step, thereby continuously updating the parameters of the simplified physical model. The continuously updated set of key model parameters, together with the key state variables in the current corrected optimal state vector estimate, constitute a dynamic digital twin consistent with the state of the physical dam.

[0009] In a preferred embodiment, the specific operation of analyzing the conditional probability distribution changes of the dam state variables represented by the dynamic digital twin before and after the construction intervention in the causal reasoning analysis module to identify the causal link is as follows: First, we define two types of core variables. The first type is intervention variables, which are vectors composed of key construction parameters extracted from construction activity data. Key construction parameters include piling energy and excavation depth. The second category is outcome variables, which are key state variables that characterize the dam response, extracted from the dam state output by the dynamic digital twin and corrected for data assimilation. Subsequently, based on historical data, a dataset is constructed, which consists of multiple data points. Each data point contains the value of the intervention variable at a time point and the value of the outcome variable at the same time point. Subsequently, a causal discovery algorithm based on conditional independence test was used to process the dataset. The goal of this algorithm is to construct a directed acyclic graph that represents the topological structure of causal relationships between variables. The directed acyclic graph (DAG) consists of a set of nodes and a set of edges. The set of nodes includes intervention variables, outcome variables, and environmental confounding variables that simultaneously affect both intervention and outcome variables. Environmental confounding variables include river water level and flow velocity. The set of edges represents the direction of direct causal influence between node variables. The causal discovery algorithm uses statistical tests to determine the conditional independence relationship between variables, gradually eliminates unnecessary correlation edges between variables, initially determines the skeleton structure of the causal graph, and then uses causal directionality rules to determine the direction of the edges in the graph, finally outputting the causal link topology structure, i.e., the directed acyclic graph.

[0010] In a preferred embodiment, the specific process of establishing the causal graph model for quantifying causal link effects is as follows: Based on the identification of the causal link topology, the effect strength of each causal link is further quantified; the quantification process is based on intervention causal reasoning theory, and for each causal link in the directed acyclic graph that points from an intervention variable to an outcome variable, its average causal effect value is calculated. The calculation process for the average causal effect value is as follows: First, determine a sufficient adjustment set based on the directed acyclic graph; Then, the average causal effect is calculated as follows: Iterate through all possible combinations of variable values ​​in the adjustment set. For each combination of values, calculate the conditional expectation of the outcome variable when the intervention variable takes the first preset value and the conditional expectation of the outcome variable when the intervention variable takes the second preset value, and calculate the difference between the two conditional expectations. Finally, all these differences are weighted and averaged according to the probability of the combination of values ​​of the adjustment set variable. The result is the average causal effect value of the causal link. Finally, the directed acyclic graph representing the topology of causal links is combined with the calculated set of average causal effect values ​​that store the strength of each causal link effect to construct a causal graph model that quantifies causal link effects.

[0011] In a preferred embodiment, the specific operation of the early warning decision module in performing multi-condition forward-looking simulation for the construction activities corresponding to the causal links identified by the causal reasoning analysis module is as follows: First, all causal links are extracted from the causal graph model, and causal links with an average causal effect value exceeding a preset threshold are selected. The construction activities corresponding to these links are then identified as key construction activities. Subsequently, based on the current construction status and the expected construction plan, a set of possible future scenarios are defined; each scenario includes the trajectory of changes in one or more key construction activity parameters and environmental parameters over a period of time in the future. The key construction activity parameters include pile driving energy, and the environmental parameters include river water level. Subsequently, the current dynamic digital twin is used as the initial state for the simulation. The change trajectories of key construction activity parameters and environmental parameters defined in each simulation scenario are used as external inputs to drive the calibrated simplified physical model in the dynamic digital twin to perform forward numerical simulation. The forward numerical simulation uses a numerical integration algorithm to start from the current displacement and stress key state variables of the dam after data assimilation and correction, and performs numerical integration calculations over a future period to predict the dynamic response sequence of the dam's displacement and stress key state variables over time. The numerical integration algorithm uses the Runge-Kutta method for calculation, predicting the evolution trajectory from the current state to the future state through multi-step iteration. This process is executed independently once for each defined simulation scenario, thereby obtaining a set of future state response sequences of the dam corresponding to different future possibilities, i.e., multi-condition forward simulation results.

[0012] In a preferred embodiment, the specific process of calculating the overall risk entropy based on complex network theory and generating forward-looking early warning information according to the evolution trend of the overall risk entropy is as follows: First, the finite element mechanical model of the dam on which the dynamic digital twin is based is abstracted into a complex network, where each finite element element is regarded as a node in the network, and the elements that share nodes or boundaries are connected to each other, thereby constructing the dam structure network. Subsequently, for the future state response sequence of the dam under each working condition obtained by forward-looking simulation, the failure probability of each node in the dam structure network at each future time is calculated. Then, the overall risk entropy is calculated based on complex network theory. The calculation process includes assigning a weight coefficient representing the topological importance of each node in the dam structure network. The specific calculation method for the overall risk entropy is as follows: For a given future moment under the scenario of interest, the calculation of the overall risk entropy follows these steps: The first step is to traverse every node in the dam structure network. For each node, calculate the product of the node's failure probability at the current moment and the node's weight coefficient to obtain the weighted failure probability of the node. The second step is to calculate the natural logarithm of the weighted failure probability of the node obtained in the first step. The third step is to multiply the weighted failure probability of the node obtained in the first step by the natural logarithm of the weighted failure probability calculated in the second step to obtain the contribution of the node to the overall risk entropy at the current moment. The fourth step is to sum the contribution values ​​of all nodes in the dam structure network at the current moment; Fifth, take the negative value of the summation result obtained in the fourth step. The final value is the overall risk entropy of the simulated working condition at that future moment. Finally, the trend of the overall risk entropy over time under each simulated working condition is analyzed. The trend analysis includes calculating the rate of change of the overall risk entropy over time. If the value of the overall risk entropy exceeds a preset threshold or its rate of change exceeds a preset growth rate threshold, an early warning level will be determined. The early warning level will be determined by comprehensively considering the value of the overall risk entropy, the rate of change, and the importance of the corresponding key construction activity parameters, and will be classified as blue, yellow, orange, or red. Based on the determined warning level, specific forward-looking warning information is generated. The warning information includes a description of the triggered working condition, the current value and predicted peak value of the risk entropy, and suggested intervention measures. Finally, structured and actionable forward-looking warning information is output.

[0013] The beneficial effects of this invention are as follows: By using multi-source data fusion and dynamic digital twin technology, it achieves accurate perception and mechanism characterization of the entire process of dam disturbance caused by bridge construction; combined with causal reasoning analysis, it can identify key disturbance factors and their influence intensity from complex working conditions, revealing the inherent causal law between construction activities and dam response; finally, relying on forward-looking extrapolation and networked risk entropy assessment, it achieves a leap from passive monitoring to proactive early warning, significantly improving the early identification and control capabilities of dam structural safety risks, and providing intelligent decision support for construction safety and engineering protection. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a block diagram of the system structure of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0016] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0017] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application. Example

[0018] This embodiment provides, for example Figure 1-2 The system illustrates a smart monitoring system for the disturbance of a levee caused by a cross-river bridge, specifically comprising: a data fusion and governance module, a digital twin construction module, a causal reasoning and analysis module, and an early warning and decision-making module, wherein; Data fusion and governance module: It is used to synchronously collect dam response data, construction activity data and environmental data measured by sensors deployed at key sections of the dam and bridge construction points through the Internet of Things sensor network to form a multi-source heterogeneous data stream. It also uses a data fusion algorithm based on spatiotemporal dual registration to clean and align the multi-source heterogeneous data stream in real time, and generate a spatiotemporally synchronized multi-dimensional data cube. Digital twin building module: It is used to receive a multidimensional data cube and embed a simplified physical model with parameters that can be updated in real time. It uses data assimilation technology to compare and calibrate the monitoring data in the multidimensional data cube with the prediction results of the simplified physical model, and dynamically updates the parameters of the simplified physical model to build a dynamic digital twin that is consistent with the state of the physical dam. Causal Reasoning Analysis Module: Based on the structural causal model framework, this module analyzes the changes in the conditional probability distribution of dam state variables represented by dynamic digital twins before and after construction intervention, identifies the causal links between construction activities and dam response, and establishes a causal graph model that quantifies the effects of causal links. Early warning decision module: It is used to couple the causal graph model with the dynamic digital twin. For the construction activities corresponding to the causal links identified by the causal reasoning analysis module, it performs multi-condition forward-looking simulation in the dynamic digital twin, calculates the overall risk entropy based on complex network theory, and generates forward-looking early warning information based on the evolution trend of the overall risk entropy.

[0019] In this embodiment, it is specifically necessary to explain that in the data fusion governance module, the dam response data includes vibration acceleration collected by vibration sensors, micro-deformation collected by tilt sensors or strain gauges, and pore water pressure collected by pore water pressure gauges (the acquisition frequency range of vibration acceleration, micro-deformation, and pore water pressure is 10-100Hz; preferably, for vibration acceleration acquisition, the frequency is not less than 50Hz to ensure the capture of the main frequency domain components of construction impact; for relatively slow-changing micro-deformation and pore water pressure, the acquisition frequency can be set to 10-20Hz to achieve a balance between efficiency and accuracy); construction activity data includes pile driving energy, excavation depth, and machinery location information obtained in real time through the construction management system; and environmental data includes river water level and flow velocity information provided by hydrological stations. The above-mentioned original readings of vibration acceleration, micro-deformation, pore water pressure, pile driving energy, excavation depth, mechanical position, river water level and flow velocity together constitute the original observations in the multi-source heterogeneous data stream, and the multi-source heterogeneous data stream is accessed in real time through the Apache Kafka stream processing platform. The specific steps for real-time cleaning and alignment of multi-source heterogeneous data streams using a spatiotemporal dual registration-based data fusion algorithm are as follows: First, for each raw reading in the raw observations, the calibration function of the corresponding sensor is applied to convert the raw reading into an engineering value with a clear physical unit and dimension, thus completing the initial standardization of the raw observations. The calibration function is a conversion rule calibrated in advance for each sensor, used to convert voltage or current signals into physical quantities. The specific calibration process includes: applying a series of known standard physical quantity inputs to the sensor under standard laboratory or field conditions, recording the corresponding output electrical signals, and then using curve fitting methods such as the least squares method to establish a linear or nonlinear mapping relationship between the input physical quantities and the output electrical signals, which yields the calibration function of the sensor. Subsequently, a dual spatiotemporal registration is performed. Time registration involves attaching a microsecond-level precision time identifier (preferably IEEE 1588 Precision Time Protocol (PTP)) to each engineering value after initial standardization, so that all data are unified on the same time reference axis. Spatial registration assigns each sensor node a unique three-dimensional position identifier in the global coordinate system of the embankment. The global coordinate system of the embankment is an engineering coordinate system established by total station or RTK GPS measurement technology with the embankment axis as the reference. The measurement error of the three-dimensional position identifier of the sensor node should be controlled within ±5 cm. The three-dimensional position identifier is defined by three rectangular coordinate components, thereby binding each engineering value to its specific acquisition source location in physical space. Ultimately, a set of discrete data points with spatiotemporal labels is formed, defined by time markers, three-dimensional location markers, and corresponding engineering values. This set of discrete data points will serve as the basis for subsequent reconstruction of the continuous physical field, and the subsequent processing of the spatiotemporal dual-registration data fusion algorithm will use this set of discrete data points as input. The process of generating a spatiotemporally synchronized multidimensional data cube is as follows: After obtaining a set of discrete data points with spatiotemporal labels, a data fusion method based on physical field reconstruction theory is adopted. This method operates on a continuous spatial distribution field representing the physical state of the dam to be reconstructed. The continuous spatial distribution field is used to represent the physical state of the dam, including the displacement field reconstructed from micro-deformation data, the acceleration field reconstructed from vibration acceleration data, and the pressure field reconstructed from pore water pressure data. The method includes constructing an optimization objective function and solving the function to obtain the optimal continuous spatial physical field. The optimization objective function aims to find a physical field distribution that both fits all sensor observations as closely as possible and satisfies physical smoothness constraints (such as minimizing non-physical drastic fluctuations). The objective function is defined as the sum of a data fitting term and a physical regularization term, whereby the data fitting term is calculated as follows: For each data point in the discrete data point set, calculate the square of the difference between the estimated value of the continuous spatial distribution field to be reconstructed at the three-dimensional location marker of the data point and the engineering value recorded at the data point. Then multiply this squared difference by a weighting coefficient pre-assigned to the corresponding sensor of the data point. Finally, sum the above calculation results for all data points. This item reflects the degree of fit between the continuous spatial distribution field to be reconstructed and the measured data. The smaller the value, the better the fit. The physical regularization term is calculated as follows: For the continuous spatial distribution field to be reconstructed, the integral of the square of the Laplace operator of the field is obtained over the entire defined dam spatial domain. The Laplace operator is used to measure the spatial smoothness of the field. This term serves as a physical constraint, penalizing drastic spatial changes in the continuous spatial distribution to be reconstructed, making its solution more consistent with actual physical laws (such as continuous deformation of elastic bodies), and avoiding overfitting noise. The sum of the data fitting term and the physical regularization term is obtained by scaling the physical regularization term with a regularization parameter and then adding it to the unscaled data fitting term. The regularization parameter is used to balance the weight between data fitting accuracy and field smoothness. Its value is determined by cross-validation or the L-curve method according to the specific scenario, and the typical value range is between 0.01 and 10.0, for example, 1.0. By adjusting the regularization parameter and solving for the physical field spatial distribution that minimizes this weighted sum, the optimal continuous spatial physical field is obtained. The solution process can use numerical optimization algorithms such as the conjugate gradient method or the quasi-Newton method for iterative calculation until the preset convergence tolerance is met (for example, the relative change of the objective function value is less than one ten-thousandth). For each processing time point, the corresponding continuous spatial physical field is reconstructed through the above process. Then, the continuous spatial physical fields reconstructed at different time points are discretized and their values ​​are taken at each node of a predefined regular three-dimensional spatial grid. The field values ​​from different physical states (e.g., displacement field value, acceleration field value, pressure field value) at the same time point and the same spatial grid node are combined into a multidimensional vector. Then, the multidimensional vectors on all grid nodes are arranged and organized in chronological order to generate a spatiotemporally synchronized multidimensional data cube. This data cube can be logically regarded as a four-dimensional array, with its four dimensions corresponding to the spatial X-axis index, spatial Y-axis index, spatial Z-axis index, and time axis index, respectively. The value stored in the array at each index position is the multi-physical quantity vector of the corresponding spatial point and the corresponding time point. This structure ensures strict alignment and synchronization of data in spatiotemporal and multi-physical dimensions, providing a standardized and structured high-quality data foundation for subsequent digital twin construction and causal analysis, effectively overcoming the problems of data silos and spatiotemporal inconsistencies in traditional monitoring.

[0020] In this embodiment, it is specifically necessary to explain the following operations in the digital twin construction module: embedding a simplified physical model with real-time updated parameters and receiving a multi-dimensional data cube. First, the simplified physical model is defined as a finite element mechanical model of a dam based on the linear elastic assumption. The dynamic behavior of this model is described by the mass matrix, damping matrix, and stiffness matrix, and is driven by the external load vector. The dynamic response of the dam finite element mechanical model is manifested as displacement vector, velocity vector, and acceleration vector. The basic relationship satisfies that the sum of the product of the mass matrix and acceleration vector, the product of the damping matrix and velocity vector, and the product of the stiffness matrix and displacement vector equals the external load vector. The mass matrix, damping matrix, and stiffness matrix are constructed based on the geometric dimensions of the dam, material density, and the initial set of elastic modulus parameters. The external load vector is equivalently calculated from construction activity data (such as pile driving impact force) and environmental data (such as hydrostatic pressure) extracted from the multidimensional data cube. To balance computational efficiency and accuracy, the simplified physical model can preferentially use one-dimensional or two-dimensional beam elements or plate elements for discretization. The number of elements can be controlled between 1000 and 5000, rather than performing fine three-dimensional solid element division. Next, a state vector is defined, which is a column vector composed of two parallel parts: the first part is the transpose of the key model parameter set to be dynamically updated in the simplified physical model, and the second part is the transpose of some key state variables in the finite element mechanical model of the dam. The key model parameter set includes the equivalent elastic modulus parameters reflecting the stiffness of the dam soil, and the key state variables include the stress vector and displacement vector of the key nodes of the dam. Specifically, the key model parameter set is the elastic modulus value of each finite element element, or a set of equivalent elastic modulus parameters representing different soil layer regions. The key state variables do not necessarily include the displacement of all nodes, but rather the stress and displacement components of the key nodes (such as nodes near the top, toe, and piers of the dam) that are most sensitive to the load response, selected based on sensitivity analysis. This can significantly reduce the dimensionality of the state vector and improve the computational efficiency of subsequent data assimilation. The dimensionality of the state vector can be set between tens and hundreds of dimensions depending on the complexity of the model. The multidimensional data cube provides initial boundary conditions, load conditions, and observation data sources for comparison to simplify the physical model; the process of dynamically updating the parameters of the simplified physical model is the process of dynamically updating the key model parameter set in the state vector. The specific process of constructing a dynamic digital twin by comparing and calibrating the monitoring data in a multidimensional data cube with the prediction results of a simplified physical model using data assimilation technology is as follows: The data assimilation technique is implemented using the ensemble Kalman filter algorithm. Its processing flow includes iterative prediction and correction steps. In the ensemble Kalman filter algorithm, the number of ensemble members used to characterize uncertainty is usually set between 50 and 200, for example, 100. The statistical characteristics of the model process noise vector and the observation noise covariance matrix can be preset according to the sensor accuracy and historical data statistics. For example, the observation noise covariance matrix can be set as a diagonal matrix, and its diagonal elements are determined according to the measurement error variance of the corresponding sensor. In the prediction step, the optimal state vector estimate obtained from the correction step is used as the initial condition for the current prediction. The initial condition is input into the model evolution operator, which represents the dynamic evolution law of the simplified physical model, to calculate a preliminary predicted value of the state vector at the current moment. This preliminary predicted value is then superimposed with a model process noise vector, which characterizes the inaccuracy of the simplified physical model itself, to finally obtain the predicted value of the state vector at the current moment. Based on this predicted value of the state vector, the predicted value of the observed physical quantity at the current moment is calculated through an observation operator. The predicted value of the observed physical quantity refers to the displacement and acceleration values ​​at the sensor deployment location predicted by the simplified physical model. The implementation of the model evolution operator in the computer is the process of solving the dynamic equations corresponding to the simplified physical model by numerical integration (such as using the Newmark-β method or the central difference method) for one time step. The observation operator is essentially an index matrix that selects the physical quantities at the corresponding sensor locations from the state variables of the entire field (such as the displacements of all nodes). In the calibration step, the current real monitoring data is extracted from the multidimensional data cube. The real monitoring data refers to the actual measured values ​​of the dam response data corresponding to the predicted values ​​of the observed physical quantities in the multidimensional data cube. The calibration step further includes the following sub-steps: A1. Generate a set containing multiple members around the predicted state vector value to characterize the uncertainty of the predicted state. The method of generating the set is usually to apply a random perturbation that satisfies a multivariate Gaussian distribution around the predicted state vector value. A2. Calculate the Kalman gain matrix, which is determined in the following way: The covariance matrix of the predicted state is calculated based on the set calculation, and combined with the observation operator and its transpose, as well as an observation noise covariance matrix that characterizes the uncertainty of the observation data. This Kalman gain matrix is ​​used to optimally balance the relative reliability between the model prediction and the observation data. The calculation of the Kalman gain matrix involves matrix inversion. When the dimension of the state vector is too high, dimensionality reduction techniques such as truncated singular value decomposition can be used to ensure numerical stability. A3. Perform state updates, which involves using the Kalman gain matrix to correct the predicted state vector of each member in the set. The correction amount is proportional to the difference between the actual monitoring data and the predicted observation value corresponding to each member, thereby obtaining the corrected state vector of each member. The mean of the corrected state vectors of all members is the optimal state vector estimate at the current time after correction. This estimate includes the calibrated model parameters and system state. The correction process is essentially "pulling" the model prediction towards the actual observation value, so that the updated state vector and its contained model parameters are more consistent with the current measured response. The convergence condition of the iterative calculation is set to the change of the Euclidean norm of the state vector between two consecutive iterations being less than a preset threshold, such as one ten-thousandth. This process is repeated at each time step, continuously updating the parameters of the simplified physical model. The continuously updated set of key model parameters, together with the key state variables in the current corrected optimal state vector estimate, constitute a dynamic digital twin consistent with the state of the physical dam. Logically, this dynamic digital twin is a combination of the updated physical model and the corrected key state variables. The core value of the final dynamic digital twin lies in the fact that it is no longer an offline, parameter-fixed simulation model, but an online model whose key parameters (such as the equivalent elastic modulus of the soil) and real-time states (such as the displacement field) are synchronized with the physical dam. This twin can more accurately reflect the true mechanical state of the dam under construction disturbances, providing a high-quality state sequence constrained by physical mechanisms and corrected by data for subsequent causal analysis, effectively avoiding misjudgments of causal relationships directly from raw data that may contain noise. At the same time, as a reliable inference platform, this twin can be used to assess the potential impact of subsequent construction activities or extreme hydrological conditions on dam safety, achieving a leap from passive monitoring to proactive early warning.

[0021] In this embodiment, it is specifically necessary to explain the following operation in the causal reasoning analysis module: analyzing the changes in the conditional probability distribution of the dam state variables represented by the dynamic digital twin before and after the construction intervention, in order to identify the causal link: First, we define two types of core variables. The first type is intervention variables, which are vectors composed of key construction parameters extracted from construction activity data. Key construction parameters include piling energy and excavation depth. The second category is outcome variables, which are key state variables (i.e., stress vectors and displacement vectors of key nodes of the dam) extracted from the dam state output by the dynamic digital twin after data assimilation correction. Subsequently, based on historical data, a dataset is constructed, which consists of multiple data points. Each data point contains the value of the intervention variable at a time point and the value of the outcome variable at the same time point. Subsequently, a causal discovery algorithm based on conditional independence testing is used to process the dataset. The goal of this algorithm is to construct a directed acyclic graph (DAG) representing the topological structure of causal relationships between variables. Specifically, the PC algorithm or its variants can be used. This algorithm first assumes that all variables are connected, forming a completely undirected graph. Then, it evaluates the independence of variables given a subset of other variables using statistical tests (such as partial correlation coefficient tests or the G-test based on the Gaussian hypothesis). The significance level threshold for rejecting the conditional independence hypothesis is set to 0.05 or 0.01. The algorithm progressively increases the size of the condition set, removing edges between variables confirmed as conditionally independent, forming the skeleton of the graph. Finally, it orients the edges in the skeleton using causal direction principles such as the V-structure rule, outputting the final directed acyclic graph. This process automatically learns the causal structure from the observed data, avoiding the bias that may arise from relying on prior knowledge, and is crucial for discovering potential causal mechanisms. This directed acyclic graph (DAG) consists of a set of nodes and a set of edges. The node set includes intervention variables, outcome variables, and environmental confounding variables that simultaneously affect both intervention and outcome variables. Environmental confounding variables include river water level and flow velocity. The edge set represents the direction of direct causal influence between node variables. The causal discovery algorithm uses statistical tests to determine the conditional independence relationship between variables, gradually eliminating unnecessary correlation edges between variables, initially determining the skeleton structure of the causal graph, and then using causal directionality rules to determine the direction of the edges in the graph. Finally, it outputs the causal link topology structure, i.e., the directed acyclic graph. The directed acyclic graph intuitively shows possible causal paths such as "piling energy → levee displacement" and "river water level → pore water pressure → levee stress", providing a structural foundation for subsequent accurate quantification of influence. Introducing environmental confounding variables (such as river water level) and controlling their influence is an important step in ensuring the reliability of the discovered causal links. It can effectively prevent the misjudgment of correlations caused only by common causes (such as rising water level simultaneously increasing the difficulty of piling and increasing the stress on the levee) as causal relationships. The specific process of establishing a causal graph model for quantifying causal link effects is as follows: Based on the identification of the causal link topology, the effect strength of each causal link is further quantified. The quantification process is based on intervention causal reasoning theory. For each causal link in the directed acyclic graph that points from an intervention variable to an outcome variable, the average causal effect value is calculated. The calculation process for the average causal effect value is as follows: First, a sufficient adjustment set is determined based on the directed acyclic graph. This adjustment set is a set of variables that can block the influence of all confounding variables, so that the causal effect of the intervention variable on the outcome variable can be estimated unbiasedly. The sufficient adjustment set needs to be determined based on causal graph theory such as the backdoor criterion to ensure that the environment of the randomized controlled trial is statistically simulated. For example, to estimate the effect of "piling energy" on "dike toe displacement", it may be necessary to include "river water level" and "excavation depth" in the adjustment set to control confounding. Then, the average causal effect is calculated as follows: The adjustment set is iterated through all possible combinations of variable values. For each combination, the conditional expectation of the outcome variable is calculated when the intervention variable takes the first preset value and when the intervention variable takes the second preset value. The difference between these two conditional expectations is calculated. The conditional expectation can be estimated by establishing a linear or generalized linear model (e.g., considering interaction terms). The first and second preset values ​​are usually taken as the high quantile (e.g., 90th quantile) and low quantile (e.g., 10th quantile) of the actual distribution of the intervention variable to evaluate a comparison that is meaningful for engineering purposes. For example, the average change in the displacement of the embankment toe can be calculated when the piling energy increases from 1000 kJ (low level) to 3000 kJ (high level). Finally, all these differences are weighted and averaged according to the probability of the combination of values ​​of the adjustment set variable. The result is the average causal effect value of the causal link. This calculation method is mathematically equivalent to first calculating the expected value of the outcome variable under the condition that the intervention variable is set to the first preset value, then calculating the expected value of the outcome variable under the condition that the intervention variable is set to the second preset value, and finally subtracting the two expected values. "Human intervention setting" is a theoretical operation used to eliminate the interference of confounding factors, thereby measuring the pure causal effect. Finally, the directed acyclic graph representing the topological structure of causal links is combined with the calculated set of average causal effect values ​​that store the strength of each causal link effect to construct a causal graph model that quantifies causal link effects. Logically, the causal graph model consists of two parts: a directed acyclic graph and an average causal effect value parameter set. The directed acyclic graph represents the causal topological relationship between variables, and the average causal effect value parameter set stores the average causal effect value corresponding to each directed edge in the directed acyclic graph.

[0022] In this embodiment, it is specifically necessary to explain the following operation in the early warning decision module: performing multi-condition forward-looking simulation of the construction activities corresponding to the causal links identified by the causal reasoning analysis module. First, extract all causal links from the causal graph model and filter out those whose average causal effect value exceeds a preset threshold (this threshold can be set to a specific value between 0.5 and 0.8, such as 0.7, based on engineering experience and sensitivity analysis, to distinguish whether the impact is significant or not). Identify the construction activities corresponding to these links as key construction activities. Key construction activities include construction parameters such as piling energy (piling energy can be further subdivided according to its strength range, such as low strength <500kJ, medium strength 500-800kJ, and high strength ≥800kJ, to support more refined scenario definitions). Subsequently, based on the current construction status and the expected construction plan, a set of possible future scenarios are defined. These scenarios describe different future possibilities, such as maintaining the current intensity of key construction activity parameters, increasing the intensity of key construction activity parameters by a preset percentage (this percentage can be set to specific gradients such as 10%, 20%, 30%, etc.), and environmental parameters rising to the warning water level (the warning water level can be set to the historical highest water level plus a safety margin, such as 0.5 meters). Each scenario includes the trajectory of changes in one or more key construction activity parameters and environmental parameters over a future period of time (this duration can be set according to the construction stage and risk assessment needs, such as 24 hours, 48 ​​hours, or 72 hours). Key construction activity parameters include piling energy, and environmental parameters include river water level. The scenarios include, but are not limited to, the following types: maintaining the current intensity of key construction activity parameters, increasing the intensity of key construction activity parameters by a preset percentage, and maintaining the current intensity of key construction activity parameters but with environmental parameters rising to a preset water level. Subsequently, the current dynamic digital twin is used as the initial state for the simulation. The change trajectories of key construction activity parameters and environmental parameters defined in each simulation scenario are used as external inputs to drive the calibrated simplified physical model in the dynamic digital twin to perform forward numerical simulation. The forward numerical simulation uses a numerical integration algorithm to start from the current displacement and stress key state variables of the dam after data assimilation and correction, and performs numerical integration calculations for a future time period (preferably the fourth-order Runge-Kutta method, whose cumulative truncation error is a fifth-order infinitesimal of the step size, which significantly improves accuracy compared to first-order algorithms such as the Euler method). The dynamic response sequence of key state variables of dam displacement and stress over time is measured. The numerical integration algorithm uses the Runge-Kutta method for calculation, and predicts the evolution trajectory from the current state to the future state through multi-step iteration. This process is executed independently once for each defined simulation scenario, thereby obtaining a set of dam future state response sequences corresponding to different future possibilities, i.e., multi-condition prospective simulation results. By coupling the real-time updated digital twin with the multi-condition simulation, the response of the dam under construction disturbance can be dynamically evaluated. The simulation results can better reflect the actual working conditions and provide more reliable input for risk warning. The specific process of calculating the overall risk entropy based on complex network theory and generating forward-looking early warning information based on the evolution trend of the overall risk entropy is as follows: First, the finite element mechanical model of the dam on which the dynamic digital twin is based is abstracted into a complex network, where each finite element element is regarded as a node in the network. Elements that share nodes or boundaries are connected to each other, thereby constructing the dam structure network. The total number of nodes in the network is the total number of network nodes (the total number of nodes is consistent with the size of the finite element model, which may range from hundreds to tens of thousands). Subsequently, for the future state response sequence of the dam under each working condition obtained from the forward-looking simulation (i.e., the evolution data of key state variables such as displacement and stress), the failure probability of each node in the dam structure network at each future time is calculated. The node failure probability is determined by comparing the stress state at the node with material strength criteria (such as the Mohr-Coulomb criterion, the Drucker-Prag criterion, etc.) and taking into account uncertainties (such as the spatial variability of material parameters, load uncertainties, etc.). (The probability can be calculated using reliability theory, such as the first second moment method or Monte Carlo simulation). Then, based on complex network theory, the overall risk entropy is calculated. This overall risk entropy is a scalar value used to quantify the degree of disorder in the dam system's risk. The calculation process includes assigning a weight coefficient to each node in the dam structure network, representing its topological importance. The weight coefficient is determined by the node's topological importance based on its connection structure characteristics within the network (for example, eigenvector centrality, betweenness centrality, etc., can be used to measure node importance; nodes with higher importance are more critical in the force transmission path, and their failure has a greater impact on the system). The specific calculation method for the overall risk entropy is as follows: For a given future moment under the scenario of interest, the calculation of the overall risk entropy follows these steps: The first step is to traverse every node in the dam structure network. For each node, calculate the product of the node's failure probability at the current moment and the node's weight coefficient to obtain the weighted failure probability of the node. The second step is to calculate the natural logarithm of the weighted failure probability of the node obtained in the first step. The third step is to multiply the weighted failure probability of the node obtained in the first step by the natural logarithm of the weighted failure probability calculated in the second step to obtain the contribution of the node to the overall risk entropy at the current moment. The fourth step is to sum the contribution values ​​of all nodes in the dam structure network at the current moment; Fifth, take the negative value of the summation result obtained in the fourth step. The final value is the overall risk entropy of the projected working condition at that future moment. The overall risk entropy characterizes the degree of overall risk disorder of the dam system at that working condition at that moment. The higher the overall risk entropy value, the greater the possibility that the system is in a high-risk chaotic or disordered state, and the higher the risk of system failure. Finally, the trend of the overall risk entropy over time under each simulated working condition is analyzed. The trend analysis includes calculating the rate of change of the overall risk entropy over time (the instantaneous rate of change can be calculated by numerical difference method, or by fitting the slope of the linear trend line). If the overall risk entropy exceeds a preset threshold (this threshold can be determined based on the dike engineering level, historical data analysis, and expert experience; for example, it can be set to 0.85 for a first-level dike and 0.75 for a second-level dike) or its rate of change exceeds a preset growth rate threshold (for example, it can be set to 0.15 per hour), then an early warning level is determined. The early warning level is determined by comprehensively considering the magnitude of the overall risk entropy, the magnitude of the rate of change, and the importance of the corresponding key construction activity parameters, and a blue, yellow, orange, or red warning level is determined (for example: blue - low risk, yellow - medium risk, orange - relatively high risk, red - high risk; the judgment rule can be designed as follows: when the risk entropy exceeds threshold 1 and the rate of change exceeds the threshold, a red warning is triggered; when it only exceeds threshold 1 or the rate of change threshold, an orange warning is triggered; and so on). Based on the determined warning level, specific forward-looking warning information is generated. The warning information includes a description of the triggered working condition, the current value and predicted peak value of the risk entropy, and suggested intervention measures (for example, the warning information may indicate that "the current piling energy is running at a high level, and it is predicted that the risk entropy of the embankment toe displacement will exceed the threshold within the next 24 hours. It is recommended to immediately reduce the piling energy to below 500kJ and strengthen the monitoring of the displacement of this section to once per minute"). Finally, structured and actionable forward-looking warning information is output (this structured information can be easily integrated into the construction management platform or sent to the mobile terminals of relevant personnel).

[0023] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0024] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0025] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0026] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0027] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0028] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0029] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A smart monitoring system for the disturbance of a levee to a cross-river bridge, characterized in that, Specifically, it includes: The system comprises a data fusion and governance module, a digital twin construction module, a causal reasoning and analysis module, and an early warning and decision-making module; Data fusion and governance module: It is used to synchronously collect dam response data, construction activity data and environmental data measured by sensors deployed at key sections of the dam and bridge construction points through the Internet of Things sensor network to form a multi-source heterogeneous data stream. It also uses a data fusion algorithm based on spatiotemporal dual registration to clean and align the multi-source heterogeneous data stream in real time, and generate a spatiotemporally synchronized multi-dimensional data cube. Digital twin building module: It is used to receive a multidimensional data cube and embed a simplified physical model with parameters that can be updated in real time. It uses data assimilation technology to compare and calibrate the monitoring data in the multidimensional data cube with the prediction results of the simplified physical model, and dynamically updates the parameters of the simplified physical model to build a dynamic digital twin that is consistent with the state of the physical dam. Causal Reasoning Analysis Module: Based on the structural causal model framework, this module analyzes the changes in the conditional probability distribution of dam state variables represented by dynamic digital twins before and after construction intervention, identifies the causal links between construction activities and dam response, and establishes a causal graph model that quantifies the effects of causal links. Early warning decision module: It is used to couple the causal graph model with the dynamic digital twin. For the construction activities corresponding to the causal links identified by the causal reasoning analysis module, it performs multi-condition forward-looking simulation in the dynamic digital twin, calculates the overall risk entropy based on complex network theory, and generates forward-looking early warning information based on the evolution trend of the overall risk entropy.

2. The intelligent monitoring system for the disturbance of a levee to a cross-river bridge according to claim 1, characterized in that: In the data fusion and governance module, the dam response data includes vibration acceleration collected by vibration sensors, micro-deformation collected by tilt sensors, and pore water pressure collected by pore water pressure gauges; the construction activity data includes pile driving energy, excavation depth, and machinery location information obtained in real time through the construction management system; and the environmental data includes river water level and flow velocity information provided by hydrological stations. The raw readings of vibration acceleration, micro-deformation, pore water pressure, pile driving energy, excavation depth, machine position, river water level and flow velocity constitute the raw observations in the multi-source heterogeneous data stream.

3. The intelligent monitoring system for the disturbance of a levee to a cross-river bridge according to claim 2, characterized in that: The specific steps for real-time cleaning and alignment of multi-source heterogeneous data streams using a spatiotemporal dual registration-based data fusion algorithm are as follows: First, for each raw reading in the raw observations, the calibration function of the corresponding sensor is applied to convert the raw reading into an engineering value, thus completing the initial standardization of the raw observations. Subsequently, a dual registration of time and space is performed. Time registration involves attaching a time identifier based on a high-precision clock synchronization protocol to each engineering value after initial standardization, so that all data are unified on the same time reference axis. Spatial registration is the process of assigning each sensor node a unique three-dimensional location identifier within the global coordinate system of the embankment. Ultimately, a set of discrete data points with spatiotemporal labels is formed, defined by time markers, three-dimensional location markers, and corresponding engineering values.

4. The intelligent monitoring system for the disturbance of a levee to a cross-river bridge according to claim 3, characterized in that: The process of generating a spatiotemporally synchronized multidimensional data cube is as follows: After obtaining a set of discrete data points with spatiotemporal labels, a data fusion method based on physical field reconstruction theory is adopted. This method operates on a continuous spatial distribution field representing the physical state of a dam to be reconstructed. The method includes constructing an optimization objective function and solving the function to obtain the optimal continuous spatial physical field. The objective function is defined as the sum of a data fitting term and a physical regularization term, whereby the data fitting term is calculated as follows: For each data point in the discrete data point set, calculate the square of the difference between the estimated value of the continuous spatial distribution field to be reconstructed at the three-dimensional location marker of the data point and the engineering value recorded at the data point. Then multiply this squared difference by a weighting coefficient pre-assigned to the corresponding sensor of the data point. Finally, sum the above calculation results for all data points. The physical regularization term is calculated as follows: For the continuous spatial distribution field to be reconstructed, find the integral of the square of the Laplace operator of the field over the entire defined dam space domain; The sum of the data fitting term and the physical regularization term is obtained by scaling the physical regularization term with a regularization parameter and then adding it to the unscaled data fitting term. By adjusting the regularization parameter and solving for the physical field spatial distribution that minimizes this weighted sum, the optimal continuous spatial physical field is obtained. For each processing time point, the corresponding continuous spatial physical field is reconstructed through the above process. Then, the continuous spatial physical fields obtained from different time points are discretized and their values ​​are taken at each node of a predefined regular three-dimensional spatial grid. The field values ​​from different physical states at the same time point and the same spatial grid node are combined into a multidimensional vector. Then, the multidimensional vectors on all grid nodes are arranged and organized in chronological order to generate a spatiotemporally synchronized multidimensional data cube.

5. The intelligent monitoring system for the disturbance of a levee to a cross-river bridge according to claim 4, characterized in that: In the digital twin construction module, the specific operations of embedding a simplified physical model with real-time updated parameters and receiving a multi-dimensional data cube are as follows: First, a simplified physical model is defined as a finite element mechanical model of a dam based on the linear elastic assumption. The dynamic behavior of this model is described by the mass matrix, damping matrix, and stiffness matrix, and is driven by the external load vector. The dynamic response of the dam finite element mechanical model is manifested as displacement vector, velocity vector, and acceleration vector. The basic relationship between them is: the sum of the product of the mass matrix and the acceleration vector, the product of the damping matrix and the velocity vector, and the product of the stiffness matrix and the displacement vector is equal to the external load vector. Next, a state vector is defined, which is a column vector composed of two parallel parts: the first part is the transpose of the key model parameter set to be dynamically updated in the simplified physical model, and the second part is the transpose of some key state variables in the finite element mechanical model of the dam; the key model parameter set includes the equivalent elastic modulus parameter reflecting the stiffness of the dam soil, and the key state variables include the stress vector and displacement vector of the key nodes of the dam. The multidimensional data cube provides initial boundary conditions, load conditions, and observation data sources for comparison to simplify the physical model; the process of dynamically updating the parameters of the simplified physical model is the process of dynamically updating the key model parameter set in the state vector.

6. The intelligent monitoring system for the disturbance of a levee to a cross-river bridge according to claim 5, characterized in that: The specific process of constructing a dynamic digital twin by comparing and calibrating the monitoring data in the multidimensional data cube with the prediction results of the simplified physical model using data assimilation technology is as follows: The data assimilation technique is implemented using an ensemble Kalman filter algorithm, whose processing flow includes iterative prediction and correction steps. In the prediction step, the estimated value of the optimal state vector obtained through the correction step is used as the initial condition for the current prediction. The initial condition is input into the model evolution operator that represents the dynamic evolution law of the simplified physical model, and a preliminary predicted value of the state vector at the current moment is calculated. This initial prediction is then superimposed with a model process noise vector used to characterize the inaccuracy of the simplified physical model itself, and finally the predicted state vector value at the current moment is obtained. Based on this state vector prediction, the predicted value of the observed physical quantity at the current moment is calculated through an observation operator. The predicted value of the observed physical quantity refers to the displacement and acceleration values ​​at the sensor deployment location predicted by the simplified physical model. In the calibration step, the current real monitoring data is extracted from the multidimensional data cube. The real monitoring data refers to the actual measured values ​​of the dam response data corresponding to the predicted values ​​of the observed physical quantities in the multidimensional data cube. The calibration step further includes the following sub-steps: A1. Generate a set containing multiple members based on the predicted values ​​of the state vector; A2. Calculate the Kalman gain matrix, which is determined in the following way: The covariance matrix of the predicted state is calculated based on set calculation, and is combined with the observation operator and its transpose, as well as an observation noise covariance matrix that characterizes the uncertainty of the observation data. A3. Perform state update, that is, use the Kalman gain matrix to correct the predicted state vector of each member in the set. The correction amount is proportional to the difference between the actual monitoring data and the predicted observation value corresponding to each member, so as to obtain the corrected state vector of each member. The mean of the corrected state vectors of all members is the optimal state vector estimate after correction at the current time. This process is repeated at each time step, thereby continuously updating the parameters of the simplified physical model. The continuously updated set of key model parameters, together with the key state variables in the current corrected optimal state vector estimate, constitute a dynamic digital twin consistent with the state of the physical dam.

7. A smart monitoring system for the disturbance of a levee to a cross-river bridge according to claim 6, characterized in that: In the causal reasoning analysis module, the specific operation of analyzing the changes in the conditional probability distribution of the dam state variables represented by the dynamic digital twin before and after the construction intervention to identify the causal link is as follows: First, we define two types of core variables. The first type is intervention variables, which are vectors composed of key construction parameters extracted from construction activity data. Key construction parameters include piling energy and excavation depth. The second category is outcome variables, which are key state variables that characterize the dam response, extracted from the dam state output by the dynamic digital twin and corrected for data assimilation. Subsequently, based on historical data, a dataset is constructed, which consists of multiple data points. Each data point contains the value of the intervention variable at a time point and the value of the outcome variable at the same time point. Subsequently, a causal discovery algorithm based on conditional independence test was used to process the dataset. The goal of this algorithm is to construct a directed acyclic graph that represents the topological structure of causal relationships between variables. The directed acyclic graph (DAG) consists of a set of nodes and a set of edges. The set of nodes includes intervention variables, outcome variables, and environmental confounding variables that simultaneously affect both intervention and outcome variables. Environmental confounding variables include river water level and flow velocity. The set of edges represents the direction of direct causal influence between node variables. The causal discovery algorithm uses statistical tests to determine the conditional independence relationship between variables, gradually eliminates unnecessary correlation edges between variables, initially determines the skeleton structure of the causal graph, and then uses causal directionality rules to determine the direction of the edges in the graph, finally outputting the causal link topology structure, i.e., the directed acyclic graph.

8. The intelligent monitoring system for the disturbance of a levee to a cross-river bridge according to claim 7, characterized in that: The specific process for establishing a causal graph model to quantify causal link effects is as follows: Based on the identification of the causal link topology, the effect strength of each causal link is further quantified; the quantification process is based on intervention causal reasoning theory, and for each causal link in the directed acyclic graph that points from an intervention variable to an outcome variable, its average causal effect value is calculated. The calculation process for the average causal effect value is as follows: First, determine a sufficient adjustment set based on the directed acyclic graph; Then, the average causal effect is calculated as follows: Iterate through all possible combinations of variable values ​​in the adjustment set. For each combination of values, calculate the conditional expectation of the outcome variable when the intervention variable takes the first preset value and the conditional expectation of the outcome variable when the intervention variable takes the second preset value, and calculate the difference between the two conditional expectations. Finally, all these differences are weighted and averaged according to the probability of the combination of values ​​of the adjustment set variable. The result is the average causal effect value of the causal link. Finally, the directed acyclic graph representing the topology of causal links is combined with the calculated set of average causal effect values ​​that store the strength of each causal link effect to construct a causal graph model that quantifies causal link effects.

9. A smart monitoring system for the disturbance of a levee to a cross-river bridge according to claim 8, characterized in that: In the aforementioned early warning decision module, the specific operation of performing multi-condition forward-looking simulation of construction activities corresponding to the causal links identified by the causal reasoning analysis module is as follows: First, all causal links are extracted from the causal graph model, and causal links with an average causal effect value exceeding a preset threshold are selected. The construction activities corresponding to these links are then identified as key construction activities. Subsequently, based on the current construction status and the expected construction plan, a set of possible future scenarios are defined; each scenario includes the trajectory of changes in one or more key construction activity parameters and environmental parameters over a period of time in the future. The key construction activity parameters include pile driving energy, and the environmental parameters include river water level. Subsequently, the current dynamic digital twin is used as the initial state for the simulation. The change trajectories of key construction activity parameters and environmental parameters defined in each simulation scenario are used as external inputs to drive the calibrated simplified physical model in the dynamic digital twin to perform forward numerical simulation. The forward numerical simulation uses a numerical integration algorithm to start from the current displacement and stress key state variables of the dam after data assimilation and correction, and performs numerical integration calculations over a future period to predict the dynamic response sequence of the dam's displacement and stress key state variables over time. The numerical integration algorithm uses the Runge-Kutta method for calculation, and predicts the evolution trajectory from the current state to the future state through multi-step iteration. This process is executed independently for each defined scenario, resulting in a set of future state response sequences for the dam corresponding to different future possibilities, i.e., multi-condition forward-looking simulation results.

10. A smart monitoring system for the disturbance of a levee to a cross-river bridge according to claim 9, characterized in that: The specific process of calculating the overall risk entropy based on complex network theory and generating forward-looking early warning information according to the evolution trend of the overall risk entropy is as follows: First, the finite element mechanical model of the dam on which the dynamic digital twin is based is abstracted into a complex network, where each finite element element is regarded as a node in the network, and the elements that share nodes or boundaries are connected to each other, thereby constructing the dam structure network. Subsequently, for the future state response sequence of the dam under each working condition obtained by forward-looking simulation, the failure probability of each node in the dam structure network at each future time is calculated. Then, the overall risk entropy is calculated based on complex network theory. The calculation process includes assigning a weight coefficient representing the topological importance of each node in the dam structure network. The specific calculation method for the overall risk entropy is as follows: For a given future moment under the scenario of interest, the calculation of the overall risk entropy follows these steps: The first step is to traverse every node in the dam structure network. For each node, calculate the product of the node's failure probability at the current moment and the node's weight coefficient to obtain the weighted failure probability of the node. The second step is to calculate the natural logarithm of the weighted failure probability of the node obtained in the first step. The third step is to multiply the weighted failure probability of the node obtained in the first step by the natural logarithm of the weighted failure probability calculated in the second step to obtain the contribution of the node to the overall risk entropy at the current moment. The fourth step is to sum the contribution values ​​of all nodes in the dam structure network at the current moment; Fifth, take the negative value of the summation result obtained in the fourth step. The final value is the overall risk entropy of the simulated working condition at that future moment. Finally, the trend of the overall risk entropy over time under each simulated working condition is analyzed. The trend analysis includes calculating the rate of change of the overall risk entropy over time. If the value of the overall risk entropy exceeds a preset threshold or its rate of change exceeds a preset growth rate threshold, an early warning level will be determined. The early warning level will be determined by comprehensively considering the value of the overall risk entropy, the rate of change, and the importance of the corresponding key construction activity parameters, and will be classified as blue, yellow, orange, or red. Based on the determined warning level, specific forward-looking warning information is generated. The warning information includes a description of the triggered working condition, the current value and predicted peak value of the risk entropy, and suggested intervention measures. Finally, structured and actionable forward-looking warning information is output.

Images