A dike engineering monitoring and detection fusion early warning method and system based on data-physical model double coupling

Abstract

The application discloses a kind of based on data-physical model double coupling's embankment engineering monitoring detection fusion early warning method and system, method includes obtaining embankment engineering multi-source data, forms multi-source data set;Using Abaqus to build embankment simulation model, carry out seepage field analysis and slope stability analysis, obtain the position of wetting front and slope stability safety factor;Build data evaluation model with cloud model as core, combined with analytic hierarchy process, quantitatively depict data uncertainty and output embankment comprehensive safety level and score;The embankment simulation model and data evaluation model are bidirectionally coupled verification and parameter inversion, optimize simulation model parameters and fuse double model evaluation results, obtain high-precision comprehensive safety evaluation;Build long time series prediction model, input historical time series data, predict and carry out graded early warning.The application realizes the multidimensional comprehensive judgment to embankment safety state by fusing including real-time monitoring data such as seepage pressure, displacement, water level and detection report data such as transient electromagnetic method report.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring and early warning technology for dike engineering, and in particular to a method and system for monitoring, detection and early warning of dike engineering based on data-physical model dual coupling. Background Technology

[0002] Dike engineering is an important component of flood control and disaster reduction systems. During long-term operation, dikes are continuously affected by multiple factors, including water erosion, seepage, and changes in soil structure, making them highly susceptible to safety hazards such as piping, leakage, and landslides. To ensure the safe operation of dike projects, comprehensive, timely, and accurate monitoring and testing of the dike body and foundation are necessary to conduct scientific hazard identification and early warning.

[0003] Currently, the safety management of dike projects mainly relies on a combination of regular manual inspections and local monitoring instruments. Manual inspections are inefficient and struggle to achieve continuous real-time monitoring. Various monitoring systems are mostly deployed independently with fragmented data storage, lacking a unified platform for multi-source data fusion analysis and intelligent early warning. Regular reports obtained through geophysical exploration and other detection methods are also not organically linked with daily monitoring data. Regarding early warning analysis methods, single data-driven models lack characterization of the physical mechanisms of dike structural stress deformation and seepage failure, resulting in insufficient model interpretability. Furthermore, physical models based on soil mechanics and seepage mechanics have weak dynamic response capabilities to real-time monitoring data. How to effectively couple data-driven models with physical models to achieve deep integration of monitoring information and detection results, and construct an early warning method and system for dike projects that combines real-time performance and interpretability, is a pressing technical problem that needs to be solved in this field.

[0004] Chinese invention patent application CN202511256025.2 discloses a method and system for safety assessment of dike engineering. It obtains dike temperature change data, calculates the temperature gradient distribution, and combines it with historical seepage field data to predict piping paths and early warning levels using a random forest model. This method focuses on identifying piping risks by utilizing the correlation between the temperature field and the seepage field, but it does not involve deep integration of monitoring data and inspection report data, nor does it construct a two-way coupling verification mechanism between the data-driven model and the physical simulation model. Therefore, it lacks sufficient characterization of the physical mechanisms of dike structural stress deformation and seepage failure. Chinese invention patent application CN202410024971.3 discloses a method and system for flood control safety evaluation of dike engineering. It constructs an evaluation index system from four aspects: the importance of the protected object, the vulnerability of the project, the risk during flood season, and the soundness of operation and management. It uses the analytic hierarchy process (AHP) to determine the weights and classify safety levels. This method focuses on the macro-level evaluation of flood control safety during the flood season, but the evaluation indicators it relies on are mostly derived from static statistics and expert experience, lacking the ability to dynamically access and predict real-time monitoring data, making it difficult to achieve early warning of the evolution of dike safety status. Furthermore, current technologies do not provide a comprehensive safety early warning method and system for dike projects that organically integrates monitoring and detection data and synergistically couples data evaluation models with physical simulation models, thus failing to simultaneously meet the comprehensive early warning requirements of real-time performance, accuracy, and interpretability. Summary of the Invention

[0005] The purpose of this invention is to provide a safety early warning method and system for dike engineering that integrates monitoring and detection data, coupled data and physical models, in order to solve the dilemma of existing technologies that are difficult to integrate monitoring and detection data, realize the synergistic coupling of data-driven prediction and physical mechanism analysis, improve the real-time performance and accuracy of dike hazard early warning, and enhance the interpretability of early warning results.

[0006] Technical solution:

[0007] A data-physical model-based dual-coupling method for monitoring, detection, and early warning of dike engineering includes the following steps:

[0008] 1) Acquire multi-source data of the dike project to form a multi-source dataset; the multi-source data includes real-time monitoring data and detection data; the real-time monitoring data includes seepage pressure data, displacement data, and water level data; the detection data includes transient electromagnetic method detection reports, historical hazard record reports, and geological survey material parameters;

[0009] 2) Based on the multi-source data, a levee simulation model was constructed using Abaqus to conduct seepage field analysis and slope stability analysis, and to obtain the location of the seepage surface and the slope stability safety factor.

[0010] 3) Based on the preprocessed multi-source monitoring data, a data evaluation model with cloud model as the core and combined with the analytic hierarchy process is constructed to quantify the uncertainty of the data and output the comprehensive safety level and score of the dike.

[0011] 4) The levee simulation model and the data evaluation model are coupled and verified in two directions and the parameters are inverted. The simulation model parameters are optimized and the evaluation results of the two models are integrated to obtain a high-precision comprehensive safety evaluation.

[0012] 5) Construct a GRU-DMA long-term prediction model based on AdaBoosting ensemble learning, input historical time series data, predict the future multi-step evolution trends of seepage pressure, water level and displacement, and provide graded early warning based on the prediction results.

[0013] Further, the seepage pressure data mentioned in step 1) is collected by seepage gauges installed on the top of the dike, the upstream slope, the downstream slope, the toe of the slope, and inside the dike body; the displacement data is collected by displacement gauges installed on the top of the dike and inside the dike foundation; and the water level data is collected by water level gauges installed at representative cross sections of the river channel.

[0014] The geological exploration material parameters include density, elastic modulus, permeability coefficient, cohesion, and internal friction angle;

[0015] The historical incident record report includes the incident type, location, corresponding water level conditions, monitoring anomaly range, and handling results.

[0016] Further, the seepage field analysis in step 2) includes: modeling the cross-section of the embankment, assigning material parameters, setting displacement constraints, water level boundaries, gravity loads and predefined fields, using CPE4P mesh generation, and calculating the pore water pressure distribution and the location of the wetting surface;

[0017] The slope stability analysis adopts the strength reduction method, iteratively reducing cohesion and internal friction angle, and using the reduction coefficient corresponding to the inflection point of slope top displacement as the slope stability safety factor, which is then compared with the safety threshold for scoring.

[0018] The levee safety simulation model is used to analyze seepage field data and slope stability. The specific analysis includes: using Abaqus to perform seepage analysis on the levee cross-section based on the acquired piezometer data, taking the location of the seepage surface as the main feature, and judging the changes in the location of the seepage surface under different working conditions to represent the safety status of the levee and whether there are any hidden dangers; at the same time, combined with the test report data, the slope stability safety factor is calculated using the strength reduction method and compared with the specified safety threshold to judge the stability and safety issues of the levee.

[0019] Furthermore, the data evaluation model in step 3) includes:

[0020] The monitoring data were processed by removing 3σ outliers, performing linear interpolation completion, and standardizing the data to [0,1].

[0021] The expected value of the cloud model Ex is determined based on the security level threshold, the entropy En is determined by the mean absolute deviation, and the hyperentropy He is determined by the difference between the sample variance and the squared entropy, thus constructing a normal cloud model.

[0022] The certainty μ of the monitoring data is calculated using a forward cloud generator;

[0023] The weights of each indicator were calculated using the analytic hierarchy process (AHP), and the certainty of each indicator was weighted and integrated to obtain the overall safety score and grade of the dike.

[0024] Furthermore, step 3) also includes: extracting the secondary induced electromotive force, measurement point coordinates and depth data from the transient electromagnetic method detection report, and identifying the location and scale of hidden dangers such as cavities, cracks and seepage channels inside the dike through wavelet denoising and electrical inversion; correlating and comparing the hidden danger characteristics with the monitoring data, and iteratively optimizing the expectation, entropy and hyperentropy parameters of the cloud model.

[0025] The data evaluation model is built around a cloud model and combines the characteristics of multi-source heterogeneous data from levee engineering safety monitoring and detection to achieve accurate evaluation and risk assessment of key safety indicators such as levee seepage, settlement, displacement, and cracks. Based on the levee engineering safety level classification standard, three core parameters of the cloud model are constructed: expectation, entropy, and hyperentropy. The expectation corresponds to the standard threshold of each safety indicator, the entropy characterizes the fuzziness of the data, and the hyperentropy reflects the discrete characteristics of entropy. By synergistically constructing a normal cloud model through the three parameters, the uncertainty mapping from monitoring data to safety level is realized.

[0026] Further, step 4) of the bidirectional coupling verification and parameter inversion includes:

[0027] Using the output of the data evaluation model as a constraint, small-range perturbation iterations are performed on the cohesion and internal friction angle, and the safety factor is inverted into the embankment simulation model. The distribution of the inverted safety factor is compared with the historical safety factor distribution, and the optimal material parameters are obtained after convergence. The data evaluation model score and the simulation model safety factor score are weighted and fused to obtain the final comprehensive safety score.

[0028] Further, the GRU-DMA long-term prediction model in step 5) includes:

[0029] The base learner is composed of alternating GRU and DMA, with the historical time series of seepage pressure, water level, and displacement as input.

[0030] GRU is used to extract nonlinear long-range dependency features of sequences, while DMA is used to extract linear trends and correct phase lag.

[0031] The AdaBoosting algorithm is used to dynamically adjust the sample weights based on the validation set error, and the outputs of each base learner are weighted and integrated to obtain the prediction results for the next 72 time steps.

[0032] The input to the long-term prediction model is Where X represents the input feature matrix, P represents the historical time series data of seepage pressure, W represents the historical time series data of channel water level, and D represents the historical time series data of embankment settlement and displacement.

[0033] The output of the long-term prediction model is Where H(x) represents the prediction result of the prediction model for each safety variable at a predetermined future step size (e.g., 72 steps), and M represents the number of iteration rounds and the total number of base learners. Indicates the first The ensemble weights of each base learner Indicates the first The predicted output of each base learner.

[0034] The long-term prediction model is based on the AdaBoosting ensemble learning framework. The base learner consists of alternating cycles of GRU and DMA. Both DMA (double-weighted moving average) and GRU (gated recurrent unit) structures are existing technologies and will not be elaborated here.

[0035] The historical time series data is used as the input of the prediction system. During the iterative training of the model, the weight distribution of the training samples in the current round is dynamically adjusted according to the mean absolute error (MAE) or root mean square error (RMSE) of the base learner on the validation set in the previous round. This allows the alternately introduced GRU or DMA models to specifically learn the time series fluctuation features with large errors.

[0036] When the Individual learners When using the DMA model, it is employed for linear trend extraction and time lag correction of the data. The specific analysis includes: setting a fixed sliding window size and assigning custom weights to the historical data within the window with exponential decay; calculating a single-weighted sliding average and a double-weighted sliding average; and constructing a trend correction equation using the difference between the two. This step effectively eliminates random noise in the settlement displacement sequence and compensates for the phase lag caused by the single sliding average.

[0037] When the Individual learners When used as a GRU model, it is employed to extract nonlinear dynamic features from the data. Specific analysis includes: utilizing internal update and reset mechanisms to regulate the memory and forgetting of hidden states in long-term osmotic pressure responses caused by drastic water level fluctuations, thereby outputting a nonlinear trend prediction vector with long-range dependence. Through weighted alternating integration of GRU and DMA, a high-precision comprehensive prediction result is ultimately output.

[0038] Furthermore, the tiered early warning system described in step 5) includes:

[0039] Green alert: The forecast curve is stable and within a safe range, with a risk probability of <20%;

[0040] Blue alert: Steps 48-72 will exceed the first-level threshold, with a risk probability of approximately 50%.

[0041] Orange alert: Significant mutations occur within 24 steps, with a risk probability >80%;

[0042] Red alert: The risk probability is approaching 100% as the system continues to exceed its limit threshold and exhibits unstable characteristics.

[0043] This invention also provides a data-physical model-based dual-coupling system for monitoring, detection, and early warning of dike engineering, comprising:

[0044] The real-time monitoring module is used to collect and display real-time data on dike seepage pressure, displacement, water level, and rainfall.

[0045] The data management module is used for monitoring equipment information maintenance and multi-source data storage;

[0046] The historical report module is used to store historical monitoring data, detection reports, and incident records, and supports traceability queries;

[0047] The simulation analysis module is used to perform simulation calculations of levee seepage and slope stability based on Abaqus;

[0048] The data evaluation module is used to evaluate the safety level of dikes based on cloud models and the analytic hierarchy process.

[0049] The coupling verification module is used for bidirectional coupling of the data model and the physical model, parameter inversion, and fusion evaluation.

[0050] The prediction and early warning module is used for long-term time-series prediction and hierarchical early warning based on the AdaBoosting-GRU-DMA model;

[0051] The geographic information module is used to mark monitoring points on electronic maps and display their status.

[0052] Furthermore, the coupling verification module is configured to: use the data evaluation results to invert and optimize the cohesion and internal friction angle parameters of the simulation model, and to weightedly fuse the evaluation results of the two models;

[0053] The prediction and early warning module is configured to output prediction curves for seepage pressure, water level, and displacement for the next 72 steps, and to output early warning information in four levels: green, blue, orange, and red.

[0054] Compared with the prior art, the present invention has the following advantages:

[0055] 1. This invention achieves a multi-dimensional comprehensive assessment of the safety status of dikes by integrating real-time monitoring data, including seepage pressure, displacement, and water level, with test report data such as transient electromagnetic method reports. This method incorporates geological interpretation information from test reports into the routine safety analysis process, avoiding the bias that may result from a single monitoring indicator, and breaking down the barriers between monitoring data and test information, thereby improving the comprehensiveness and accuracy of dike hazard identification.

[0056] 2. This invention, based on a cloud model, quantifies the uncertainty of monitoring data, improving the objectivity and accuracy of evaluation. Compared to single-indicator evaluation, it combines the analytic hierarchy process (AHP) to achieve weighted fusion of multiple indicators, avoiding bias and enhancing the comprehensiveness of the evaluation. Simultaneously, it integrates the transient electromagnetic method report analysis path to accurately uncover hidden internal hazard data, compensating for the shortcomings of existing models. Furthermore, the model supports dynamic iterative optimization of parameters and is deeply embedded in an integrated platform for real-time computation, significantly improving decision-making efficiency compared to offline models. Its adaptability and practicality far exceed existing technologies, providing reliable support for dike safety management.

[0057] 3. This invention utilizes a coupled verification of dike safety using a dike simulation model and a data evaluation model. Compared to relying entirely on historical data from inspection reports, the data evaluation model is used to perform parameter inversion on the physical model, further improving the accuracy of the dike simulation model and making early warnings more precise. Simultaneously, the safety factor score calculated by the physical model is fused with the safety level score obtained from the data evaluation cloud model, resulting in a more comprehensive safety evaluation perspective and providing more reliable assurance for dike safety early warnings.

[0058] 4. This invention constructs an early warning system based on time series autocorrelation prediction coupled with data and physical models for safety verification. It can capture the dynamic changes of various indicators related to the safety assessment of dike projects, focusing not only on the safety status of real-time monitoring but also on real-time prediction. Risk rating and early warning functions provide decision-makers with intuitive results; the spatiotemporal evolution characteristics of potential risks are transformed into visualized early warning curves, which helps to strengthen the risk foresight period and implement preventative measures in advance, realizing a shift from a passive response to an early-predictive dike hazard identification model. Attached Figure Description

[0059] Figure 1 A flowchart illustrating the method of the present invention is shown;

[0060] Figure 2 A schematic diagram of the data evaluation model process of the present invention is shown;

[0061] Figure 3 A schematic diagram of the data-physical model dual-coupling verification process of the present invention is shown. Detailed Implementation

[0062] The technical solution of the present invention will be further described below with reference to the accompanying drawings in the embodiments of the present invention.

[0063] A data-physical model-based dual-coupling method for monitoring, detection, and early warning of dike engineering, such as... Figure 1 As shown, it includes the following steps:

[0064] Step 1: Obtain seepage pressure data, displacement data, water level data, and test report data of the area to be monitored in the dike, and store them in the database to form a multi-source dataset of the dike.

[0065] Piezometers are deployed at the top, slope, toe, and key sections within the levee to form a seepage pressure monitoring grid. Displacement gauges are deployed within the levee foundation and at key locations on the levee top to form a displacement deformation monitoring grid. Water level gauges are deployed at representative sections of the river section where the levee is located. For example, on a typical monitoring section, seepage measurement points can be deployed at the levee top, upstream slope, downstream slope, and toe, while displacement measurement points can be deployed at the levee top and within the foundation. Real-time data collected by the sensors is wirelessly transmitted to a remote server. Simultaneously, historical transient electromagnetic method testing reports and hazard record reports of the levee are compiled. Historical testing report data is stored in a database with a structured index, used to extract geological hazard characteristics, historical hazard experience information, and material data for the levee.

[0066] Geological hazard characteristic data includes hazard type: cavities, fissures, weak layers, seepage channels, concentrated seepage zones; spatial information: station number, depth, lateral location, burial depth, and area dimensions; and physical property parameters: resistivity / conductivity anomaly ranges, dielectric constant, and anomaly intensity. Historical hazard experience information includes hazard type: piping, soil erosion, landslide, cracks, and diffuse seepage; occurrence time, water level conditions, rainfall conditions, and treatment measures; hazard location, development duration, degree of damage, recurrence probability, and corresponding abnormal ranges of monitoring indicators (seepage pressure / displacement / water level), etc.

[0067] Data on seepage pressure, displacement, water level, geological hazards, historical risks, and materials are all stored in the database, forming a multi-source dataset for dikes.

[0068] This method of acquiring monitoring and detection data from multiple sources overcomes the limitations of traditional levee safety management, which relies solely on monitoring data or detection reports. By aggregating real-time monitored physical field data with information from detection reports, it not only improves the comprehensiveness of the levee's operational performance but also provides a multi-dimensional information foundation for the subsequent synergistic coupling of data-driven models and physical models, effectively enhancing the accuracy of levee hazard identification and the scientific nature of early warning decisions.

[0069] Step 2: Based on the real-time piezometer data and material data in the test report, establish a simulation model of the embankment to conduct seepage analysis and slope stability analysis.

[0070] Abaqus was used to simulate and analyze the cross-section of the dike, and the distribution map of the pore water pressure in the cross-section was output.

[0071] First, a model of the levee cross-section is created. Then, the model is filled with levee material data from the inspection report, such as density, elastic modulus, and permeability coefficient. A steady-state analysis step is set for the model, and field outputs are defined, including void ratio, seepage velocity, pore water pressure, and displacement. Boundary conditions are set for the model, employing displacement and rotation constraints in the x and y directions for the entire model. A pore water pressure boundary condition is set on the water-facing surface of the levee using data from a piezometer. Gravity loads are applied to the model, and two predefined fields—void ratio and saturation—are defined. Finally, the model is meshed using the CPE4P mesh type. After obtaining the calculation results, the required pore water pressure output diagram can be obtained from the visualization interface.

[0072] Pore ​​water pressure distribution maps can reflect the location of the wetting surface. The elevation of the wetting surface can indicate whether there are potential safety hazards in the levee. A higher elevation means a higher outlet point near the downstream slope toe or dam toe, and a larger hydraulic gradient. This significantly increases the risk of seepage failure, such as piping, soil erosion, contact scour, or uplift, which may lead to levee instability. At the same time, a high wetting surface extends the saturation zone, increases seepage flow, and may cause downstream waterlogging. A lower elevation indicates a smoother seepage path, greater head loss, a lower outlet point, or even the presence of an outlet point within the drainage body, resulting in a lower risk of seepage failure and a safer levee condition. Using this simulation analysis, accurate early warning models can be constructed.

[0073] For the stability analysis of embankment slopes, the strength reduction method is used for simulation and to output a stability safety factor. Specifically, the strength reduction method involves continuously reducing the shear strength parameters of the soil until the slope becomes unstable and fails; the reduction factor at this point is the safety factor. In embankments, two key parameters satisfying the Mohr-Coulomb criterion are the internal friction angle and cohesion.

[0074] Abaqus finds the critical point through iterative calculation, as expressed by the following formula: ; Where c represents cohesion. This represents the internal friction angle. When defining the material properties of the model, the Mohr-Coulomb parameter is defined, and the reduction factor F is defined to increase from 0.5 to 2.0 in a step of 0.25. The cohesion and internal friction angle at each step are iteratively calculated and input using the above formula. The remaining steps are similar to those in seepage analysis and will not be elaborated here. After obtaining the calculation results, the output cloud map of the defined field variables is obtained in the visualization interface. Then, a monitoring point is set at the top of the slope, and the relationship curve between its displacement and the reduction factor is plotted. The field variable corresponding to the obvious inflection point of the curve is the safety factor. After obtaining the safety factor, it is compared with the threshold given in the inspection report, scored on a percentage scale, and the existence of safety hazards in the embankment is analyzed.

[0075] The strength reduction method provides more accurate and comprehensive information than traditional methods, enabling a more refined assessment of dike safety.

[0076] Step 3: Based on the acquired multi-source dataset of the dikes, use a data evaluation model to conduct safety analysis and output the safety evaluation results, such as... Figure 2 .

[0077] First, the collected raw monitoring data were preprocessed to remove outliers caused by sensor malfunctions and environmental interference. A 3-standard-deviation interval was set for the data; only values ​​exceeding this interval were considered outliers. Fluctuations within this interval were not considered outliers, but rather likely sensor noise fluctuations. Linear interpolation was used to fill in the gaps, and standardization was applied to map all monitoring data to the [0,1] evaluation interval, resulting in the preprocessed levee monitoring sample set. , of which Effective monitoring data for a single monitoring indicator This represents the number of valid samples.

[0078] Secondly, based on the current standards for classifying the safety levels of dike projects, the safety grading standards and quantitative thresholds for core evaluation indicators such as seepage pressure, displacement, and water level are clarified. A cloud model for each core evaluation indicator is then constructed, and the expected value of the cloud model is determined. ,entropy hyperentropy The three core parameters are calculated as follows:

[0079] Expectations of the cloud model The formula for calculating the central level of the monitoring indicator is as follows: ;

[0080] Entropy is determined by the mean absolute deviation of the monitored data. This characterizes the fuzziness and dispersion of the indicator, and the calculation formula is as follows:

[0081] ;

[0082] Then calculate the hyperentropy based on the squared difference between the sample variance and the entropy. It reflects the fluctuations and uncertainties of entropy itself, and the calculation formula is:

[0083] .

[0084] A normal cloud model is constructed by synergistically using the above three parameters. A one-dimensional normal cloud is generated using a forward cloud generator to visualize the uncertainty of single-index monitoring data. Specifically, firstly, random entropy is generated with entropy as the expected value and hyperentropy as the standard deviation. The calculation formula is: Then, using the expectation as the center and the absolute value of the random entropy as the standard deviation, the cloud droplet abscissa is generated. The calculation formula is: Finally, calculate the degree of certainty. This characterizes the degree to which the monitoring data belongs to the normal operating status of the dike, and the calculation formula is as follows: .

[0085] By constructing a judgment matrix, performing consistency checks, and calculating weights, the weight proportions of core indicators such as seepage flow, levee crest settlement, and horizontal displacement are clarified. To address the weight differences among different monitoring indicators, the Analytic Hierarchy Process (AHP) is introduced to weight and fuse the determination results output by the cloud models of each single indicator, avoiding the one-sidedness of evaluation by a single indicator. Ultimately, a comprehensive evaluation of the overall safety level of the levee is completed, outputting a clear safety level classification and corresponding quantitative score.

[0086] Meanwhile, the model supports linked analysis with historical monitoring reports. The specific analysis technique for transient electromagnetic method (TEM) reports is as follows: First, the original observation data in the TEM report is extracted, including core information such as the decay curve of the secondary induced electromotive force, the coordinates of the observation point, and the detection depth. Wavelet analysis is used to remove noise signals. Combined with physical parameters such as the dielectric constant and conductivity of the embankment soil and rock, the electrical distribution characteristics of the soil and rock inside the embankment are obtained through inversion. The location, scale, and development degree of hidden hazards such as cavities, cracks, and seepage channels inside the embankment are accurately identified. The inversion results of the TEM report are compared and correlated with the current evaluation data and historical monitoring data of the cloud model to establish a correspondence between the development degree of hazards and the anomalies in the monitoring data, providing targeted data support for the iteration of cloud model parameters.

[0087] By continuously integrating multi-source monitoring data, the three core parameters of the cloud model—expectation, entropy, and hyperentropy—are dynamically and iteratively optimized to gradually reduce evaluation errors and improve evaluation accuracy, ensuring a high degree of match between the evaluation results and the actual safety status of the dikes. Furthermore, the model incorporates historical data traceability capabilities, enabling precise retrospective analysis of the changing trends of various monitoring indicators, parameter adjustment processes, and the evolution of evaluation results. This helps managers accurately locate potential dike hazards, analyze the causes of these hazards, effectively improve dike repair efficiency, reduce safety management costs, and ensure the long-term stable operation of dike projects.

[0088] Step 4: Perform coupled verification of the dike safety based on the dike safety simulation model and the data evaluation model to improve the accuracy of the early warning model and the comprehensive evaluation of the dike safety status.

[0089] First, coupled validation is used to improve model accuracy, such as... Figure 3 .

[0090] First, extract the historical safety factor of the dike section from the historical monitoring report database, and use the data evaluation model positive cloud generator to process the data and output the probability distribution map. The specific method has been explained in step 3 and will not be repeated here.

[0091] Secondly, considering that key parameters such as the internal friction angle and cohesion of the levee cross-section are affected by moisture content, density, and seepage environment, the parameter values ​​extracted in the test report are not accurate, and consequently, the accuracy of the model is also insufficient. Therefore, the safety factor value of the simulation model is obtained by inverting the evaluation results of the data evaluation model. The cohesion and internal friction angle are subjected to multiple small-scale perturbations through the inversion algorithm and input into the levee simulation model. The safety factor results of multiple inversions are output, thus obtaining a new probability distribution map of the safety factor.

[0092] Third, the safety factor distribution map obtained by inversion is compared with the distribution cloud map obtained from historical data. When the two are almost overlapping, they can be optimized into new parameter values, thereby improving the accuracy of the model and making safety assessment and hazard warning more accurate.

[0093] In step 2, the safety factor obtained from the levee simulation model is scored on a percentage scale within the safety threshold. In step 3, the fuzzy boundaries of different safety levels are characterized by the digital features (expectation, entropy, hyperentropy) of the cloud model. The safety levels are evaluated and scored according to the data certainty, and the real-time monitoring data is converted into level scores.

[0094] By employing a fusion evaluation algorithm, the grade scores of the two are weighted and calculated to obtain a more comprehensive safety analysis result that integrates levee simulation early warning and data probability evaluation, making the levee safety evaluation more accurate.

[0095] Step 5: Based on the historical autocorrelation sequences of the seepage pressure data, channel water level data, and dike settlement and displacement data, the evolution trend of each variable in the next 72 time steps is predicted by the constructed AdaBoosting long-term prediction model, and the risk probability and warning level are determined accordingly.

[0096] The AdaBoosting long-term prediction model employs an ensemble learning framework, with its base learners consisting of gated recurrent units (GRUs) and DMA (dual moving average) methods. For each monitored variable, the measured historical sequence at a specific measurement point is used as input. The AdaBoosting algorithm is then used to weight and integrate multiple alternately trained GRUs and DMA weak predictors to achieve predictions for data at the next 72 time steps.

[0097] Specifically, the early warning system models the autocorrelation characteristics of dike safety data. For variables with strong inertial characteristics, such as seepage pressure, water level, and displacement, the system extracts deep nonlinear dependencies in the sequence using GRU, and uses DMA to correct potential trend biases and time lag effects in predictions with a step length of up to 72, ensuring the physical rationality of long-term predictions.

[0098] The GRU model in the base learner utilizes its gating structure to perform feature mapping on the autocorrelation sequence. Taking channel water level as an example, the model learns the periodicity of water level fluctuations and its own feedback mechanism, enabling it to capture the nonlinear rise or fall patterns of water level within 72 steps. The hidden states of the GRU can retain information from historical key nodes, thereby suppressing information loss in long-distance prediction.

[0099] The DMA model in the base learner is specifically designed to extrapolate the linear trend of the autocorrelation sequence. Since single-step moving averages exhibit phase lag when processing displacement or pressure sequences with clear trends, this invention establishes a trend prediction equation using quadratic moving averages to compensate for and correct the linear increments within 72 steps. When the historical sequence shows a continuous upward trend, DMA can provide a stable linear baseline, helping the model maintain trend consistency in long-term predictions.

[0100] The AdaBoosting ensemble process involves iteratively optimizing the weights of the base learners. During the 72-step prediction process, to address the large cumulative error generated by the preceding model at later steps, the AdaBoosting algorithm automatically adjusts the sample weights, guiding subsequent GRU or DMA learners to focus on correcting the bias at the end of the prediction. Finally, by weightedly combining the outputs of the weak predictors, a complete prediction curve for each safety variable over the next 72 steps is obtained.

[0101] Using the predicted results of each variable over the next 72 steps as input variables, and through the coupled safety verification of the aforementioned data model and physical model, the safety status of the dike is assessed, thus constructing an early warning level assessment system based on multi-step trend evolution:

[0102] 1. Green Alert: If the prediction curves of all variables remain stable within the next 72 steps, all within the safe threshold range and without any signs of acceleration, the system outputs a risk probability of <20%;

[0103] 2. Blue Alert: If the predicted sequence shows that it may break through the first-level threshold within the next 48-72 step range, or the slope of the DMA trend term continues to increase, the system outputs a risk probability of about 50%.

[0104] 3. Orange alert: If the prediction curve shows a significant jump or accelerated rise within the next 24 steps, and the multivariate prediction results show unfavorable coupling, the system outputs a risk probability of over 80%.

[0105] 4. Red Alert: If the predicted value within 72 steps continuously exceeds the ultimate bearing threshold, and the GRU captures typical unstable nonlinear characteristics, the system outputs a risk probability of 100%, triggering the highest level of alert.

[0106] Example 2:

[0107] A safety monitoring and early warning platform for dike engineering includes:

[0108] The real-time monitoring module is used to visually display real-time data from various sensors in the monitored levee area in the form of charts, including rainfall data, water level change data, deep pressure data, and structural safety monitoring data.

[0109] The data management module is used to maintain and manage the basic information of sensor devices, enabling the addition, deletion, modification, and query operations of monitoring devices, and ensuring the integrity and accuracy of sensor information;

[0110] The historical report module is used to record and store historical monitoring data in chronological order, and generate monitoring reports with timestamps to facilitate subsequent data tracing and comparative analysis.

[0111] The details viewing module is used to display detailed information on various monitoring data such as rainfall, water level, monitoring points, piezometers, and structural safety, and supports data drill-down and related queries;

[0112] The geographic information module is used to mark the spatial distribution of monitoring points on an electronic map and to display the real-time status of each monitoring point in conjunction with map visualization technology.

[0113] The safety evaluation module analyzes various monitoring data based on a preset safety assessment algorithm, outputs a safety evaluation level, and generates corresponding early warning information.

[0114] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for monitoring, detection, and early warning of dike engineering based on data-physical model dual coupling, characterized in that, Includes the following steps: 1) Acquire multi-source data of the dike project to form a multi-source dataset; the multi-source data includes real-time monitoring data and detection data; the real-time monitoring data includes seepage pressure data, displacement data, and water level data; the detection data includes transient electromagnetic method detection reports, historical hazard record reports, and geological survey material parameters; 2) Based on the multi-source data, a levee simulation model was constructed using Abaqus to conduct seepage field analysis and slope stability analysis, and to obtain the location of the seepage surface and the slope stability safety factor. 3) Based on the preprocessed multi-source monitoring data, a data evaluation model with cloud model as the core and combined with the analytic hierarchy process is constructed to quantify the uncertainty of the data and output the comprehensive safety level and score of the dike. 4) The levee simulation model and the data evaluation model are coupled and verified in two directions and the parameters are inverted. The simulation model parameters are optimized and the evaluation results of the two models are integrated to obtain a high-precision comprehensive safety evaluation. 5) Construct a GRU-DMA long-term prediction model based on AdaBoosting ensemble learning, input historical time series data, predict the future multi-step evolution trends of seepage pressure, water level and displacement, and provide graded early warning based on the prediction results.

2. The method according to claim 1, characterized in that, The seepage pressure data mentioned in step 1) is collected by seepage gauges installed on the top of the dike, the upstream slope, the downstream slope, the toe of the slope, and inside the dike body; the displacement data is collected by displacement gauges installed on the top of the dike and inside the dike foundation; the water level data is collected by water level gauges installed at representative cross sections of the river channel. The geological exploration material parameters include density, elastic modulus, permeability coefficient, cohesion, and internal friction angle; The historical incident record report includes the incident type, location, corresponding water level conditions, monitoring anomaly range, and handling results.

3. The method according to claim 1, characterized in that, Step 2) The seepage field analysis includes: modeling the cross-section of the embankment, assigning material parameters, setting displacement constraints, water level boundaries, gravity loads and predefined fields, using CPE4P mesh generation, and calculating the pore water pressure distribution and the location of the wetting surface; The slope stability analysis adopts the strength reduction method, iteratively reducing cohesion and internal friction angle, and using the reduction coefficient corresponding to the inflection point of slope top displacement as the slope stability safety factor, which is then compared with the safety threshold for scoring.

4. The method according to claim 1, characterized in that, Step 3) The data evaluation model includes: The monitoring data were processed by removing 3σ outliers, performing linear interpolation completion, and standardizing the data to [0,1]. The expected value of the cloud model Ex is determined based on the security level threshold, the entropy En is determined by the mean absolute deviation, and the hyperentropy He is determined by the difference between the sample variance and the squared entropy, thus constructing a normal cloud model. The certainty μ of the monitoring data is calculated using a forward cloud generator; The weights of each indicator were calculated using the analytic hierarchy process (AHP), and the certainty of each indicator was weighted and integrated to obtain the overall safety score and grade of the dike.

5. The method according to claim 4, characterized in that, Step 3) also includes: extracting the secondary induced electromotive force, measurement point coordinates and depth data from the transient electromagnetic method detection report, and identifying the location and scale of hidden dangers such as cavities, cracks and seepage channels inside the dike through wavelet denoising and electrical inversion; correlating and comparing the hidden danger characteristics with the monitoring data, and iteratively optimizing the expectation, entropy and hyperentropy parameters of the cloud model.

6. The method according to claim 1, characterized in that, Step 4) The bidirectional coupling verification and parameter inversion includes: Using the output of the data evaluation model as a constraint, small-range perturbation iterations are performed on the cohesion and internal friction angle, and the safety factor is inverted by substituting them into the levee simulation model. By comparing the inverted safety factor distribution with the historical safety factor distribution, the optimal material parameters are obtained after convergence. The final comprehensive safety score is obtained by weighting and integrating the data evaluation model score and the simulation model safety coefficient score.

7. The method according to claim 1, characterized in that, Step 5) The GRU-DMA long-term prediction model includes: The base learner is composed of alternating GRU and DMA, with the historical time series of seepage pressure, water level, and displacement as input. GRU is used to extract nonlinear long-range dependency features of sequences, while DMA is used to extract linear trends and correct phase lag. The AdaBoosting algorithm is used to dynamically adjust the sample weights based on the validation set error, and the outputs of each base learner are weighted and integrated to obtain the prediction results for the next 72 time steps.

8. The method according to claim 7, characterized in that, Step 5) The tiered early warning includes: Green alert: The forecast curve is stable and within a safe range, with a risk probability of <20%; Blue alert: Steps 48-72 will exceed the first-level threshold, with a risk probability of approximately 50%. Orange alert: Significant mutations occur within 24 steps, with a risk probability >80%; Red alert: The risk probability is approaching 100% as the system continues to exceed its limit threshold and exhibits unstable characteristics.

9. A monitoring, detection, and early warning system for dike engineering based on data-physical model dual coupling, characterized in that, include: The real-time monitoring module is used to collect and display real-time data on dike seepage pressure, displacement, water level, and rainfall. The data management module is used for monitoring equipment information maintenance and multi-source data storage; The historical report module is used to store historical monitoring data, detection reports, and incident records, and supports traceability queries; The simulation analysis module is used to perform simulation calculations of levee seepage and slope stability based on Abaqus; The data evaluation module is used to evaluate the safety level of dikes based on cloud models and the analytic hierarchy process. The coupling verification module is used for bidirectional coupling of the data model and the physical model, parameter inversion, and fusion evaluation. The prediction and early warning module is used for long-term time-series prediction and hierarchical early warning based on the AdaBoosting-GRU-DMA model; The geographic information module is used to mark monitoring points on electronic maps and display their status.

10. The system according to claim 9, characterized in that: The coupling verification module is configured to: use the data evaluation results to invert and optimize the cohesion and internal friction angle parameters of the simulation model, and to weight and fuse the evaluation results of the two models; The prediction and early warning module is configured to output prediction curves for seepage pressure, water level, and displacement for the next 72 steps, and to output early warning information in four levels: green, blue, orange, and red.

Images