A dam safety state evaluation method and system fusing multi-source monitoring data

Abstract

The present application relates to dam safety monitoring and evaluation technical field, specifically to a kind of dam safety state evaluation method and system fusing multi-source monitoring data, including acquisition multi-source heterogeneous real-time monitoring data;Obtain standardized data set;Generation spatiotemporal consistent fusion data set;Through multi-physical field coupling solution, inversion calculation dam internal physical field distribution;Extract internal physical field distribution corresponding model predicted displacement;In combination with the actual monitoring displacement corresponding in fusion data set, calculate displacement deviation;Generation corrected physical model;In response to displacement deviation is not greater than preset displacement deviation threshold, then initial physical model is determined as corrected physical model;Generation multidimensional risk feature vector;Determine current comprehensive health index;Generation predicted health index;Calculate future instability probability;Output early warning level and decision countermeasure;The present application overcomes the hysteresis of traditional early warning, improves the scientificity and foresight of dam safety management.

Description

Technical Field

[0001] This invention relates to the field of dam safety monitoring and assessment technology, specifically to a method and system for assessing the safety status of dams by integrating multi-source monitoring data. Background Technology

[0002] With the continuous advancement of large-scale water conservancy infrastructure construction, dam structures are becoming increasingly complex, posing severe challenges to their safety management. Currently, dam safety monitoring systems have widely deployed multi-source heterogeneous sensors to collect real-time monitoring data such as seepage, deformation, and temperature. However, these data often suffer from the following problems:

[0003] Data heterogeneity and inconsistency: Different monitoring data sources, varying types, and inconsistent sampling frequencies and spatial distributions make direct data fusion and comparison difficult, hindering the construction of a comprehensive and consistent view of dam operation status. Specifically, while there are numerous dam safety monitoring data sources, effectively integrating seepage pressure monitoring data (reflecting pore water pressure within the dam), displacement monitoring data (reflecting structural response), and temperature monitoring data (reflecting environmental impacts) spatiotemporally and uniformly inputting them into the physical model is the foundation and primary challenge for building an accurate assessment model. Difficulty in directly measuring internal conditions: The critical internal physical conditions of the dam cannot be directly obtained through sensors, making traditional monitoring data analysis methods inadequate for accurately assessing the true risks within the dam. The physical model... Static limitations: The initial physical model is built based on design data, but during the long-term operation of the dam, its material parameters will change due to aging and damage, causing the model parameters to be inconsistent with the actual physical characteristics of the dam. The model prediction results will deviate from the actual monitoring data, and the model fidelity will decrease. In particular, during the long-term service of the dam, its internal materials, such as elastic modulus and permeability coefficient, will deteriorate, and traditional physical models cannot dynamically reflect this change, resulting in serious distortion of the assessment results based on static models and multi-source monitoring data. Lack of foresight in assessment: Traditional safety assessments are mostly based on current or historical state thresholds, lacking the ability to predict the future trend and instability probability of the dam, making it difficult to achieve timely and forward-looking early warning.

[0004] Therefore, how to effectively integrate multi-source heterogeneous monitoring data, dynamically revise physical models to construct high-fidelity digital twins, and realize closed-loop management from state assessment to trend prediction based on this has become a key technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for assessing the safety status of dams by integrating multi-source monitoring data. This method and system aim to effectively integrate heterogeneous monitoring data from multiple sources, overcoming the static limitations of traditional physical models that do not match the actual physical characteristics of dams due to variations in material parameters. It constructs a high-fidelity digital twin by dynamically correcting the physical model and achieves forward-looking closed-loop management of dam safety status from current assessment to future trend prediction. Specifically, the technical solution of this invention is:

[0006] A method for assessing the safety status of dams by integrating multi-source monitoring data includes:

[0007] Collect real-time monitoring data from multiple heterogeneous sources;

[0008] Based on preset minimum and maximum reference values, the real-time monitoring data is subjected to minimum-maximum normalization to obtain a standardized dataset.

[0009] By combining standardized datasets, sensor 3D spatial coordinate information, and data acquisition timestamps, a spatiotemporal alignment process is performed using a preset interpolation algorithm to generate a spatiotemporally consistent fused dataset.

[0010] The fused dataset is used as real-time boundary conditions and internal loads, and input into the initial physical model constructed based on the preset static data of the dam. Through multi-physics coupling solution, the distribution of physical fields inside the dam body is inverted and calculated. Multi-physics coupling solution specifically refers to the control equations that couple the seepage field, stress field and temperature field. The theoretical core of this step is: the seepage field obtained after fusing seepage pressure monitoring data, that is, the distribution of pore water pressure, is introduced into the calculation of stress field based on the effective stress principle, so as to accurately invert the actual impact of pore water pressure changes on the actual stress and deformation of the dam body.

[0011] Extract the model-predicted displacement corresponding to the internal physical field distribution;

[0012] Calculate the displacement deviation by combining the actual monitored displacements in the fused dataset;

[0013] Determine whether the displacement deviation is greater than the preset displacement deviation threshold;

[0014] If the displacement deviation exceeds a preset displacement deviation threshold, the key parameters of the initial physical model are automatically corrected using a preset inversion algorithm to generate a corrected physical model.

[0015] If the displacement deviation is not greater than the preset displacement deviation threshold, the initial physical model is determined as the corrected physical model.

[0016] The internal physical field distribution output by the corrected physical model is extracted using a preset pattern recognition algorithm to generate a multidimensional risk feature vector. The multidimensional risk feature vector is converted into multiple normalized single-item risk indicators, including at least: a seepage risk indicator corresponding to the seepage field, a stress risk indicator corresponding to the stress field, and a deformation risk indicator corresponding to the deformation, in order to quantitatively assess the safety status of the dam under different physical dimensions.

[0017] Based on a multidimensional risk feature vector, the current comprehensive health index is determined by a preset mapping function and a linear weighted summation.

[0018] The comprehensive health index is combined into a historical index sequence, and a predicted health index is generated through a pre-trained time series prediction model.

[0019] Calculate the probability of future instability based on the predicted health index and confidence interval, the preset critical failure threshold and historical index sequence;

[0020] Based on the current comprehensive health index and the probability of future instability, combined with the preset graded early warning logic, the system outputs the early warning level and decision-making countermeasures.

[0021] Optionally, spatiotemporal alignment is performed using a preset interpolation algorithm, including:

[0022] Based on the three-dimensional spatial coordinate information and data acquisition timestamps of each sensor, the time dimension of the standardized dataset is synchronized using linear interpolation or Kriging interpolation algorithms.

[0023] The synchronized data is mapped onto preset unified dam body 3D spatial grid nodes to construct a spatiotemporally consistent fused dataset.

[0024] Optionally, solutions can be obtained through multiphysics coupling, including:

[0025] The governing equations that couple the seepage field, stress field and temperature field are solved. The stress field is calculated based on the effective stress principle, which is used to describe the relationship between stress, pore pressure and temperature stress in porous media.

[0026] Optionally, key parameters of the initial physical model can be automatically corrected using a preset inversion algorithm, including:

[0027] The least squares method is used to dynamically adjust the elastic modulus and permeability coefficient of the initial physical model to minimize the displacement deviation, thereby generating the corrected physical model.

[0028] Optionally, based on a multidimensional risk feature vector, the current comprehensive health index is determined, including:

[0029] The multidimensional risk feature vector is converted into multiple normalized single risk indicators through a preset mapping function. The single risk indicators include: seepage risk indicator, stress risk indicator and deformation risk indicator.

[0030] By using linear weighted summation, multiple individual risk indicators are weighted with their corresponding preset risk weight coefficients to obtain the cumulative risk value.

[0031] Subtract the accumulated risk value from 1 to obtain the comprehensive health index.

[0032] Optionally, a preset risk weighting coefficient can be used, and the determination methods include:

[0033] Based on a pre-set expert knowledge base, the initial values ​​of the risk weight coefficients are set using the analytic hierarchy process.

[0034] Using historical incident data, the initial values ​​are trained and dynamically optimized to calibrate the risk weight coefficients, and the risk weight coefficients satisfy the normalization constraint.

[0035] Optionally, based on preset tiered early warning logic, the system outputs early warning levels and corresponding decision-making strategies, including:

[0036] When the current comprehensive health index is less than or equal to the critical failure threshold, or the probability of future instability is greater than the second preset probability threshold, a dangerous state is output.

[0037] When the current comprehensive health index is greater than the critical failure threshold and the probability of future instability is not greater than the second preset probability threshold, and the current comprehensive health index is less than or equal to the second preset threshold, or the probability of future instability is greater than the first preset probability threshold, an early warning status is output.

[0038] When the current comprehensive health index is greater than the second preset threshold, and the current comprehensive health index is less than or equal to the first preset threshold, and the probability of future instability is not greater than the first preset probability threshold, output the attention status;

[0039] When the current comprehensive health index is greater than the first preset threshold and the probability of future instability is not greater than the first preset probability threshold, the normal state is output.

[0040] A dam safety status assessment system integrating multi-source monitoring data includes:

[0041] The data fusion module is used to collect real-time monitoring data from multiple heterogeneous sources and perform min-max normalization and spatiotemporal alignment processing to generate a spatiotemporally consistent fused dataset.

[0042] The physical inversion module is used to input the fused dataset into the initial physical model and solve it through multi-physics coupling to invert and calculate the physical field distribution inside the dam body;

[0043] The model correction module is used to calculate the displacement deviation between the model-predicted displacement and the actual monitored displacement, and to generate a corrected physical model based on the comparison result of the displacement deviation with the preset displacement deviation threshold.

[0044] The status assessment module is used to extract the internal physical field distribution output by the corrected physical model into a multi-dimensional risk feature vector and calculate the current comprehensive health index.

[0045] The trend prediction module is used to combine comprehensive health indices into a historical index sequence to generate a predicted health index and calculate the probability of future instability.

[0046] The early warning decision module is used to output the early warning level and decision-making countermeasures based on the current comprehensive health index and the probability of future instability, combined with the preset hierarchical early warning logic.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] 1. This invention achieves effective fusion of multi-source heterogeneous monitoring data. It eliminates the dimensional differences between different data through min-max normalization and solves the problems of inconsistent sampling frequencies and uneven spatial distribution through spatiotemporal alignment processing, generating a high-quality spatiotemporally consistent fusion dataset, which provides an accurate and reliable data foundation for subsequent physical inversion.

[0049] 2. This invention constructs a dynamically corrected, high-fidelity digital twin of a dam. By utilizing the deviation between monitoring data and model-predicted displacement, the key parameters of the initial physical model, such as the elastic modulus and permeability coefficient, are automatically corrected through an inversion algorithm. This ensures that the physical model can reflect the aging or damage changes of the dam materials in real time, greatly improving the accuracy of the internal physical field inversion calculation.

[0050] 3. This invention establishes a quantitative assessment system from complex physical fields to intuitive health indicators. By coupling multiple physical fields, it deduces the seepage field and stress field inside the dam that cannot be directly measured, and extracts them as a multi-dimensional risk feature vector. Finally, it calculates a single, quantitative comprehensive health index, which makes it easy for managers to quickly grasp the overall safety status of the dam. The quantitative assessment system closely links the physical and mechanical mechanisms of the dam with its safety characteristics. By combining quantitative indicators with qualitative logic, it ensures the feasibility and consistency of the assessment results.

[0051] 4. This invention realizes closed-loop management from state assessment to trend prediction. It not only judges the state based on the current comprehensive health index, but also generates a predicted health index through a time series prediction model and calculates the probability of future instability. Combined with the hierarchical early warning logic of dual criteria, it overcomes the lag of traditional early warning and improves the scientificity and foresight of dam safety management. Attached Figure Description

[0052] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0053] Figure 1 This is a flowchart of the method of the present invention;

[0054] Figure 2 This is a structural diagram of the system of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0056] Example 1:

[0057] The method of this invention is particularly applicable to automated safety monitoring and assessment scenarios for hydraulic structures such as large and medium-sized reservoir earth-rock dams. Please refer to [link / reference]. Figure 1 A method for assessing the safety status of dams by integrating multi-source monitoring data, comprising:

[0058] Collect real-time monitoring data from multiple heterogeneous sources;

[0059] Based on preset minimum and maximum reference values, the real-time monitoring data is subjected to minimum-maximum normalization to obtain a standardized dataset; the preset minimum and maximum reference values ​​are determined by the dam design safety limits or the statistical extreme values ​​of long-term monitoring data.

[0060] By combining standardized datasets, sensor 3D spatial coordinate information, and data acquisition timestamps, a spatiotemporal alignment process is performed using a pre-defined interpolation algorithm to generate a spatiotemporally consistent fused dataset. Spatiotemporal alignment is a key data governance step aimed at solving the heterogeneity and inconsistency problems of multi-source heterogeneous data, providing a high-quality data foundation for subsequent analysis.

[0061] The fused dataset is used as real-time boundary conditions and internal loads, and input into the initial physical model constructed based on the preset static data of the dam. The distribution of physical fields inside the dam body is calculated by multi-physics coupling solution.

[0062] Extract the model-predicted displacement corresponding to the internal physical field distribution;

[0063] Calculate the displacement deviation by combining the actual monitored displacements in the fused dataset;

[0064] Determine whether the displacement deviation is greater than the preset displacement deviation threshold;

[0065] If the displacement deviation exceeds a preset displacement deviation threshold, the key parameters of the initial physical model are automatically corrected using a preset inversion algorithm to generate a corrected physical model.

[0066] If the displacement deviation is not greater than the preset displacement deviation threshold, the initial physical model is determined as the corrected physical model.

[0067] The internal physical field distribution output by the corrected physical model is extracted using a preset pattern recognition algorithm to generate a multidimensional risk feature vector.

[0068] Based on a multidimensional risk feature vector, the current comprehensive health index is determined by a preset mapping function and a linear weighted summation.

[0069] The comprehensive health index is combined into a historical index sequence, and a predicted health index is generated through a pre-trained time series prediction model.

[0070] Calculate the probability of future instability based on the predicted health index and confidence interval, the preset critical failure threshold and historical index sequence;

[0071] Based on the current comprehensive health index and the probability of future instability, combined with the preset graded early warning logic, the system outputs the early warning level and decision-making countermeasures.

[0072] This embodiment provides a method for assessing the safety status of dams by integrating multi-source monitoring data. The method aims to construct a dynamically corrected and predictable closed-loop assessment system to achieve comprehensive, accurate, and forward-looking management of the safety status of dams.

[0073] The process of this method involves collecting real-time monitoring data from multiple sources; multi-source heterogeneous data refers to data from different sources, of different types and units, with the aim of comprehensively sensing the dam's operating status; in this embodiment, this includes, but is not limited to: seepage pressure values ​​collected by seepage sensors, deformation values ​​collected by displacement sensors, and temperature values ​​collected by temperature sensors.

[0074] Based on preset minimum and maximum reference values, the real-time monitoring data undergoes min-max normalization to obtain a standardized dataset; the purpose of this step is to eliminate dimensional differences between different monitoring data, making them comparable; [The text then abruptly shifts to a different topic:] ...judgment... Is it equal to ,like That is, if the denominator is zero, then The value is always 0 or 0.5; if Then it is calculated using the following formula:

[0075] ;

[0076] The normalized dimensionless value is the output of this step.

[0077] These are the raw monitoring values, obtained in real time from previous steps;

[0078] The minimum reference value is a preset value, derived from the safety limits of the dam design or the statistical extreme values ​​of long-term historical monitoring data.

[0079] This is the preset maximum reference value, sourced from the same source. Determined based on design limits or statistical extreme values;

[0080] Through this processing, all raw monitoring data All are linearly mapped to the [0,1] interval. To enhance robustness, a pruning operation can be further performed, i.e., if Then let ;like Then let Ensure that all data are strictly within the range [0,1] to form a standardized dataset;

[0081] By combining standardized datasets, sensor 3D spatial coordinate information, and data acquisition timestamps, a spatiotemporal alignment process is performed using a preset interpolation algorithm to generate a spatiotemporally consistent fusion dataset. The purpose of this step is to solve the problems of different sampling frequencies and uneven spatial distribution among different sensors. In this embodiment, the preset interpolation algorithm may include linear interpolation or kriging interpolation, and the spatiotemporal alignment process specifically includes synchronization of the time dimension and mapping of spatial positions to unified 3D grid nodes. The final fusion dataset is a high-quality, high-density data foundation with a unified spatiotemporal reference.

[0082] The fused dataset is used as real-time boundary conditions and internal loads, input into the initial physical model constructed based on pre-set dam static data. Through multiphysics coupling, the internal physical field distribution of the dam body is inverted and calculated. The purpose of this step is to use monitoring data to invert and calculate key physical states within the dam body that cannot be directly measured. The pre-set dam static data refers to the dam's design drawings, geological exploration data, and material mechanics parameters. The initial physical model is a three-dimensional finite element (FEM) model constructed based on the aforementioned static data. This model defines initial material constitutive relations, such as the initial elastic modulus. and initial permeability coefficient Multiphysics coupling solution specifically refers to the simultaneous solution of the governing equations of the seepage field, stress field, and temperature field.

[0083] Extract the model-predicted displacement corresponding to the internal physical field distribution; this model-predicted displacement is denoted as... , is the displacement value calculated by the physical model under the current load at the same location as the actual monitoring point;

[0084] The displacement deviation is calculated by combining the actual monitored displacements in the fused dataset; the actual monitored displacement is denoted as... This data originates from the fused dataset collected and processed in the first step; the displacement deviation is denoted as... The absolute error is defined by the following formula:

[0085] ;

[0086] The displacement deviation is the calculation output of this step;

[0087] The displacement is predicted for the model, and its source is calculated from the preceding physics field inversion step;

[0088] The actual monitored displacement is obtained from data fusion module processing.

[0089] Determine whether the displacement deviation is greater than a preset displacement deviation threshold; the preset displacement deviation threshold is denoted as... It is a critical value determined based on engineering tolerance error or historical data statistics, used to judge the degree of distortion of the current physical model;

[0090] In response to a displacement deviation greater than a preset displacement deviation threshold, The key parameters of the initial physical model are automatically corrected using a preset inversion algorithm to generate a corrected physical model. The fused dataset is then input again as real-time boundary conditions and internal loads into this corrected physical model. Through multiphysics coupling, the corrected internal physical field distribution of the dam body is obtained through inversion calculation. This step constitutes a closed loop of dynamic correction. The preset inversion algorithm, such as the least squares method, aims to adjust the key parameters to reduce displacement deviations. Approaching a minimum; key parameters specifically refer to parameters that significantly affect the safety state of the dam, such as the elastic modulus from Revised to and permeability coefficient from Revised to ;

[0091] In response to the displacement deviation not exceeding a preset displacement deviation threshold, i.e. If so, the initial physical model is determined as the corrected physical model; this means that the parameters of the current model still accurately reflect the actual state of the dam, and no adjustment is needed, so it can be directly used for subsequent analysis.

[0092] The corrected internal physical field distribution is extracted using a preset pattern recognition algorithm to generate a multidimensional risk feature vector. To further clarify, the preset pattern recognition algorithm, such as a CNN, may specifically include: constructing a three-dimensional convolutional neural network, the input of which is the corrected physical field distribution data, for example, represented as... The data consists of three-dimensional mesh nodes, where each node contains physical quantities such as seepage pressure and stress. The network structure may include two three-dimensional convolutional layers, for example, using... A convolutional kernel and a ReLU activation function are followed by a 3D max pooling layer, for example, a pooling window. Finally, the multidimensional risk feature vector is output through one or more fully connected layers. The CNN model is pre-trained based on historical incident data, with physical field data as input and corresponding historical risk states, such as stress concentration and high seepage gradient, as labels.

[0093] This step aims to automatically extract deep-level features representing risk from complex physical field data; the preset pattern recognition algorithm, such as a convolutional neural network (CNN) trained on historical data, can automatically identify regions with high seepage gradients, stress concentration areas, etc.; the output multidimensional risk feature vector is denoted as... This is a highly condensed summary of the current risk situation;

[0094] Based on a multidimensional risk feature vector, the current comprehensive health index is determined through a pre-defined mapping function and linear weighted summation. This step aims to reduce the dimensionality of the multidimensional and complex risk feature vector and quantify it into a single, intuitive indicator. The pre-defined mapping function is denoted as... , used to convert vectors This is converted into multiple normalized individual risk indicators, such as seepage risk. Stress risk and deformation risk The mapping function It is a transformation model trained based on expert experience or historical data; the current comprehensive health index is calculated using the following linear weighted summation formula, denoted as... :

[0095] ;

[0096] The comprehensive health index has a value range of [0,1], where 1 represents the optimal health status and 0 represents structural failure. It is the calculation output of this step.

[0097] The risk weight coefficient represents the contribution of each risk to overall health, and is obtained through pre-calibration or dynamic optimization.

[0098] This is a normalized single-item risk indicator, with a value range of [0,1], derived from the mapping function. Calculated;

[0099] The comprehensive health index is combined into a historical index series, and a predicted health index is generated through a pre-trained time series prediction model; to further clarify, the pre-trained time series prediction model... For example, LSTM can specifically include: constructing a network structure containing two stacked LSTM layers, for example, each layer with 64 hidden units, followed by a fully connected layer to output the future... Predicted values ​​at time steps The model The input is a fixed length Historical index series The model is pre-trained using supervised learning, employing a long-term historical health index sequence of the dam as the training dataset. For example, using the previous... Using the data from the nth time moment as input features, the data from the next nth time moment... The data at each time point is used as a label, and the mean squared error (MSE) is used as a loss function for optimization.

[0100] The purpose of this step is to shift from assessing the present situation to predicting the future; the historical index sequence is denoted as... It is composed of current and past health indices, i.e. The pre-trained time series prediction model is denoted as... For example, Long Short-Term Memory (LSTM) networks are based on historical sequences. Extrapolation is performed, and the calculation method is as follows:

[0101] ;

[0102] This step is the output of the calculation to predict health index.

[0103] For pre-trained prediction models such as LSTM;

[0104] This is a historical index sequence, derived from previous steps. Composed of values;

[0105] The time window for prediction is a preset parameter, such as the next 7 days;

[0106] Based on the predicted health index and confidence interval, the preset critical failure threshold, and the historical index sequence, the probability of future instability is calculated; this step quantifies the likelihood of future danger; the preset critical failure threshold is denoted as... It is a critical health index characterizing dam failure, determined based on dam design specifications or operational risk management systems, for example... The probability of future instability is denoted as It is defined by the following conditional probability formula:

[0107] ;

[0108] This is the output of the calculation in this step, used to predict the probability of instability.

[0109] To predict events where the health index falls below the failure threshold; The confidence interval, i.e. the predicted distribution, is determined by the model. supply;

[0110] The known historical index sequence is the condition;

[0111] Based on the current comprehensive health index and the probability of future instability, combined with a pre-defined tiered early warning logic, the system outputs early warning levels and corresponding decision-making strategies. This step aims to transform the assessment and prediction results into actionable management actions. The tiered early warning logic is based on... That is, the current state and This means a dual judgment mechanism for future trends, which outputs corresponding early warning levels such as normal, attention, warning, danger, and decision-making countermeasures such as routine monitoring, increasing frequency, preparing contingency plans, and immediate response.

[0112] The method described in this embodiment constructs a high-fidelity digital twin of the dam by fusing multi-source heterogeneous monitoring data and using data-driven physical models for dynamic correction. It can not only accurately invert the physical field distribution inside the dam, but also realize the transformation from state assessment to trend prediction by integrating two dimensions: health index and future instability probability. This method forms a complete closed loop of data acquisition, model correction, state assessment, trend prediction, and decision warning, which significantly improves the scientificity, timeliness, and foresight of dam safety management and provides strong decision support for disaster prevention and mitigation.

[0113] Example 2:

[0114] Spatiotemporal alignment is performed using a preset interpolation algorithm, including:

[0115] Based on the three-dimensional spatial coordinate information and data acquisition timestamps of each sensor, the time dimension of the standardized dataset is synchronized using linear interpolation or Kriging interpolation algorithms.

[0116] The synchronized data is mapped onto preset unified dam body 3D spatial grid nodes to construct a spatiotemporally consistent fused dataset.

[0117] To further clarify, the method for assessing the safety status of a dam by fusing multi-source monitoring data described in Example 1, wherein the step of performing spatiotemporal alignment processing through a preset interpolation algorithm further limits the way to construct a spatiotemporally consistent fused dataset, providing high-quality input for subsequent multi-physics coupling solutions;

[0118] This step may specifically include:

[0119] Based on the three-dimensional spatial coordinate information and data acquisition timestamps of each sensor, linear interpolation or Kriging interpolation algorithms are used to synchronize the time dimension of the standardized dataset. Linear interpolation is suitable for scenarios where data changes are relatively stable and is simple to calculate. Kriging interpolation is an unbiased optimal interpolation based on spatial statistics, which is suitable for data with strong spatial correlation and can provide the variance of the interpolation results. The choice of algorithm can be preset according to the density of sensors and the spatiotemporal variability of data.

[0120] The synchronized data is mapped onto the pre-defined unified dam body 3D spatial grid nodes to construct a spatiotemporally consistent fused dataset; the unified dam body 3D spatial grid refers to the grid divided when constructing the initial physical model such as the FEM model; this mapping process ensures that monitoring data from different sources, such as seepage pressure and temperature, correspond spatially to the calculation nodes of the physical model;

[0121] Through the synchronization of the time dimension and the grid mapping of the spatial dimension, this embodiment ensures that the data input to the physical model is completely consistent in terms of spatiotemporal reference, avoiding inversion distortion caused by mismatch in data frequency and location; this greatly improves the quality of the fused dataset, which is a key prerequisite for the accuracy of subsequent physical field inversion calculations.

[0122] Example 3:

[0123] Solving through multiphysics coupling includes:

[0124] The governing equations that couple the seepage field, stress field and temperature field are solved. The stress field is calculated based on the effective stress principle, which is used to describe the relationship between stress, pore pressure and temperature stress in porous media.

[0125] To further clarify, the dam safety status assessment method that integrates multi-source monitoring data described in Example 1 further limits the complex interaction between multiple physical fields such as seepage, stress, and temperature inside the dam through the multi-physics field coupling solution step;

[0126] This step may specifically include:

[0127] Solve the governing equations that couple the seepage field, stress field, and temperature field; this means that the three physical fields are no longer calculated independently, but are closely linked through the interaction terms in the governing equations, such as the influence of pore water pressure on the stress field, and the influence of temperature changes on the stress field and seepage field.

[0128] The governing equations for seepage not only describe the motion of fluid in porous media, but also couple the effects of stress and temperature fields; specifically, the permeability coefficient of the dam body... It is not a constant initial value, but rather the effective stress. and temperature The function is represented as This is used to characterize the dynamic changes in the permeability of the dam body when it is compacted or relaxed and when the fluid viscosity changes.

[0129] The temperature field governing equations not only describe the heat conduction within the dam body but also couple the effects of the seepage field. Specifically, a heat convection term must be added to the heat conduction equations, which is related to the seepage velocity of the pore water. and fluid heat capacity The correlation is used to characterize the redistribution of temperature distribution inside the dam by seepage, i.e., heat-water coupling;

[0130] The stress field governing equations are based on the mechanical equilibrium equations and are coupled with the effects of the seepage field and the temperature field.

[0131] This three-field fully coupled solution mechanism ensures that the model can capture factors such as pore pressure caused by high water levels. Increase, thereby reducing effective stress Simultaneous seepage Changed the temperature field , and temperature The changes, in turn, affect the permeability coefficient. and thermal stress This series of complex chain reactions;

[0132] The stress field is calculated based on the effective stress principle, which describes the relationship between stress, pore pressure, and temperature stress in porous media. In the specific technical context of this invention, this principle can be expressed as:

[0133] ;

[0134] The effective stress tensor is the actual stress that causes deformation of the soil and rock mass, and is the output of this step.

[0135] The total stress tensor is derived from external loads such as water pressure;

[0136] The Biot coefficient is the permeation coupling coefficient, a dimensionless material property, which is provided by the initial physical model or the modified physical model.

[0137] The pore water pressure scalar is derived from the seepage field model based on the integrated seepage pressure monitoring data;

[0138] Unit tensor;

[0139] This is the temperature stress tensor, derived from the temperature field model based on fused temperature monitoring data;

[0140] This formula ensures pore water pressure and temperature stress For total stress The reduction or superposition effect is accurately calculated, thus obtaining a more realistic effective stress. distributed;

[0141] By employing a multiphysics coupling solution based on the effective stress principle, this embodiment can more realistically reflect the mechanical properties of the dam as a porous medium; it accurately captures the direct influence of seepage and temperature changes on the internal stress state of the dam body, and its inverse calculation of the internal physical field distribution, such as the effective stress field, is also reflected. and seepage field Compared to solving a single field, it has higher fidelity and provides a more reliable basis for subsequent state assessment.

[0142] Example 4:

[0143] The key parameters of the initial physical model are automatically corrected through a preset inversion algorithm, including:

[0144] The least squares method is used to dynamically adjust the elastic modulus and permeability coefficient of the initial physical model to minimize the displacement deviation, thereby generating the corrected physical model.

[0145] To further clarify, the method for assessing the safety status of a dam by integrating multi-source monitoring data described in Example 1, wherein the step of automatically correcting the key parameters of the initial physical model through a preset inversion algorithm further defines the establishment of a digital twin model that can reflect the current real physical characteristics of the dam in real time.

[0146] This step may specifically include:

[0147] The least squares method is used to dynamically adjust the elastic modulus and permeability coefficient of the initial physical model to minimize the displacement deviation, thereby generating the corrected physical model.

[0148] Least squares is a mathematical optimization technique, and its application here aims to find an optimal set of parameters. This enables the model to predict displacement. Compared with actual monitored displacement The sum of squares of the differences between them is Minimum;

[0149] Elastic modulus E and permeability coefficient K were selected as key parameters because they dominate the stress field deformation and seepage field, respectively, and are most susceptible to changes with dam aging or damage; when displacement deviation Exceeding the limit At that time, the system automatically triggers the inversion algorithm to convert the initial model... Revised to be new This new model, the revised physical model, will be used for calculations in the next time step.

[0150] By employing the least squares method to dynamically invert and correct the elastic modulus and permeability coefficient, this embodiment achieves an adaptive model correction mechanism. This makes the physical model no longer a static initial model based on design values, but a high-fidelity digital twin that can reflect the changes in modulus caused by aging and damage of dam material properties in real time, greatly improving the accuracy of the internal physical field distribution output by the model.

[0151] Example 5:

[0152] Based on a multidimensional risk feature vector... the current comprehensive health index is determined, including:

[0153] The multidimensional risk feature vector is converted into multiple normalized single risk indicators through a preset mapping function. The single risk indicators include: seepage risk indicator, stress risk indicator and deformation risk indicator.

[0154] By using linear weighted summation, multiple individual risk indicators are weighted with their corresponding preset risk weight coefficients to obtain the cumulative risk value.

[0155] Subtract the accumulated risk value from 1 to obtain the comprehensive health index.

[0156] To further clarify, the dam safety status assessment method integrating multi-source monitoring data described in Example 1, particularly the step of determining the current comprehensive health index based on a multi-dimensional risk feature vector, further limits the application of high-dimensional, complex risk feature vectors. This is transformed into a single, intuitive, and quantifiable health assessment indicator. ;

[0157] This step may specifically include:

[0158] A pre-defined mapping function is used to convert multidimensional risk feature vectors into multiple normalized single-item risk indicators; to further clarify, the pre-defined mapping function... This can be implemented in the following example: Let there be a multidimensional risk feature vector. , It can be defined as a simplified set of regression models or logical rules. For example, seepage risk indicators. Through the Specific characteristic components that characterize high seepage gradients (such as...) Logistic regression calculation yielded the following result: ,in These parameters are obtained through training and calibration based on historical data. Similarly, the stress risk index... and deformation risk indicators It can also be done in a similar way, by utilizing The different feature components and corresponding calibration parameters are calculated; the preset mapping function is obtained. It is a conversion model trained based on expert experience or historical data; it will Features associated with specific risks, such as high seepage gradient and high stress concentration, are mapped as follows:

[0159] Seepage risk indicator is marked as ;

[0160] Stress risk indicator is marked as ;

[0161] Deformation risk indicator is marked as ;

[0162] All three indicators are normalized to the [0,1] range, with larger values ​​indicating higher individual risk.

[0163] By using linear weighted summation, multiple individual risk indicators are weighted with their corresponding preset risk weight coefficients to obtain the cumulative risk value.

[0164] Cumulative risk value ;

[0165] in Preset risk weight coefficients for each risk;

[0166] Subtracting the accumulated risk value from 1 yields the comprehensive health index:

[0167] ;

[0168] This 1-risk structure makes The value, i.e., the comprehensive health index, is positively correlated with the health status of the dam. This indicates optimal health, which is defined as a total risk of 0. Indicating failure as a total risk of 1 is more intuitive;

[0169] By employing this feature extraction-single-item quantization-weighted synthesis approach, this embodiment successfully reduces complex, multidimensional physical field information to a single comprehensive health index H; this index is not only intuitive and easy to understand, but its composition is also... Clarity allows managers to see When the value decreases, it is possible to trace which individual risk factor, seepage, stress, or deformation is playing a dominant role, providing more refined guidance for subsequent decision-making.

[0170] Example 6:

[0171] Preset risk weighting coefficients, determined in the following ways:

[0172] Based on a pre-set expert knowledge base, the initial values ​​of risk weight coefficients are set using the analytic hierarchy process. This constitutes an initial assignment stage that combines theory with expert experience, providing a scientific and reasonable starting point for subsequent data-driven dynamic optimization using historical risk case data, thus achieving an organic combination of theory and practice.

[0173] Using historical incident data, the initial values ​​are trained and dynamically optimized to calibrate the risk weight coefficients, and the risk weight coefficients satisfy the normalization constraint.

[0174] To further clarify, the method for determining the preset risk weight coefficient in the dam safety status assessment method that integrates multi-source monitoring data described in Example 5 is specifically limited; the purpose of this step is to ensure the weight used to calculate the comprehensive health index H. It is objective, accurate, and dynamically optimized;

[0175] The methods for determining weights include:

[0176] Based on a pre-defined expert knowledge base, the Analytic Hierarchy Process (AHP) is used to set initial values ​​for risk weight coefficients. The expert knowledge base contains engineering experience and judgments in the field of dam safety. AHP is a decision analysis method that transforms qualitative judgments into quantitative weights. Through AHP, the relative importance of seepage, stress, and deformation to the overall safety of the dam can be systematically compared, resulting in a set of initial weights. ;

[0177] Using historical hazard case data, the initial values ​​are trained and dynamically optimized to calibrate the risk weight coefficients; this step introduces data-driven correction; historical hazard case data refers to past dam hazards such as seepage failure, slippage instability and their corresponding monitoring data and R index; through machine learning or optimization algorithms, the initial weights are trained so that the H value can accurately reflect the high-risk state when historical hazards occur, ensuring that the H index has accurate sensitivity to different types of failure modes;

[0178] The calibrated risk weight coefficients satisfy the normalization constraint:

[0179] ;

[0180] This constraint ensures that the total weight is 1, making the calculation of the H value mathematically sound;

[0181] By combining initial values ​​assigned by AHP experts with training and optimization using historical data, the risk weight coefficients determined in this embodiment not only incorporate valuable expert experience but are also objectively verified and dynamically calibrated using real data. This makes the weight allocation more scientific and reasonable, avoids the one-sidedness of purely subjective weighting, and significantly improves the accuracy and reliability of the comprehensive health index H as an evaluation indicator.

[0182] Example 7:

[0183] Based on the pre-defined tiered early warning logic, the system outputs the early warning level and corresponding decision-making strategies, including:

[0184] When the current comprehensive health index is less than or equal to the critical failure threshold, or the probability of future instability is greater than the second preset probability threshold, a dangerous state is output.

[0185] When the current comprehensive health index is greater than the critical failure threshold, and the probability of future instability is not greater than the second preset probability threshold, and (the current comprehensive health index is less than or equal to the second preset threshold, or the probability of future instability is greater than the first preset probability threshold), an early warning status is output.

[0186] When the current comprehensive health index is greater than the second preset threshold, and the current comprehensive health index is less than or equal to the first preset threshold, and the probability of future instability is not greater than the first preset probability threshold, output the attention status;

[0187] When the current comprehensive health index is greater than the first preset threshold and the probability of future instability is not greater than the first preset probability threshold, the normal state is output.

[0188] To further clarify, the dam safety status assessment method integrating multi-source monitoring data described in Example 1, which combines a preset hierarchical early warning logic to output early warning levels and decision-making countermeasures, further limits the quantitative assessment results. and prediction results Transformed into a clear and actionable four-level early warning response;

[0189] The preset tiered early warning logic is based on the current comprehensive health index. Probability of future instability Make a judgment; in this logic, For example, the critical failure threshold ; The second preset probability threshold, or danger probability threshold, is, for example... ; The first preset probability threshold, or attention probability threshold, is, for example... The above All of these are derived from dam design specifications or operational risk management frameworks; For example, the first preset threshold. ; For example, the second preset threshold. ; Similarly, it is derived from the operational risk management framework or determined through historical data statistical analysis, and fully considers the safety requirements of the dam throughout its planning and design life cycle;

[0190] when or When a dangerous situation is detected, an emergency response should be initiated immediately, including measures such as reservoir discharge or personnel evacuation.

[0191] when and ,and or When an early warning status is issued, an emergency plan should be prepared and a detailed on-site investigation should be organized.

[0192] when and ,and When this happens, the status of attention should be displayed, and the monitoring frequency should be increased to analyze the dominant risks.

[0193] when and When the output is in normal condition, normal monitoring is maintained.

[0194] The warning logic in this embodiment innovatively combines the current state. Value and future trends The value is used for dual judgment; it overcomes the lag of traditional threshold early warning by introducing... or This enabled proactive early warning and bought valuable time for risk management.

[0195] Example 8:

[0196] Please see Figure 2 A dam safety status assessment system integrating multi-source monitoring data, comprising:

[0197] The data fusion module is used to collect real-time monitoring data from multiple heterogeneous sources and perform min-max normalization and spatiotemporal alignment processing to generate a spatiotemporally consistent fused dataset.

[0198] The physical inversion module is used to input the fused dataset into the initial physical model and solve it through multi-physics coupling to invert and calculate the physical field distribution inside the dam body;

[0199] The model correction module is used to calculate the displacement deviation between the model-predicted displacement and the actual monitored displacement, and to generate a corrected physical model based on the comparison result of the displacement deviation with the preset displacement deviation threshold.

[0200] The status assessment module is used to extract the internal physical field distribution output by the corrected physical model into a multi-dimensional risk feature vector and calculate the current comprehensive health index.

[0201] The trend prediction module is used to combine comprehensive health indices into a historical index sequence to generate a predicted health index and calculate the probability of future instability.

[0202] The early warning decision module is used to output the early warning level and decision-making countermeasures based on the current comprehensive health index and the probability of future instability, combined with the preset hierarchical early warning logic.

[0203] This embodiment provides a dam safety status assessment system that integrates multi-source monitoring data. The system includes:

[0204] The data fusion module is configured to collect real-time monitoring data from multiple heterogeneous sources, and performs minimum-maximum normalization and spatiotemporal alignment processing through built-in normalization and spatiotemporal interpolation algorithms to generate a spatiotemporally consistent fusion dataset.

[0205] The physical inversion module is configured with an initial physical model built based on the static data of the dam, and receives the fused dataset generated by the data fusion module as input. The fused dataset is input into the initial physical model, and the physical field distribution inside the dam is inverted and calculated through multi-physics coupling solution.

[0206] The model correction module obtains the model predicted displacement from the physical inversion module and the actual monitored displacement from the data fusion module to calculate the displacement deviation between the model predicted displacement and the actual monitored displacement. Based on the comparison result of the displacement deviation with the preset displacement deviation threshold, this module calls the inversion algorithm to generate a corrected physical model when the deviation exceeds the limit.

[0207] The status assessment module receives the internal physical field distribution output by the model correction module, extracts the internal physical field distribution into a multi-dimensional risk feature vector, and calculates the current comprehensive health index.

[0208] The trend prediction module obtains the current comprehensive health index from the state assessment module, combines the comprehensive health index into a historical index sequence, and calls a pre-trained time series prediction model to generate a predicted health index, thereby calculating the probability of future instability.

[0209] The early warning decision module, based on the current comprehensive health index output by the status assessment module and the future instability probability output by the trend prediction module, combined with the preset hierarchical early warning logic, outputs the early warning level and decision-making countermeasures.

[0210] The system described in this embodiment, through the clear division and collaborative work of six major functional modules, solidifies the complex data fusion, physical inversion, model correction, state assessment, trend prediction, and early warning decision-making process into an automated and intelligent technical platform; it provides a complete physical solution for the methods defined in embodiments 1-7, and can realize real-time closed-loop, dynamic prediction, and scientific decision-making of dam safety status, and has extremely high engineering application value.

[0211] Example 9:

[0212] The method of this invention is executed on a dam that has been equipped with multi-source heterogeneous sensors and has experienced a flood season with high water levels and subsequent rapid water level decline.

[0213] The system's data fusion module automatically collects real-time monitoring data such as seepage pressure, deformation, and temperature at key monitoring sections of the dam. Through min-max normalization and spatiotemporal alignment, it generates a spatiotemporally consistent fused dataset. The physical inversion module inputs the fused dataset into an initial three-dimensional finite element (FEM) physical model constructed based on the dam design data. The model correction module calculates the model-predicted displacement of the dam area. Compared with actual monitored displacement deviation Displacement deviation greater than the preset threshold The system then triggers an inversion algorithm based on the least squares method, automatically determining the elastic modulus of the initial physical model of the region. Appropriately lower the permeability coefficient By appropriately adjusting the parameters, a revised physical model reflecting the current material properties of the dam is generated. The state assessment module, based on the internal physical field distribution derived from the revised physical model, extracts multi-dimensional risk feature vectors using a pre-trained convolutional neural network (CNN) and calculates the current comprehensive health index. The value is 0.72; subsequently, the trend prediction module will... Value added to historical index sequence It uses a pre-trained Long Short-Term Memory (LSTM) network model to generate a predicted health index and calculates the probability of future instability within the next 7 days. The figure is 3%; the early warning decision module receives the current... and the future Based on the preset tiered early warning logic, a threshold for the probability of attention is set. Danger probability threshold First preset threshold The second preset threshold Due to the current Less than And the probability of future instability Greater than The system determines that the triggering conditions for the early warning state are met, automatically outputs the early warning level, and provides decision-making strategies for preparing emergency plans and organizing detailed on-site investigations.

[0214] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for assessing the safety status of a dam by integrating multi-source monitoring data, characterized in that, include: Collect real-time monitoring data from multiple heterogeneous sources; Based on preset minimum and maximum reference values, the real-time monitoring data is subjected to minimum-maximum normalization to obtain a standardized dataset. By combining standardized datasets, sensor 3D spatial coordinate information, and data acquisition timestamps, a spatiotemporal alignment process is performed using a preset interpolation algorithm to generate a spatiotemporally consistent fused dataset. The fused dataset is used as real-time boundary conditions and internal loads, and input into the initial physical model constructed based on the preset static data of the dam. The distribution of physical fields inside the dam body is calculated by multi-physics coupling solution. Extract the model-predicted displacement corresponding to the internal physical field distribution; Calculate the displacement deviation by combining the actual monitored displacements in the fused dataset; Determine whether the displacement deviation is greater than the preset displacement deviation threshold; If the displacement deviation exceeds a preset displacement deviation threshold, the key parameters of the initial physical model are automatically corrected using a preset inversion algorithm to generate a corrected physical model. If the displacement deviation is not greater than the preset displacement deviation threshold, the initial physical model is determined as the corrected physical model. The internal physical field distribution output by the corrected physical model is extracted using a preset pattern recognition algorithm to generate a multidimensional risk feature vector. Based on a multidimensional risk feature vector, the current comprehensive health index is determined by a preset mapping function and a linear weighted summation. The comprehensive health index is combined into a historical index sequence, and a predicted health index is generated through a pre-trained time series prediction model. Calculate the probability of future instability based on the predicted health index and confidence interval, the preset critical failure threshold and historical index sequence; Based on the current comprehensive health index and the probability of future instability, combined with the preset graded early warning logic, the early warning level and decision-making countermeasures are output. Solving through multiphysics coupling includes: The governing equations that couple the seepage field, stress field and temperature field are solved. The stress field is calculated based on the effective stress principle, which is used to describe the relationship between stress, pore pressure and temperature stress in porous media. The key parameters of the initial physical model are automatically corrected through a preset inversion algorithm, including: The least squares method is used to dynamically adjust the elastic modulus and permeability coefficient of the initial physical model to minimize the displacement deviation, thereby generating the corrected physical model. The current comprehensive health index is determined based on a multidimensional risk feature vector, including: The multidimensional risk feature vector is converted into multiple normalized single risk indicators through a preset mapping function. The single risk indicators include: seepage risk indicator, stress risk indicator and deformation risk indicator. By using linear weighted summation, multiple individual risk indicators are weighted with their corresponding preset risk weight coefficients to obtain the cumulative risk value. Subtract the accumulated risk value from 1 to obtain the comprehensive health index; Preset risk weighting coefficients, determined in the following ways: Based on a pre-set expert knowledge base, the initial values ​​of the risk weight coefficients are set using the analytic hierarchy process. Using historical incident data, the initial values ​​are trained and dynamically optimized to calibrate the risk weight coefficients, and the risk weight coefficients satisfy the normalization constraint.

2. The method for assessing the safety status of a dam by integrating multi-source monitoring data according to claim 1, characterized in that, Spatiotemporal alignment is performed using a preset interpolation algorithm, including: Based on the three-dimensional spatial coordinate information and data acquisition timestamps of each sensor, the time dimension of the standardized dataset is synchronized using linear interpolation or Kriging interpolation algorithms. The synchronized data is mapped onto preset unified dam body 3D spatial grid nodes to construct a spatiotemporally consistent fused dataset.

3. The method for assessing the safety status of a dam by integrating multi-source monitoring data according to claim 1, characterized in that, Based on the pre-defined tiered early warning logic, the system outputs the early warning level and corresponding decision-making strategies, including: When the current comprehensive health index is less than or equal to the critical failure threshold, or the probability of future instability is greater than the second preset probability threshold, a dangerous state is output. When the current comprehensive health index is greater than the critical failure threshold and the probability of future instability is not greater than the second preset probability threshold, and the current comprehensive health index is less than or equal to the second preset threshold, or the probability of future instability is greater than the first preset probability threshold, an early warning status is output. When the current comprehensive health index is greater than the second preset threshold, and the current comprehensive health index is less than or equal to the first preset threshold, and the probability of future instability is not greater than the first preset probability threshold, output the attention status; When the current comprehensive health index is greater than the first preset threshold and the probability of future instability is not greater than the first preset probability threshold, the normal state is output.

4. A dam safety status assessment system integrating multi-source monitoring data, applied to the dam safety status assessment method integrating multi-source monitoring data as described in any one of claims 1-3, characterized in that, include: The data fusion module is used to collect real-time monitoring data from multiple heterogeneous sources and perform min-max normalization and spatiotemporal alignment processing to generate a spatiotemporally consistent fused dataset. The physical inversion module is used to input the fused dataset into the initial physical model and solve it through multi-physics coupling to invert and calculate the physical field distribution inside the dam body; The model correction module is used to calculate the displacement deviation between the model-predicted displacement and the actual monitored displacement, and to generate a corrected physical model based on the comparison result of the displacement deviation with the preset displacement deviation threshold. The status assessment module is used to extract the internal physical field distribution output by the corrected physical model into a multi-dimensional risk feature vector and calculate the current comprehensive health index. The trend prediction module is used to combine comprehensive health indices into a historical index sequence to generate a predicted health index and calculate the probability of future instability. The early warning decision module is used to output the early warning level and decision-making countermeasures based on the current comprehensive health index and the probability of future instability, combined with the preset hierarchical early warning logic.

Images