Concrete dam intelligent online monitoring method based on multi-source data fusion

Abstract

The invention discloses a concrete dam intelligent online monitoring method based on multi-source data fusion. The method comprises the following steps: S1, acquiring monitoring data; s2, judging the validity of the monitoring data; s3, monitoring data abnormity identification; s4, performing on-site inspection information abnormity identification; s5, structure safety rechecking is carried out; s6, comprehensively evaluating the safety condition of the dam; according to the invention, a full-service process closed-loop management system is constructed, so that full-link intellectualization from data acquisition to security decision is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent online monitoring of gravity dams, and in particular to an intelligent online monitoring method for concrete dams based on multi-source data fusion. Background Art

[0002] As the scale of water conservancy and hydropower projects expands and their service life increases, dam safety management faces increasingly severe challenges, and various new models, algorithms and technologies are gradually being applied to dam safety monitoring.

[0003] In terms of data collection, manual observation was the main method in the early days, using simple instruments with low efficiency and poor accuracy. With the development of electronic technology and computers, the 1970s saw the beginning of the automated monitoring stage, with the emergence of semi-automated or automated monitoring instruments, continuous improvement of collection devices, and greatly enhanced data transmission technology.

[0004] In terms of data analysis, since the 1950s, statistical models (such as Fanelli mixture model) and machine learning (such as ANN, SVM) have been gradually applied to dam monitoring. After the 1990s, localized models such as grey theory, fuzzy mathematics, and chaos analysis (such as DGM dynamic grey model) have emerged and applied to dam monitoring data analysis.

[0005] In terms of monitoring systems, hydropower development and operation units such as the Three Gorges Group have made huge investments in dam informatization and intelligentization, developed a number of excellent application systems, and played an important role in project construction and operation. Examples include the Three Gorges Group's intelligent dam construction information platform (iDam2.0 system), the State Power Investment Corporation's dam safety management and monitoring information system, and the Huaneng Lancang River Company's basin dam intelligent online monitoring platform.

[0006] However, to date, dam safety monitoring still has some deficiencies in data processing, model application, algorithm adaptation, and business design, specifically: 1. Insufficient data validity: Monitoring data are affected by instrument failures and noise interference, and often contain a large number of gross errors. Traditional threshold methods are difficult to adapt to complex data distributions, and the accuracy of gross error identification is low. The data reliability is insufficient, which has an adverse impact on subsequent structural property analysis.

[0007] 2. Inadequate Modeling: Existing dam safety monitoring relies primarily on mathematical models, which analyze and predict dam performance based on statistical laws. This makes it difficult to provide effective results for conditions exceeding historical operating conditions or extreme operating conditions. Mechanistic models should be added to mathematical models to analyze dam structural performance from physical mechanisms such as thermal mechanics. This would also provide the necessary support for structural safety reviews and engineering safety rehearsals.

[0008] 3. Insufficient model applicability: The accuracy of model algorithms is closely related to data characteristics, data quality, and distribution. Currently, a single algorithm is difficult to adapt to the influence of these multiple factors. To improve model recognition accuracy and algorithm applicability, it is necessary to accurately match the algorithm itself with the physical quantity to be monitored, so that each algorithm can handle the problem most appropriately.

[0009] 4. Insufficient domestic substitution: Although most of the current dam safety monitoring systems have been independently developed, the necessary components supporting the operation of the systems still use foreign commercial software, especially simulation software and three-dimensional visualization engines, which are heavily dependent on imports.

[0010] 5. Unclear business processes: Taking the validity determination of monitoring data as an example, the current practice is to close the process after the data is initially judged to be unqualified. There is a lack of specific and feasible business process designs for subsequent data retesting, multi-physical quantity evaluation, on-site comparison testing, instrument troubleshooting, etc.

[0011] 6. Insufficient comprehensive evaluation capabilities: After completing sub-item evaluations such as monitoring data analysis, structural safety review, and defect information identification, there is a lack of effective data fusion methods, making it impossible to integrate the individual evaluation results to conduct a comprehensive quantitative evaluation of dam safety. Summary of the Invention

[0012] The purpose of the present invention is to overcome the above-mentioned shortcomings and provide an intelligent online monitoring method for concrete dams based on multi-source data fusion, which realizes the full-link intelligence from data collection to safety decision-making by building a closed-loop management system for the entire business process.

[0013] To solve the above technical problems, the technical solution adopted by the present invention is: an intelligent online monitoring method for concrete dams based on multi-source data fusion, comprising the following steps: S1. Acquisition of monitoring data: Acquire real-time and historical monitoring data of all monitoring points from the database according to business needs; S2. Validity determination of monitoring data: Apply mathematical models to check the validity of monitoring data and eliminate invalid measurements; S3. Monitoring data anomaly identification: Based on design indicators, historical extreme values, limit states, and small probability methods, formulate monitoring indicators for each monitoring point in each monitoring project from the dimensions of threshold, rate, and acceleration, and use monitoring indicators to identify anomalies in monitoring data; S4. Identification of anomalies in on-site inspection information: Through semantic recognition, inspection records are structured, and various defect information and defect change information found during on-site inspections are used as input conditions for comprehensive dam safety assessment; S5. Structural safety review: When the measured operational performance is judged to be abnormal, the abnormal situation is used as an input condition to conduct a safety review of the corresponding building; when the actual load and its combination, mechanical parameters, calculation model are significantly different from the design, a review is conducted based on the actual operational situation; S6. Comprehensive evaluation of dam safety status: By establishing a dam safety evaluation rule system, integrating the results of measured operational performance analysis, structural safety review results and on-site inspection results, a comprehensive quantitative evaluation of the dam safety status is carried out, the dam safety level is determined, and early warnings are issued based on the evaluation results.

[0014] Preferably, step S2 specifically includes the following process: S2.1, model construction and adaptation, intelligently matching the optimal discrimination model according to the measurement point type and data time series characteristics; S2.2, Data Preliminary Screening: Daily new data is preliminarily judged by the model constructed in S2.1. Values ​​that pass the preliminary judgment are identified as valid and enter the monitoring data anomaly identification stage. Invalid values ​​enter S2.3; S2.3, Data Retest: The invalid values ​​determined in S2.2 are first retested by the station. After the retest is completed, the model is used for secondary discrimination. If the retested data is a valid value, the gross error record of the measuring point is saved and the process ends. If it is still determined to be an invalid value, proceed to S2.4. S2.4, Risk Classification and Disposal: For important measuring points, direct instructions are issued directly to the station to remind staff to conduct on-site inspections. For general measuring points, multiple physical quantity and multi-measurement evaluations are conducted, and data from nearby measuring points of the same type or different physical quantities in the same location are correlated. If all are qualified, the point is marked as "low risk" and included in the list of key measurement points, with increased monitoring frequency, and proceeding to S2.5. Otherwise, it is considered to be a high risk and a direct instruction is issued to the station to remind staff to conduct on-site inspections. S2.5: Monitor the measurement points listed in S2.4 for five days and then use the model to evaluate them. If they return to normal, the process ends. If they still fail, send a command to the station to remind staff to conduct an on-site inspection. If there is an instrument failure, repair it as soon as possible. If the measured value is indeed the true value, enter the monitoring data anomaly identification module and send an inspection command to the station, and then enter S2.6. S2.6. On-site inspection and defect handling: If cracks or water seepage defects are found after the inspection, a repair plan will be provided and included in the defect management, and the cusp gross error data will be retrieved. If the on-site inspection results are normal, the area will be added as an inspection area and included in the inspection management module. S2.7. Manual review and model optimization: Regularly conduct manual review of the identification results and verify the reliability of the model based on the review results; mark failed models and re-match the algorithm.

[0015] Preferably, in S2.1, model building includes the following methods: Logical discrimination method: Eliminate over-limit data based on the monitoring instrument range, monitoring accuracy, physical meaning of monitoring data and preset thresholds; Statistical discrimination method: Use historical data to establish a multiple regression model, define a confidence interval of 6 times the standard deviation, and judge it as invalid if it exceeds the interval; Cluster analysis method: using cluster analysis model to identify outlier data as gross errors; Adaptation means intelligently matching the optimal discrimination model based on the measurement point type and data timing characteristics.

[0016] Preferably, the S3 specifically includes the following process: S3.1. Determine monitoring targets and indicators: Select monitoring indicators based on dam failure modes, establish a risk transmission chain based on existing dam risk assessment results, and implement the selection of monitoring indicator items for risk control; select monitoring points based on the weak points of the dam, simulate and analyze the stress evolution process of the dam under different working conditions based on project practice, and analyze the failure transmission path; establish a perception time chain for perceptible effect quantities and determine the monitoring content of the monitoring items; select monitoring points based on dam monitoring data, and first conduct monitoring data evaluation for the selected monitoring indicators; S3.2, monitoring indicator comparison and abnormality identification; S3.3, Abnormal classification: According to the degree of deviation of the measured value from the monitoring index, the classification is combined with the structural safety stage; S3.4. Multi-point fusion and correlation analysis: Combine multi-point data to analyze the overall state of the structure and identify potential risk transmission paths; establish a monitoring data time chain and analyze timeliness responses; S3.5. Diagnosis and feedback optimization: Combine finite element analysis, inspection reports, and hidden danger records to locate the cause of abnormalities; regularly update monitoring indicator thresholds based on the latest monitoring data and risk assessment results; feed back abnormality identification results to the "four prevention" system to improve the risk management chain.

[0017] Preferably, in S3.2, the comparison dimension is threshold, speed, and acceleration, and the specific method is as follows: Threshold determination: Compare monitoring data with preset thresholds, which include historical extreme values, design values, and statistical model confidence intervals. Use the convex hull method to delineate historical operating condition extremes and determine whether new data exceeds the historical range. Use the 3σ method, POT model, and cloud model statistical methods to set threshold intervals. Rate discrimination: Analyze the rate of change of the measured value and determine the rate threshold based on the finite element calculation results; Acceleration discrimination: monitor the acceleration trend of measured value changes and identify nonlinear mutation signals.

[0018] Preferably, in S3.3, the specific process of grading in combination with the structural safety stage is as follows: Elastic stage: Level 1: The measured value does not exceed the historical extreme value, and the structure is safe; Level 2: The measured value exceeds the historical extreme value but does not exceed the threshold range, and there is no trend growth; Level 3: The measured value exceeds the historical extreme value and individual measuring points are close to the material strength; Elastic-plastic stage: Level 4: The measured value exceeds the material strength, and there are local cracks or irreversible deformation; Instability and damage stage: Level 5: Multiple indicators show a trend of growth and the structure is on the verge of failure.

[0019] Preferably, the S4 specifically includes the following process: S4.1. Structuring on-site inspection information: Through semantic recognition, on-site inspection records are structured to clearly define inspection time, personnel, route, scope, results, and conclusions. Information on various defects and defect changes discovered during on-site inspections is used as input for comprehensive engineering safety assessments. S4.2. Intelligent text classification and information extraction: Deep learning models are used to implement text classification and feature extraction. The details are as follows: Preprocessing: data cleaning, deduplication, and semantic segmentation; Vectorization: Character-level vectorization processing avoids reliance on specialized lexicons; Feature extraction: extract text features and reduce dimensionality through convolutional layers; Classification output: Use Softmax classifier to output inspection text classification results; S4.3. Quantitative identification of concrete cracks based on machine learning; S4.4. Synchronize and visualize abnormal information, using time and project location as indexes to save the abnormal information matrix; in the digital twin platform, visually present the abnormal location through text descriptions or pictures; S4.5 Abnormality classification and grading. Abnormality categories of hydraulic structures are classified according to damage defects, abnormal seepage defects, and abnormal deformation defects. Abnormality levels are graded based on the defect location, scale, shape, development trend, and impact on safety. The details are as follows: Level Ⅰ: Major defects that threaten structural safety or function; Level II: Major defects that affect structural safety or function; Grade III: general defect, slightly affected but may progress; Level IV: Minor defects that do not affect safety for the time being; S4.6. Abnormal handling: For abnormalities in hydraulic structures, a defect management ledger should be established in the Group's dam system to record relevant information in detail and conduct information management. The abnormality management ledger should include the location, method, time, and nature of the defect discovery, as well as the category, type, level, identification method, identification time, defect control plan and program, implementation and acceptance status information. The specific abnormality handling measures are as follows: Focus on abnormal developments and changes, conduct monitoring and testing, or take effective measures to ensure the safe operation of hydraulic structures; A defect management plan should be formulated for Level I, Level II, and Level III anomalies; Special monitoring plans should be formulated for Level I and Level II anomalies to strengthen monitoring, supervision, analysis and judgment; The treatment plan for Level I anomalies should undergo special design, special review, safety assessment, special construction and special acceptance; Safety monitoring and inspection work should be strengthened during the abnormal management process; After the abnormal control passes the acceptance, the acceptance report and relevant materials should be compiled and archived in a timely manner, and submitted to the dam management unit and the group dam safety center in accordance with regulations, and reported to the government and relevant industry regulatory agencies in accordance with regulations.

[0020] Preferably, the S4.3 specifically includes the following process: Grayscale model establishment: convert RGB images into grayscale images to enhance contrast; Image filtering: Use median filter to remove noise and retain edge details; Grayscale enhancement and binarization: Improve the distinction between the crack area and the background through grayscale stretching and compression to generate a binary image; Output quantitative crack length, width, and distribution parameters to support defect level determination.

[0021] Preferably, the S5 specifically includes the following process: S5.1. Model Construction and Parameter Inversion: Based on the BIM model and engineering data, a multi-scale finite element mesh model is established to accurately reflect the dam's geometric structure, material zoning, and underlying geological characteristics. A hexahedral mesh is primarily used, with local densification to improve accuracy, supporting multi-physics field coupling analysis of temperature, seepage, deformation, and stress fields. Utilizing historical monitoring data, a neural network and genetic algorithm are used to rapidly invert the elastic modulus of the dam concrete and the variable modulus parameters of the bedrock. S5.2. Conduct dam stability, stress, and seepage analysis primarily using the finite element method, combined with structural mechanics and material mechanics methods. The finite element method supports full-process simulation, linear elasticity, contact nonlinearity, material nonlinearity analysis, and "water-heat-mechanics" multi-physics coupling. Structural mechanics methods are used for anti-sliding stability verification. Material mechanics methods are used to calculate edge and internal stresses on the dam body. S5.3. Initial dam state construction: Based on the inversion parameters and real environmental conditions, an initial physical field consistent with the measured data is constructed to support subsequent twin synchronous simulation, structural safety review, and special working condition rehearsal; S5.4. Algorithm packaging and deployment: This includes integrated automated pre- and post-processing, integrated boundary condition prediction models, integrated efficient parallel solvers, and integrated dedicated post-processing engines. Automated pre- and post-processing can automatically extract monitoring data, generate finite element input files, and automatically extract results after calculations are complete. The boundary condition prediction model uses historical data regression analysis to predict future reservoir water levels and temperatures, supporting pre-rehearsal scenarios. The parallel solver supports single-node multi-threading / multi-node multi-process parallel computing to improve efficiency. The post-processing engine can display the temperature, deformation, stress, and seepage fields of the dam body and foundation based on finite element analysis results, for ease of use by engineering operation and management personnel. S5.5. After module deployment and system integration, rapid review of structural safety, rapid review of extreme working conditions, and rapid rehearsal of risky working conditions are achieved based on the configured models, parameters, algorithms, and software.

[0022] Preferably, the S6 specifically includes the following process: S6.1. Classify the building types and categorize the monitored objects according to the structural characteristics and functional differences of the dam; S6.2. Construction of the indicator system: For each type of building, specific evaluation indicators are formulated in combination with the safety evaluation content required by the regulations to form a type-specific safety evaluation indicator system; S6.3. Quantitative standard setting for evaluation indicators includes the following two methods: Design index method: setting index thresholds based on standard design values; Abnormal reasoning method: Use historical monitoring data and abnormal identification results to dynamically adjust indicator thresholds; S6.4. Indicator weighting and weight calculation: a weight distribution table is formed to clarify the contribution ratio of each indicator to the overall safety level. The weighting method is as follows: Judgment matrix method: construct a judgment matrix through expert scoring and calculate the relative weight of each indicator; Expert experience empowerment: weights are directly assigned based on project importance and failure risk factors; S6.5. Calculation and grading of overall safety: Multiply the quantitative results of each indicator by its weight and add them up to obtain the overall safety score of the project. The formula is as follows: ; in: For overall safety, is the indicator weight, Score the indicator.

[0023] Beneficial effects of the present invention: 1. This invention proposes a full-process intelligent monitoring method, establishing a closed-loop management system from data collection, validity determination, anomaly identification, on-site inspection, structural safety review to comprehensive safety assessment, thus achieving full-chain intelligent dam safety management. By integrating multi-source data (monitoring data, inspection records, simulation results), it improves analysis accuracy and decision-making efficiency. 2. This invention proposes an efficient data management and anomaly identification method, relying on an automated monitoring system to acquire data in real time, saving labor costs and improving collection efficiency. It uses logical discrimination, statistical discrimination, and cluster analysis to intelligently eliminate invalid data to ensure data reliability. It also combines three-dimensional dynamic monitoring of thresholds (historical extreme values, design values), rates (annual variation, trend growth), and acceleration (nonlinear mutation signals), with graded alarms (elastic, elastoplastic, and instability and destruction stages). 3. This invention proposes a precise structural safety analysis method, constructing a finite element mesh based on the BIM model, supporting multi-physics field coupling analysis of temperature field, seepage field, deformation field, and stress field, and truly reflecting the operational performance of the dam; 4. Use neural networks and genetic algorithms to invert dam material parameters (elastic modulus, bedrock modulus) to improve the matching between the model and actual working conditions; simulate risk scenarios such as floods and earthquakes by predicting boundary conditions (reservoir water level and temperature) to quickly verify structural safety; 5. This invention proposes an intelligent on-site inspection and defect management method, using deep learning models (LSTM, CNN) to achieve inspection text classification and structured extraction of defect information; using image processing technology (grayscale enhancement, median filtering) to quantitatively identify concrete crack parameters (length, width, distribution); formulating special treatment plans based on defect severity (Level I to Level IV), establishing a defect ledger, and dynamically tracking repair results; 6. This invention proposes a dynamic safety assessment method for dams. Evaluation indicators (such as stress threshold and seepage pressure reduction coefficient) are customized according to the dam structure type (gravity dam, arch dam, etc.). The contribution of each indicator to safety is quantified through the AHP method or expert experience weighting (e.g., stability 40%, seepage 30%). The overall safety factor is calculated using a weighted summation method and classified into five levels, from A (safe) to E (dangerous), with clear status and treatment measures. 7. This invention proposes an intelligent decision-making and early warning method, combining the confidence interval method, limit state method, and small probability method to dynamically generate monitoring indicators and trigger graded alarms (early warning, emergency response); relying on knowledge graphs and expert systems, it analyzes the causes of abnormalities (material aging, foundation defects) and pushes matching engineering plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Schematic diagram of the process of determining the validity of monitoring data in S2 of the present invention; Figure 2 Schematic diagram of the process of detecting abnormalities in monitoring data in S3 of the present invention; Figure 3 Schematic diagram of the process of structural safety review in S5 of the present invention; Figure 4 This is a flow chart of the comprehensive evaluation of the dam safety status in S6 of the present invention. DETAILED DESCRIPTION

[0025] The present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.

[0026] Example 1: An intelligent online monitoring method for gravity dams based on multi-source data fusion, including acquisition of monitoring data, determination of monitoring data validity, identification of monitoring data anomalies, identification of on-site inspection information anomalies, structural safety review, and comprehensive evaluation of dam safety status.

[0027] Acquisition of the monitoring data, i.e., acquiring the real-time and historical monitoring data of all monitoring points from the database according to business needs; The validity determination of the monitoring data is to use a mathematical model to check the validity of the monitoring data and eliminate invalid measurements; The monitoring data anomaly identification is to formulate monitoring indicators for each monitoring point of each monitoring project from the dimensions of threshold, rate, acceleration, etc. based on design indicators, historical extreme values, limit states, small probability methods, etc., and use the monitoring indicators to identify anomalies in the monitoring data; The on-site inspection information anomaly recognition, through semantic recognition, structures the inspection records, and various defect information and defect change information found during on-site inspections are used as input conditions for comprehensive dam safety assessment; The structural safety review, when the measured operational performance is judged to be abnormal, takes the abnormal situation as the input condition and conducts a safety review of the corresponding building; when the actual load and its combination, mechanical parameters, calculation model, etc. are significantly different from the design, a review is conducted based on the actual operational situation; The comprehensive assessment of the dam safety status is carried out by establishing a dam safety evaluation rule system, integrating the results of the measured operational performance analysis, structural safety review results and on-site inspection results, conducting a comprehensive quantitative assessment of the dam safety status, determining the dam safety level, and issuing early warnings based on the assessment results.

[0028] As a further solution of the present invention, the steps for determining the validity of the monitoring data are as follows: Step 1: Model construction and adaptation. The model includes the following methods: 1. Logical discrimination method: Eliminate over-limit data based on the range of the monitoring instrument, monitoring accuracy, the physical meaning of the monitoring data and the preset threshold; 2. Statistical discrimination method: Use historical data to establish a multiple regression model, define a confidence interval of 6 times the standard deviation, and judge it as invalid if it exceeds the interval; 3. Cluster analysis method: Use the cluster analysis model to determine outlier data as gross errors; Intelligently match the optimal discrimination model based on the measurement point type and data time series characteristics.

[0029] Step 2: Data initial screening: Daily new data is initially judged by the model constructed in step 1. Those that pass the initial judgment are identified as valid values ​​and enter the monitoring data anomaly identification stage. Invalid values ​​enter step 3; Step 3: Data retesting: The invalid values ​​determined in step 2 are first retested by the station. After the retesting is completed, the model is used for secondary discrimination. If the retested data is a valid value, the gross error record of the measuring point is saved and the process ends. If it is still determined to be an invalid value, proceed to step 4. Step 4: Risk classification and disposal. For important measuring points (the division between important and ordinary measuring points is described in the monitoring data anomaly identification section), directives are issued directly to the station to remind staff to conduct on-site inspections. For general measuring points, multiple physical quantity and multi-measurement evaluations are conducted, and data from nearby measuring points of the same type and different physical quantities in the same location are correlated. If all data are qualified, the point is marked as "low risk" and included in the list of key measurement points. The monitoring frequency is increased, and the process proceeds to step 5. Otherwise, it is considered to be a high risk, and directives are issued directly to the station to remind staff to conduct on-site inspections. Step 5: Monitor the measurement points listed in the key focus measurement point list in Step 4 for five consecutive days and then use the model to evaluate them. If they return to normal, the process ends. If they still fail, a command is issued to the station to remind staff to conduct an on-site inspection. If there is an instrument failure, it is repaired as soon as possible. If the measured value is indeed the true value, the monitoring data anomaly identification module is entered, and an inspection command is issued to the station to proceed to Step 6. Step 6: On-site inspection and defect handling. If defects such as cracks and water seepage are found after the inspection, a repair plan will be provided and included in the defect management. At the same time, the cusp gross error data will be retrieved. If the on-site inspection results are normal, the area will be added as an inspection area and included in the inspection management module. Step 7: Manual review and model optimization. Regularly review the identification results manually and verify the reliability of the model based on the review results; mark the failed models and re-match the algorithm.

[0030] As a further solution of the present invention, the steps for implementing the monitoring data anomaly identification are as follows: Step 1: Determine the monitoring objects and monitoring indicators. Select monitoring indicators based on dam failure modes. Combined with existing dam risk assessment results, establish a risk transmission chain and implement the selection of monitoring indicator projects for risk control. Select monitoring points based on the weak parts of the dam. Combined with the actual project, simulate and analyze the stress evolution process of the dam under different working conditions and analyze the failure transmission path. Establish a perception time chain for the perceptible effect quantity and determine the monitoring content of the monitoring project. Select monitoring points based on dam monitoring data and first evaluate the monitoring data of the selected monitoring indicators. Step 2: Compare monitoring indicators and identify anomalies. The comparison dimensions are mainly thresholds, rates, and accelerations. The specific methods are as follows: 1. Threshold determination: Compare monitoring data with preset thresholds (historical extreme values, design values, and statistical model confidence intervals); use the convex hull method to delineate historical operating condition extremes and determine whether new data exceeds the historical range; use statistical methods such as the 3σ method, POT model, and cloud model to set threshold intervals; 2. Rate discrimination: Analyze the rate of change of measured values ​​(such as annual variation, trend growth rate), and determine the rate threshold based on the finite element calculation results; 3. Acceleration discrimination: monitor the acceleration trend of measured value changes (such as deformation and secondary derivative analysis of seepage pressure) and identify nonlinear mutation signals.

[0031] Step 3: Abnormal classification: According to the degree of deviation of the measured value from the monitoring index, the classification is carried out in combination with the structural safety stage (elasticity, elastoplasticity, instability and failure): 1. Elasticity stage: Level 1: The measured value does not exceed the historical extreme value, and the structure is safe. Level 2: The measured value exceeds the historical extreme value but does not exceed the threshold range, and there is no trend of growth. Level 3: The measured value exceeds the historical extreme value and some measuring points are close to the material strength.

[0032] 2. Elastic-plastic stage (Level 4): The measured value exceeds the material strength (tensile stress / compressive stress exceeds the limit), and local cracks or irreversible deformation exist.

[0033] 3. Instability and damage stage (Level 5): Multiple indicators show a trend of growth (such as through-going cracks and a sudden increase in leakage), and the structure is on the verge of failure.

[0034] Step 4: Fusion and correlation analysis of multiple measurement points: combining multi-point data (such as deformation, seepage pressure, and stress) to analyze the overall state of the structure and identify potential risk transmission paths (such as curtain failure → increased seepage pressure → reduced anti-sliding stability); establishing a monitoring data time chain and analyzing the timeliness response (such as the lag effect of dam body stress after water level changes).

[0035] Step 5: Diagnosis and feedback optimization, combining finite element analysis, inspection reports, and hidden danger records to locate the causes of abnormalities (such as material aging and foundation defects); regularly update monitoring indicator thresholds based on the latest monitoring data and risk assessment results; and feed back abnormality identification results to the "four predictions" (forecast, warning, rehearsal, and plan) system to improve the risk management chain.

[0036] As a further solution of the present invention, the steps for implementing the abnormality identification of on-site inspection information are as follows: Step 1: Structuring on-site inspection information: Through semantic recognition, on-site inspection records are structured to clearly define inspection time, personnel, route, scope, results, conclusions, etc. Various defect information and defect change information found during on-site inspections are used as input conditions for comprehensive project safety evaluation. Step 2: Intelligent text classification and information extraction. Deep learning models (LSTM, convolutional neural network) are used to implement text classification and feature extraction. The details are as follows: 1. Preprocessing: data cleaning, deduplication, and semantic segmentation; 2. Vectorization: Character-level vectorization processing to avoid reliance on specialized lexicons; 3. Feature extraction: extract text features and reduce dimensionality through convolutional layers; 4. Classification output: Use the Softmax classifier to output the inspection text classification results (such as defect category and level).

[0037] Step 3: Quantitative identification of concrete cracks based on machine learning. The specific method is as follows: 1. Grayscale model establishment: convert RGB images into grayscale images to enhance contrast; 2. Image filtering: Use a median filter to remove noise (such as pulse and salt and pepper noise) while retaining edge details; 3. Grayscale enhancement and binarization: Enhance the distinction between the crack area and the background through grayscale stretching and compression to generate a binary image; Output quantitative crack length, width, distribution and other parameters to support defect level determination; Step 4: Synchronize and visualize abnormal information. Use time and project location as indexes to save the abnormal information matrix (including time, content, and trend evaluation). In the digital twin platform, visually present the abnormal location (such as crack location, leakage point) through text descriptions or pictures. Step 5: Abnormality classification and grading. Hydraulic structure abnormalities (defects) are classified into categories such as damage defects, abnormal seepage defects, and abnormal deformation defects. Abnormality (defect) levels are graded based on defect location, scale, morphology, development trend, and impact on safety, as follows: Level Ⅰ: Major defects that threaten structural safety or function; Level II: Major defects that affect structural safety or function; Grade III: general defect, slightly affected but may progress; Level IV: Minor defects that do not affect safety for the time being.

[0038] Step 6: Abnormal handling. Abnormalities (defects) of hydraulic structures should be recorded in a defect management ledger in the group's dam system, with detailed information recorded for information management. The abnormality (defect) management ledger should include the location, method, time, and nature of the defect, as well as the category, type, level, identification method, identification time, defect management plan and program, implementation and acceptance status, etc. The specific measures for handling abnormalities (defects) are as follows: 1. Pay close attention to the development and changes of abnormalities (defects), and conduct monitoring and testing when necessary, or take effective measures to ensure the safe operation of hydraulic structures.

[0039] 2. A defect management plan should be formulated for Level I, Level II, and Level III anomalies (defects).

[0040] 3. Special monitoring plans should be formulated for Level I and Level II anomalies (defects) to strengthen monitoring, supervision, analysis and judgment.

[0041] 4. The treatment plan for Level I anomalies (defects) should undergo special design, special review, safety assessment, special construction and special acceptance.

[0042] 5. Safety monitoring and patrol inspections should be strengthened during the abnormality (defect) management process.

[0043] 6. After the abnormality (defect) treatment passes the acceptance, the acceptance report and relevant materials should be compiled and archived in a timely manner, and submitted to the dam management unit and the group dam safety center in accordance with regulations, and reported to the government and industry and other relevant regulatory agencies in accordance with regulations.

[0044] As a further solution of the present invention, the steps for implementing the structural safety review are as follows: Step 1: Model construction and parameter inversion. Based on the BIM model and engineering data, a multi-scale finite element mesh model is established to accurately reflect the dam's geometric structure, material partitioning, basic geological characteristics, etc. The hexahedral mesh is mainly used, with local encryption to improve accuracy, to support multi-physics field coupling analysis of temperature, seepage, deformation, and stress fields. Utilizing historical monitoring data, neural networks and genetic algorithms are used to quickly invert parameters such as the elastic modulus of the dam concrete and the variable modulus of the bedrock. Step 2: Algorithm: dam stability, stress, and seepage analysis is conducted primarily using the finite element method, combined with structural mechanics and material mechanics methods. The finite element method supports full-process simulation, linear elasticity, contact nonlinearity, material nonlinearity analysis, and "water-heat-mechanics" multi-physics coupling. Structural mechanics methods (rigid body limit equilibrium method) are used for anti-sliding stability verification. Material mechanics methods primarily calculate edge and internal stresses on the dam body. Step 3: Construct the initial state of the dam. Based on the inversion parameters and real environmental conditions, an initial physical field consistent with the measured data is constructed to support subsequent twin synchronous simulation, structural safety review, and special working condition rehearsal. Step 4: Algorithm packaging and deployment, mainly including the integration of automated pre- and post-processing, integrated boundary condition prediction models, integrated efficient parallel solvers, and integrated dedicated post-processing engines; automated pre- and post-processing can automatically extract monitoring data (reservoir water level, air temperature, etc.), generate finite element input files, and automatically extract results (temperature / stress cloud maps, time history curves) after the calculation is completed; the boundary condition prediction model can predict future reservoir water level, air temperature, etc. based on historical data regression analysis to support pre-rehearsal scenarios; the parallel solver supports single-node multi-threading / multi-node multi-process parallel computing to improve efficiency; the post-processing engine can display the temperature field, deformation field, stress field, and seepage field of the dam body and foundation based on the finite element analysis results, for the convenience of engineering operation and management personnel; After completing module deployment and system integration, the configured models, parameters, algorithms and software can be used to quickly review structural safety, quickly review extreme working conditions and quickly rehearse risky working conditions.

[0045] As a further solution of the present invention, the steps for implementing the comprehensive dam safety assessment are as follows: Step 1: Classify the building types and classify the monitored objects according to the structural characteristics and functional differences of the dam (such as gravity dam, arch dam, earth-rock dam, etc.); Step 2: Construct an indicator system. For each type of building, specific evaluation indicators (such as stress threshold, seepage pressure reduction coefficient, deformation rate, etc.) are formulated in combination with the safety evaluation content required by the regulations (such as stability, seepage control, deformation monitoring, etc.), forming a type-based safety evaluation indicator system. Step 3: Setting the quantitative standards for evaluation indicators. There are two main methods: 1. Design index method: Set index thresholds based on standard design values ​​(such as anti-sliding stability safety factor and material strength limit).

[0046] 2. Abnormal reasoning method: Use historical monitoring data and abnormal identification results (such as deformation exceeding the limit and trend growth of seepage pressure) to dynamically adjust the indicator threshold.

[0047] Step 4: Assign indicators weights and calculate weights to form a weight distribution table to clarify the contribution ratio of each indicator to the overall safety level. The weighting method is as follows: 1. Judgment Matrix Method (AHP): Construct a judgment matrix through expert scoring and calculate the relative weight of each indicator (e.g. stability accounts for 40%, seepage accounts for 30%, and deformation accounts for 30%).

[0048] 2. Expert experience weighting: weights are directly assigned based on factors such as project importance and failure risk.

[0049] Step 5: Calculate and grade the overall safety level. Multiply the quantitative results of each indicator (such as the standardized score) by its weight and add them up to obtain the overall safety level of the project. The formula is as follows: ; in: For overall safety, is the indicator weight, Score the indicator.

[0050] The dam safety status evaluation level can be divided according to the overall safety score, as shown below:

[0051] Example 2: An intelligent online monitoring method for gravity dams based on multi-source data fusion includes the acquisition of monitoring data, determination of monitoring data validity, identification of monitoring data anomalies, identification of on-site inspection information anomalies, structural safety review, and comprehensive evaluation of dam safety status.

[0052] Acquisition of monitoring data: obtaining real-time and historical monitoring data of all monitoring points from the database according to business needs; Specifically, during the dam construction and operation phases, internal and external monitoring instruments were deployed within the dam body and bedrock to provide real-time monitoring of environmental variables, dam deformation, seepage and pressure, stress and strain, and temperature within the dam site. An automated data collection system was also established. The data collection system cleans the collected monitoring data, using logical judgment to eliminate out-of-range values ​​(the osmometer range is 0-1 MPa; any values ​​exceeding this range are marked as invalid) and data with incorrect sign. Multiple interpolation (MICE) algorithms were used to correct missing values, and matrix completion (minimizing the nuclear norm) was used for non-periodic data. Furthermore, data from multiple physical quantities were normalized using the Z-score algorithm to eliminate dimensionality effects.

[0053] In acquiring monitoring data, internal and external monitoring instruments deployed on the dam body and bedrock include: radar water level gauges for measuring upstream and downstream water levels; differential resistance thermometers for measuring reservoir water temperature, air temperature, dam body temperature, and bedrock temperature; vacuum lasers, tensioning lines, precision levels, and multi-point displacement meters for measuring horizontal and vertical deformation of the dam body and displacement of bedrock; piezometers and water measuring weirs for measuring dam body seepage pressure, dam foundation uplift pressure, and dam foundation leakage; and strain gauge groups, steel bar gauges, and steel plate gauges for measuring dam body stress and strain.

[0054] Reference Figure 1 ,The validity judgment of monitoring data is to apply mathematical models to check the ,validity of monitoring data and eliminate invalid measurements; Specifically, the validity of the collected monitoring data is judged, including gross error identification, that is, the LOF algorithm is used for periodic data, K-medoids clustering is used for trend data, and an integrated learning model (such as random forest + gradient boosting) is used for non-periodic data; and a multivariate regression model is constructed based on historical data. Measurements exceeding 6 times the standard deviation are judged to be invalid.

[0055] Reference Figure 2 ,monitoring data anomaly identification, that is, based on design indicators, historical extreme values, limit states, small probability methods, etc., the monitoring indicators of each monitoring point of each monitoring project are formulated from the dimensions of threshold, rate, acceleration, etc., and the monitoring indicators are used to identify anomalies in the monitoring data; Specifically, there are several methods for anomaly identification: threshold discrimination, comparing the monitoring data with the preset threshold (historical extreme value, design value, statistical model confidence interval); using the convex hull method to define the extreme value of historical working conditions and determine whether the new data exceeds the historical range; using statistical methods such as the 3σ method, POT model, and cloud model to set the threshold interval. When the historical extreme value of dam deformation is 10mm and the new measured value is 12mm, a level 2 alarm is triggered (breaking the historical extreme value but not exceeding the design threshold); rate discrimination: analyzing the rate of change of measured values ​​(such as annual amplitude, trend growth rate), and combining the finite element calculation results to determine the rate threshold. The annual amplitude threshold is set to 2mm / year. If the measured value growth rate reaches 3mm / year, the elastic-plastic stage alarm is triggered; acceleration discrimination: monitoring the acceleration trend of measured value changes (such as deformation, secondary derivative analysis of seepage pressure), identifying nonlinear mutation signals, the secondary derivative analysis of seepage pressure shows nonlinear mutation, and combined with the finite element results to determine the risk of local cracks.

[0056] On-site inspection information anomaly recognition uses semantic recognition to structure inspection records. Various defect information and defect change information found during on-site inspections are used as input conditions for comprehensive dam safety assessment. Specifically, intelligent text information classification and structured extraction, after data preprocessing, using the BiLSTM+Attention model for classification, annotated 2,000 historical inspection records (including four types of defects: cracks, leakage, deformation, and others), divided into a training set (80%), a validation set (10%), and a test set (10%). Focal Loss was introduced to address the class imbalance problem (crack samples accounted for 60%). The test set accuracy was ≥92%, and the recall rate was ≥88%. Finally, structured fields were output and stored in a MySQL database. For quantitative recognition of concrete crack images, RGB images were converted to grayscale images, and adaptive median filtering (window size 5×5) was used to remove noise (such as reflective spots). CLAHE (Contrast-Limited Adaptive Histogram Equalization) was used to improve the distinction between cracks and background. The Canny algorithm was used to extract crack contours, with threshold settings (low = 50, high = 150), and morphological closing operations were used for filling. For small fractures, LBP (local binary pattern recognition) is used to determine the age of cracks (texture roughness of old cracks > 0.15). The ResNet-50 pre-trained model is fine-tuned to distinguish crack types (shrinkage cracks, temperature cracks, and structural cracks), with a test set accuracy of ≥ 89%. Anomaly detection: YOLOv5 is used to identify non-crack anomalies (such as spalling and calcium precipitation) in images in real time, with a confidence threshold of 0.7. Anomaly information is synchronized and visualized, mapped on the digital twin platform, and multi-source data fusion is performed. Clarify anomaly classification standards to achieve coordinated treatment and closed-loop management. The specific classification standards are as follows: Level I defect: penetrating cracks (width > 2mm), leakage > 10L / s; Level II defects: surface cracks (width 1-2mm), local water seepage; Level III defect: peeling area > 0.5㎡, calcium precipitation area > 1㎡; Grade IV defects: slight honeycombing, vegetative growth; Level I defects are pushed to the operation and maintenance team through the enterprise WeChat API, requiring on-site review within 24 hours. Similar cases (such as "gate pier crack grouting plan") are retrieved from the knowledge base, and treatment suggestions are generated. After treatment, repair pictures are uploaded. The system confirms the closed loop by comparing the images before and after repair (SSIM similarity > 95%).

[0057] Reference Figure 3 , structural safety review: when the measured operational performance is judged to be abnormal, the abnormal situation will be used as input conditions to conduct a safety review of the corresponding building; when the actual load and its combination, mechanical parameters, calculation model, etc. are significantly different from the design, a review will be conducted based on the actual operational situation; Specifically, a hexahedral mesh of the gravity dam was generated based on BIM, and locally refined to an accuracy of 0.5m. A BP neural network was trained based on monitoring data to invert the elastic modulus of the dam body and the deformation modulus of the bedrock. The objective function was to minimize the error between the measured deformation and the simulated deformation. Convergence was achieved after 500 iterations, and the error was controlled within ±5%. Finite element simulation calculations were performed to simulate the sudden rise in reservoir water level, coupling the seepage field and stress field, and outputting a dam heel tensile stress cloud map to determine whether it exceeded the limit. Contact nonlinear analysis was performed on abnormal locations (such as the Erjiang spillway piers) to calculate the crack propagation path. The sliding surface stress calculated by finite element was input through the rigid body limit equilibrium method to verify the anti-sliding safety factor. Reference Figure 4 , comprehensively evaluate the safety status of the dam, establish a dam safety evaluation rule system, integrate the results of the measured operation performance analysis and structural safety review results, carry out a comprehensive quantitative evaluation of the dam safety status, determine the dam safety level, and issue early warnings based on the evaluation results.

[0058] Specifically, the evaluation indicators include the first-level indicators: deformation, seepage, stress and strain, and on-site inspection; the second-level indicators: single-point exceedance rate, multi-point correlation, and structural safety factor; the AHP method is used to determine the weights of deformation (0.35), seepage (0.25), stress and strain (0.25), and on-site inspection (0.15), and the overall safety degree is calculated. The dam safety level is divided into the following categories based on the overall safety degree: Grade A (safety): score range [8, 10], the dam is in good operating condition; Grade B (relatively safe): score range [6, 8], daily management needs to be strengthened; Level C (general safety): score range [4, 6], reinforcement measures are required; Grade D (more dangerous): score range [2, 4], requiring on-site inspection by an expert group; Level E (dangerous): score range [0, 2), there is a risk of dam collapse and emergency risk elimination is required.

[0059] In an application on a certain ultra-high gravity dam, this method increased the accuracy of anomaly identification to 92%, shortened the structural review time from the traditional 72 hours to 8 hours, and increased the defect handling response efficiency by 40%, verifying the practicality and advancement of the invention.

[0060] The above embodiments are merely preferred technical solutions of the present invention and should not be construed as limiting the present invention. The scope of protection of the present invention shall be the technical solutions set forth in the claims, including equivalent alternatives to the technical features of the technical solutions set forth in the claims. In other words, equivalent alternatives and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. An intelligent online monitoring method for concrete dams based on multi-source data fusion, characterized by: The following steps are involved: S1. Acquisition of monitoring data: Acquire real-time and historical monitoring data of all monitoring points from the database according to business needs; S2. Validity determination of monitoring data: Apply mathematical models to check the validity of monitoring data and eliminate invalid measurements; S3. Monitoring data anomaly identification: Based on design indicators, historical extreme values, limit states, and small probability methods, formulate monitoring indicators for each monitoring point in each monitoring project from the dimensions of threshold, rate, and acceleration, and use monitoring indicators to identify anomalies in monitoring data; S4. Identification of anomalies in on-site inspection information: Through semantic recognition, inspection records are structured, and various defect information and defect change information found during on-site inspections are used as input conditions for comprehensive dam safety assessment; S5. Structural safety review: When the measured operational performance is judged to be abnormal, the abnormal situation is used as an input condition to conduct a safety review of the corresponding building; when the actual load and its combination, mechanical parameters, calculation model are significantly different from the design, a review is conducted based on the actual operational situation; S6. Comprehensive evaluation of dam safety status: By establishing a dam safety evaluation rule system, integrating the results of measured operational performance analysis, structural safety review results and on-site inspection results, a comprehensive quantitative evaluation of the dam safety status is carried out, the dam safety level is determined, and early warnings are issued based on the evaluation results.

2. The intelligent online monitoring method for concrete dams based on multi-source data fusion according to claim 1 is characterized by: The step S2 specifically includes the following process: S2.1, model construction and adaptation, intelligently matching the optimal discrimination model according to the measurement point type and data time series characteristics; S2.2, Data Preliminary Screening: Daily new data is preliminarily judged by the model constructed in S2.

1. Values ​​that pass the preliminary judgment are identified as valid and enter the monitoring data anomaly identification stage. Invalid values ​​enter S2.3; S2.3, Data Retest: The invalid values ​​determined in S2.2 are first retested by the station. After the retest is completed, the model is used for secondary discrimination. If the retested data is a valid value, the gross error record of the measuring point is saved and the process ends. If it is still determined to be an invalid value, proceed to S2.

4. S2.4, Risk Classification and Disposition: For important measuring points, direct instructions are issued to the station to remind staff to conduct on-site inspections. For general measuring points, multiple physical quantity and multi-measurement evaluations are conducted, and data from nearby measuring points of the same type and different physical quantities in the same location are correlated. If all data are qualified, the point is marked as "low risk" and added to the list of key measurement points, with increased monitoring frequency, and proceeding to S2.

5. Otherwise, it is considered a high risk and direct instructions are issued to the station to remind staff to conduct on-site inspections. S2.5: Monitor the measurement points listed in S2.4 for five days and then use the model to evaluate them. If they return to normal, the process ends. If they still fail, send a command to the station to remind staff to conduct an on-site inspection. If there is an instrument failure, repair it as soon as possible. If the measured value is indeed the true value, enter the monitoring data anomaly identification module and send an inspection command to the station, and then enter S2.

6. S2.

6. On-site inspection and defect handling: If cracks or water seepage defects are found after the inspection, a repair plan will be provided and included in the defect management, and the cusp gross error data will be retrieved. If the on-site inspection results are normal, the area will be added as an inspection area and included in the inspection management module. S2.

7. Manual review and model optimization: Regularly conduct manual review of the identification results and verify the reliability of the model based on the review results; mark failed models and re-match the algorithm.

3. The intelligent online monitoring method for concrete dams based on multi-source data fusion according to claim 1 is characterized by: In S2.1, model construction includes the following methods: Logical discrimination method: Eliminate over-limit data based on the monitoring instrument range, monitoring accuracy, physical meaning of monitoring data and preset thresholds; Statistical discrimination method: Use historical data to establish a multiple regression model, define a confidence interval of 6 times the standard deviation, and judge it as invalid if it exceeds the interval; Cluster analysis method: using cluster analysis model to identify outlier data as gross errors; Adaptation means intelligently matching the optimal discrimination model based on the measurement point type and data timing characteristics.

4. The intelligent online monitoring method for concrete dams based on multi-source data fusion according to claim 1 is characterized in that: The S3 specifically The following processes are included: S3.

1. Determine monitoring targets and indicators: Select monitoring indicators based on dam failure modes, combine existing dam risk assessment results, establish a risk transmission chain, and implement the selection of monitoring indicator items for risk control; Based on the selected monitoring points at the weak parts of the dam and in combination with the actual project, the stress evolution process of the dam under different working conditions is simulated and analyzed, and the failure transmission path is analyzed; Establish a perceptual time chain for the perceptible effect size and determine the monitoring content of the monitoring project; When selecting monitoring points based on dam monitoring data, the selected monitoring indicators should be evaluated first; S3.2, monitoring indicator comparison and abnormality identification; S3.3, abnormal classification; According to the degree of deviation of the measured value from the monitoring index, the structure safety stage is classified; S3.

4. Multi-point fusion and correlation analysis: Combine multi-point data to analyze the overall state of the structure and identify potential risk transmission paths; establish a monitoring data time chain and analyze timeliness responses; S3.

5. Diagnosis and feedback optimization: Combine finite element analysis, inspection reports, and hidden danger records to locate the cause of anomalies; regularly update monitoring indicator thresholds based on the latest monitoring data and risk assessment results; and feed back anomaly identification results to the "four prevention" system to improve the risk management chain.

5. The intelligent online monitoring method for concrete dams based on multi-source data fusion according to claim 4 is characterized in that: In S3.2, the comparison dimensions are threshold, rate, and acceleration. The specific method is as follows: Threshold determination: Compare monitoring data with preset thresholds, which include historical extreme values, design values, and statistical model confidence intervals. Use the convex hull method to delineate historical operating condition extremes and determine whether new data exceeds the historical range. Use the 3σ method, POT model, and cloud model statistical methods to set threshold intervals. Rate discrimination: Analyze the rate of change of the measured value and determine the rate threshold based on the finite element calculation results; Acceleration discrimination: monitor the acceleration trend of measured value changes and identify nonlinear mutation signals.

6. The intelligent online monitoring method for concrete dams based on multi-source data fusion according to claim 4 is characterized by: In S3.3, the specific process of grading in combination with the structural safety stage is as follows: Elastic stage: Level 1: The measured value does not exceed the historical extreme value, and the structure is safe; Level 2: The measured value exceeds the historical extreme value but does not exceed the threshold range, and there is no trend growth; Level 3: The measured value exceeds the historical extreme value and individual measuring points are close to the material strength; Elastic-plastic stage: Level 4: The measured value exceeds the material strength, and there are local cracks or irreversible deformation; Instability and damage stage: Level 5: Multiple indicators show a trend of growth and the structure is on the verge of failure.

7. The intelligent online monitoring method for concrete dams based on multi-source data fusion according to claim 1 is characterized by: The S4 specifically includes the following processes: S4.

1. Structuring on-site inspection information: Through semantic recognition, on-site inspection records are structured to clearly define inspection time, personnel, route, scope, results, and conclusions. Information on various defects and defect changes discovered during on-site inspections is used as input for comprehensive engineering safety assessments. S4.

2. Intelligent text classification and information extraction: Deep learning models are used to implement text classification and feature extraction. The details are as follows: Preprocessing: data cleaning, deduplication, and semantic segmentation; Vectorization: Character-level vectorization processing avoids reliance on specialized lexicons; Feature extraction: extract text features and reduce dimensionality through convolutional layers; Classification output: Use Softmax classifier to output inspection text classification results; S4.

3. Quantitative identification of concrete cracks based on machine learning; S4.

4. Synchronize and visualize abnormal information, using time and project location as indexes to save the abnormal information matrix; in the digital twin platform, visually present the abnormal location through text descriptions or pictures; S4.5 Abnormality classification and grading. Abnormality categories of hydraulic structures are classified according to damage defects, abnormal seepage defects, and abnormal deformation defects. Abnormality levels are graded based on the defect location, scale, shape, development trend, and impact on safety. The details are as follows: Level Ⅰ: Major defects that threaten structural safety or function; Level II: Major defects that affect structural safety or function; Grade III: general defect, slightly affected but may progress; Level IV: Minor defects that do not affect safety for the time being; S4.

6. Abnormal handling: For abnormalities in hydraulic structures, a defect management ledger should be established in the Group's dam system to record relevant information in detail and conduct information management. The abnormality management ledger should include the location, method, time, and nature of the defect discovery, as well as the category, type, level, identification method, identification time, defect control plan and program, implementation and acceptance status information. The specific abnormality handling measures are as follows: Focus on abnormal developments and changes, conduct monitoring and testing, or take effective measures to ensure the safe operation of hydraulic structures; A defect management plan should be formulated for Level I, Level II, and Level III anomalies; Special monitoring plans should be formulated for Level I and Level II anomalies to strengthen monitoring, supervision, analysis and judgment; The treatment plan for Level I anomalies should undergo special design, special review, safety assessment, special construction and special acceptance; Safety monitoring and inspection work should be strengthened during the abnormal management process; After the abnormal control passes the acceptance, the acceptance report and relevant materials should be compiled and archived in a timely manner, and submitted to the dam management unit and the group dam safety center in accordance with regulations, and reported to the government and relevant industry regulatory agencies in accordance with regulations.

8. The intelligent online monitoring method for concrete dams based on multi-source data fusion according to claim 7 is characterized by: The S4.3 specifically includes the following process: Grayscale model establishment: convert RGB images into grayscale images to enhance contrast; Image filtering: Use median filter to remove noise and retain edge details; Grayscale enhancement and binarization: Improve the distinction between the crack area and the background through grayscale stretching and compression to generate a binary image; Output quantitative crack length, width, and distribution parameters to support defect level determination.

9. The intelligent online monitoring method for concrete dams based on multi-source data fusion according to claim 1 is characterized by: The S5 specifically includes the following processes: S5.

1. Model Construction and Parameter Inversion: Based on the BIM model and engineering data, a multi-scale finite element mesh model is established to accurately reflect the dam's geometric structure, material zoning, and underlying geological characteristics. A hexahedral mesh is primarily used, with local densification to improve accuracy, supporting multi-physics field coupling analysis of temperature, seepage, deformation, and stress fields. Utilizing historical monitoring data, a neural network and genetic algorithm are used to rapidly invert the elastic modulus of the dam concrete and the variable modulus parameters of the bedrock. S5.

2. Conduct dam stability, stress, and seepage analysis primarily using the finite element method, combined with structural mechanics and material mechanics methods. The finite element method supports full-process simulation, linear elasticity, contact nonlinearity, material nonlinearity analysis, and "water-heat-mechanics" multi-physics coupling. Structural mechanics methods are used for anti-sliding stability verification. Material mechanics methods are used to calculate edge and internal stresses in the dam body. S5.

3. Initial dam state construction: Based on the inversion parameters and real environmental conditions, an initial physical field consistent with the measured data is constructed to support subsequent twin synchronous simulation, structural safety review, and special working condition rehearsal; S5.

4. Algorithm packaging and deployment: This includes integrated automated pre- and post-processing, integrated boundary condition prediction models, integrated efficient parallel solvers, and integrated dedicated post-processing engines. Automated pre- and post-processing can automatically extract monitoring data, generate finite element input files, and automatically extract results after calculations are complete. The boundary condition prediction model uses historical data regression analysis to predict future reservoir water levels and temperatures, supporting pre-rehearsal scenarios. The parallel solver supports single-node multi-threading / multi-node multi-process parallel computing to improve efficiency. The post-processing engine can display the temperature, deformation, stress, and seepage fields of the dam body and foundation based on finite element analysis results, for ease of use by engineering operation and management personnel. S5.

5. After module deployment and system integration, rapid review of structural safety, rapid review of extreme working conditions, and rapid rehearsal of risky working conditions are achieved based on the configured models, parameters, algorithms, and software.

10. The intelligent online monitoring method for concrete dams based on multi-source data fusion according to claim 1, characterized in that: The S6 specifically includes the following processes: S6.

1. Classify the building types and categorize the monitored objects according to the structural characteristics and functional differences of the dam; S6.

2. Construction of the indicator system: For each type of building, specific evaluation indicators are formulated in combination with the safety evaluation content required by the regulations to form a type-specific safety evaluation indicator system; S6.

3. Quantitative standard setting for evaluation indicators includes the following two methods: Design index method: setting index thresholds based on standard design values; Abnormal reasoning method: Use historical monitoring data and abnormal identification results to dynamically adjust indicator thresholds; S6.

4. Indicator weighting and weight calculation: a weight distribution table is formed to clarify the contribution ratio of each indicator to the overall safety level. The weighting method is as follows: Judgment matrix method: construct a judgment matrix through expert scoring and calculate the relative weight of each indicator; Expert experience empowerment: weights are directly assigned based on project importance and failure risk factors; S6.

5. Calculation and grading of overall safety: Multiply the quantitative results of each indicator by its weight and add them up to obtain the overall safety score of the project. The formula is as follows: ； in: For overall safety, is the indicator weight, Score the indicator.

Images