Reservoir dam safety grading diagnosis and early warning method and system

Abstract

This invention discloses a method and system for graded diagnosis and early warning of reservoir dam safety, relating to the field of water conservancy engineering safety monitoring and information technology. The method involves collecting multi-source heterogeneous data on reservoir dams; preprocessing and spatiotemporally rasterizing the multi-source heterogeneous data to construct a multi-factor feature dataset; obtaining a structural health index based on structural deformation characteristics; constructing quantitative scoring functions for each auxiliary modality data to calculate auxiliary health indices; fusing the structural health index and auxiliary health indices to obtain a comprehensive health index; and classifying the dam's safety status based on the comprehensive health index, outputting corresponding diagnostic and early warning results. This invention unifies the modeling of features such as structural deformation, environmental hydrology, periodic monitoring, visual appearance, and textual semantics; and enables state perception and quantitative analysis using non-contact methods such as visual imaging, thereby improving the accuracy, sensitivity, and robustness of safety diagnosis for small reservoir dams.

Description

Technical Field

[0001] This invention relates to the field of water conservancy project safety monitoring and information technology, and more specifically to a method and system for safety classification diagnosis and early warning of reservoir dams. Background Technology

[0002] Currently, small reservoirs are numerous and widely distributed, forming an important part of flood control, irrigation, and water supply systems. The safety and stability of dams are directly related to the safety of life and property downstream. With the development of the Internet of Things and information technology, various automated monitoring devices (such as GNSS displacement gauges and piezometers) have been gradually applied to reservoirs and dams, providing a wealth of data support for dam safety assessments.

[0003] However, existing methods for safety monitoring and early warning of small reservoir dams still have the following significant problems in practical applications.

[0004] First, the monitoring data sources are heterogeneous and scattered, lacking a comprehensive spatiotemporal fusion mechanism. Current dam safety monitoring primarily relies on high-frequency automated sensor data (such as displacement and seepage data), while periodic inspection data (such as concrete strength testing and dam density testing), manual inspection reports (text data), and monitoring videos or drone images reflecting the operational status of the dam surface and key components (visual data) often exist in offline, scattered, or unstructured forms. Existing monitoring systems typically struggle to align and correlate these qualitative or low-frequency data with high-frequency automated monitoring data within the same time window, resulting in difficulties in collaboratively utilizing multi-source information and creating data silos.

[0005] Secondly, although non-contact monitoring methods such as visual monitoring can identify and record defects on the dam surface (such as cracks and seepage marks) and the operational status of key parts (such as the accumulation of floating objects in the spillway and the opening and closing status of gates), in existing engineering applications, the above visual information is mostly presented independently in the form of images or video results. It lacks a unified quantitative mapping and spatiotemporal alignment mechanism with structural monitoring data, environmental condition data, inspection text data, and periodic testing data, making it difficult to form a comprehensive criterion that can be directly used for safety classification diagnosis in daily online monitoring.

[0006] Furthermore, although the water conservancy industry has established relatively comprehensive guidelines and grading standards for dam safety evaluation, these evaluations typically rely on periodic manual inspections or expert consultations, representing a cyclical or lagging assessment method. In routine automated monitoring systems, there is a lack of effective mathematical models to transform the operational status information described in inspection reports or the local anomalies reflected in images into quantitative features that can be used for real-time calculations. Existing online early warning systems mostly use simple comparisons between sensor values ​​and fixed thresholds for alarms, making it difficult to comprehensively consider multiple factors such as structural deformation, environmental hydrology, apparent condition, and operational conditions for joint analysis. This makes it difficult to output safety level determination results that meet management requirements and clear analyses of the causes of dam failures in real-time during routine monitoring.

[0007] Therefore, how to provide a method and system for safety classification diagnosis and early warning of reservoir dams, realize the safety diagnosis and early warning monitoring of small reservoir dams by integrating multi-source heterogeneous data, and improve the accuracy, sensitivity and robustness of safety diagnosis of small reservoir dams is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0008] In view of this, the present invention provides a method and system for safety classification diagnosis and early warning of reservoir dams. Given the common technical problems in existing small reservoir dam safety monitoring, such as heterogeneous and isolated monitoring data sources, misaligned sampling frequencies, difficulty in quantifying qualitative information such as manual inspections and visual images for calculation, and the susceptibility to false alarms and missed alarms due to reliance on single sensor indicators, this invention constructs a multi-factor dataset matrix and a normalized relational function to uniformly model features such as structural deformation, environmental hydrology, periodic monitoring, visual appearance, and textual semantics. Especially for small reservoirs or key local areas lacking automated monitoring instruments, it can utilize non-contact methods such as visual images to achieve state perception and quantitative analysis, filling the blind spots of traditional single monitoring methods, thereby improving the accuracy, sensitivity, and robustness of small reservoir dam safety diagnosis.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: a method for safety classification diagnosis and early warning of reservoir dams, comprising: Collect multi-source heterogeneous data on reservoir dams; The multi-source heterogeneous data is preprocessed, and the preprocessed data is then spatiotemporally rasterized and aligned to construct a multi-factor feature dataset. For structural deformation features in a multi-factor feature dataset, a structural health mapping function based on engineering safety thresholds is constructed to obtain a structural health index. Construct quantitative scoring functions for each auxiliary modality data and calculate the auxiliary health index; By introducing a structural dominant factor and fusing the structural health index and auxiliary health index, a comprehensive health index is obtained. The dam's safety status is classified and determined based on the comprehensive health index, and corresponding diagnostic and early warning results are output.

[0010] Preferably, the preprocessed data undergoes spatiotemporal rasterization alignment processing, including: Using dam monitoring units as spatial indexes and standard time windows as time indexes, a spatiotemporal raster matrix is ​​established, and the preprocessed data is aligned. Within the same time window, an extreme value aggregation strategy is used to extract time-period characteristics of high-frequency or continuously collected data. A long-term retention and event-driven update strategy is adopted to process low-frequency or event-triggered data; The aligned multimodal data are concatenated into vectors within the same spatiotemporal raster to construct the multi-factor feature vector for the nth time window. .

[0011] Preferably, the long-term retention and event-driven update strategy includes: using the conclusions of the previous period when there is no new data; when new data is entered, mapping it to the current window using the time tolerance nearest neighbor matching method; if the time difference exceeds the preset tolerance, introducing a confidence decay factor or marking it as missing.

[0012] Preferably, the structural health mapping function maps structural displacement characteristics. Mapped to a normalized structural health index The expression is as follows: ; In the formula, , , These are preset safety thresholds for structural displacement at levels one, two, and three; This represents the corresponding health score coefficient.

[0013] Preferably, a quantitative scoring function is constructed for each auxiliary modality data to calculate the auxiliary health index, including: Environmental condition scoring function, text semantic scoring function, visual image scoring function, and detection data scoring function are constructed respectively. Based on the environmental condition scoring function, text semantic scoring function, visual image scoring function, and detection data scoring function, an auxiliary health index is calculated using a weighted summation model.

[0014] Preferably, a comprehensive health index is obtained by introducing a structure-dominant factor to fuse the structural health index and the auxiliary health index, including: ; in, Indicates the dominant structural factor. Indicates an auxiliary health index, This indicates the structural health index.

[0015] Preferably, the dam's safety status is classified and determined based on a comprehensive health index, including: establishing a comprehensive health index. Mapping relationship with security level : ; In the formula, Thresholds are set for security levels, and .

[0016] Preferably, it also includes: when the dam's safety status is determined to be level two or below, initiating a reverse attribution mechanism to traverse and backtrack the factors leading to the comprehensive health index. The decreasing controlling factor generates a structured diagnostic report containing disease type, location, and specific numerical values.

[0017] Preferably, a reservoir dam safety classification diagnosis and early warning system includes: The data acquisition module is used to collect multi-source heterogeneous data from the reservoir dam. The dataset construction module is used to preprocess the multi-source heterogeneous data and perform spatiotemporal rasterization alignment on the preprocessed data to construct a multi-factor feature dataset. The structural health index calculation module is used to construct a structural health mapping function based on engineering safety thresholds for structural deformation features in a multi-factor feature dataset, and obtain the structural health index. The auxiliary health index calculation module is used to construct quantitative scoring functions for each auxiliary modality data and calculate the auxiliary health index. The comprehensive health index calculation module is used to incorporate the structural dominant factor and fuse the structural health index with the auxiliary health index to obtain the comprehensive health index. The result judgment module is used to classify and determine the safety status of the dam based on the comprehensive health index, and output the corresponding diagnosis and early warning results.

[0018] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method and system for safety classification diagnosis and early warning of reservoir dams, constructs a unified analysis and spatiotemporal rasterization alignment mechanism for multimodal data, establishes a dual-channel safety evaluation model that combines structural health index (quantitative) and auxiliary health index (qualitative + semi-quantitative), and realizes quantitative classification diagnosis and disease attribution early warning of the safety status of small reservoir dams through comprehensive health index and safety level mapping rules. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of a method for safety classification diagnosis and early warning of reservoir dams provided by the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] This invention discloses a method for safety classification diagnosis and early warning of reservoir dams, such as... Figure 1 As shown, it includes: Step 1: Collect multi-source heterogeneous data of the reservoir dam; Step 2: Preprocess the multi-source heterogeneous data and perform spatiotemporal rasterization alignment on the preprocessed data to construct a multi-factor feature dataset; Step 3: For the structural deformation features in the multi-factor feature dataset, construct a structural health mapping function based on engineering safety thresholds to obtain the structural health index; Step 4: Construct quantitative scoring functions for each auxiliary modality data and calculate the auxiliary health index; Step 5: Introduce the structural dominant factor to fuse the structural health index and the auxiliary health index to obtain the comprehensive health index; Step 6: Classify and determine the safety status of the dam based on the comprehensive health index, and output the corresponding diagnosis and early warning results.

[0023] Specifically, the multi-source heterogeneous data includes structural monitoring data, environmental condition data, manual inspection text data, detection data, and dam appearance image data, etc.

[0024] Specifically, the preprocessing of the multi-source heterogeneous data includes: Standardized analysis and cleaning of multi-source heterogeneous data. A full-dimensional sensing network covering the reservoir dam is constructed to collect multi-source heterogeneous data within the monitoring period, including: automated sensor monitoring data, environmental hydrological data, manual inspection text data, dam exterior image data, and engineering inspection data including regular inspections and problem-triggered inspections. For automated sensor data (displacement, seepage pressure, etc.), a physical limit threshold filter is used to scan the original time series, removing abnormal drift noise, and a linear interpolation algorithm is used to repair missing data. For manual inspection text data, natural language processing technology is used to extract inspection timestamps, and semantic features of inspection conclusions are extracted based on keyword matching. For dam exterior image data, computer vision models are used to extract visual risk feature values ​​reflecting dam surface defects (such as cracks) and the operational status of key components (such as spillway blockage). For engineering inspection data, regularly implemented routine inspection data and special inspection data triggered by abnormal operating conditions are obtained by classification, and inspection feature values ​​reflecting the physical and mechanical properties of the dam are extracted.

[0025] Specifically, the preprocessed data undergoes spatiotemporal rasterization and alignment, including: Using dam monitoring units as spatial indexes and standard time windows as time indexes, a spatiotemporal raster matrix is ​​established, and the preprocessed data is aligned. Within the same time window, an extreme value aggregation strategy is used to extract time-period characteristics of high-frequency or continuously collected data. A long-term retention and event-driven update strategy is adopted to process low-frequency or event-triggered data; The aligned multimodal data are concatenated into vectors within the same spatiotemporal raster to construct the multi-factor feature vector for the nth time window. .

[0026] Specifically, the long-term retention and event-driven update strategy includes: using the previous conclusions when there is no new data; when new data is entered, mapping it to the current window using the time tolerance nearest neighbor matching method; if the time difference exceeds the preset tolerance, introducing a confidence decay factor or marking it as missing.

[0027] Specifically, the spatiotemporal rasterization alignment of multimodal data includes: addressing the issue of inconsistent sampling frequencies among multi-source data by using dam monitoring units as spatial indexes and standard time windows. For time indexing, a spatiotemporal raster matrix is ​​established, and the data acquired in step one is aligned. For high-frequency or continuously collected data (such as water level and displacement), an extreme value aggregation strategy is adopted to extract the most unfavorable feature value (such as maximum displacement and highest water level) within the time window to represent the state of that period. For low-frequency or event-triggered data (such as inspection reports and patrol records), a "long-term retention and event-driven update" strategy is adopted: when there is no new data, the conclusions of the previous period are used; when new data is entered, the time tolerance nearest neighbor matching method is used to map it to the current window. If the time difference exceeds the preset tolerance, the data is updated accordingly. If the confidence decay factor is introduced or the value is marked as missing, then a confidence decay factor is introduced.

[0028] In this embodiment of the invention, in order to unify monitoring data with different sampling frequencies, an extreme value aggregation strategy is adopted for high-frequency continuous sampling data within each time window, specifically including the following steps: Step 1: Divide the time window Suppose the monitoring timeline is divided into a continuous sequence of time windows: ; Among them, the Each time window is defined as: ; This is the preset window length.

[0029] Step 2: Building the Window Data Set

[0030] For any monitoring quantity Extract it within the time window The sampling set within: ; Step 3: Extreme Value Extraction Rules When the monitored quantity is a deformation or displacement index, the absolute extreme value is used as the representative value of the window: ; When the monitored quantity is a unidirectional cumulative value or an operating condition indicator (such as water level), the maximum value is used as the representative value of the window. ; This extreme value represents the most unfavorable operating condition characteristic within that time window.

[0031] Step 4: Handling Missing Tests

[0032] When there are no valid sampled values ​​within the window: If there is valid data in an adjacent window, the nearest neighbor window extreme value is used as the replacement. If consecutive missing values ​​exceed a preset threshold, the test is marked as missing and the confidence weight is reduced.

[0033] Step 5: Output window features

[0034] window extrema As the first The representative feature values ​​of each time window are used for subsequent health index calculations.

[0035] Specifically, the multi-factor feature dataset is constructed by concatenating the aligned multimodal data within the same spatiotemporal grid to construct the multi-factor feature vector for the nth time window. : ; In the formula, This represents the extreme characteristics of structural displacement. For environmental water level characteristics, To inspect the semantic features of the text. For visual image risk characteristics, To detect the feature values ​​of the data.

[0036] Specifically, the structural health mapping function maps structural displacement characteristics. Mapped to a normalized structural health index The expression is as follows: ; In the formula, , , These are the preset structural displacement safety thresholds for Level 1 (Attention), Level 2 (Warning), and Level 3 (Danger); This represents the corresponding health score coefficient.

[0037] Specifically, a quantitative scoring function is constructed for each auxiliary modality data, and an auxiliary health index is calculated, including: Environmental condition scoring function, text semantic scoring function, visual image scoring function, and detection data scoring function are constructed respectively. In this embodiment of the invention, the semantic parsing process for manually inspected text data includes the following steps: Step 1: Text Preprocessing Read the original text of the manual inspection report, remove irrelevant characters, punctuation marks and formatting marks, standardize the text, and extract the inspection date as a timestamp according to the time field.

[0038] Step 2: Keyword dictionary construction

[0039] Establish a semantic keyword dictionary for water conservancy project inspection, and classify common expressions in inspection into semantic sets with different risk levels: ; ; Semantics that do not match the above set are classified into the normal semantic set by default.

[0040] Step 3: Keyword Matching and Semantic Judgment

[0041] The preprocessed text is scanned word by word to determine whether it contains elements from the keyword set. If the text contains any high-risk keyword, it is judged as high-risk semantics; If it does not contain high-risk keywords but contains medium-risk keywords, it is judged as medium-risk semantics; If none of them match, it is considered normal semantics.

[0042] Step 4: Semantic Quantization Mapping

[0043] Map the semantic discrimination results to numerical health scores: ; This score serves as a semantic feature value of the inspection text and is used in subsequent calculations of the comprehensive health index.

[0044] In this embodiment of the invention, the process of extracting visual risk features from dam appearance image data specifically includes the following steps: Step 1: Image Acquisition and Standardization Processing Images of the dam's exterior, captured by fixed cameras or drones, are read and processed for size normalization, brightness equalization, and noise filtering to eliminate the influence of ambient light and shooting angle on the recognition results.

[0045] Step 2: Key Region Segmentation

[0046] Based on prior information about the dam structure, the images were divided into several key monitoring areas, including the upstream face of the dam, the dam crest structure, the spillway area, and the gate area.

[0047] Features are extracted separately for each region to avoid visual information from different parts interfering with each other.

[0048] Step 3: Defect and Operational Status Identification

[0049] Perform visual detection processing on key area images to identify the following two types of targets: Surface defect targets: including cracks, water seepage marks, and erosion areas. Operational status targets include: accumulation of floating debris in the spillway, abnormal gate opening, and blockage. For each type of target, calculate its area proportion or confidence probability value within the region to obtain the corresponding risk indicator: ; in, Indicates the first The relative severity of visual-like risks.

[0050] Step 4: Visual Risk Aggregation

[0051] All identified visual risk indicators are weighted and aggregated to obtain a comprehensive visual risk feature value: ; in, ; For the first Weighting coefficients for visual-like risks.

[0052] Step 5: Health Score Mapping

[0053] Mapping visual risk features to visual health scores: ; This score serves as a health characteristic of the visual modality and is used in the calculation of the comprehensive health index.

[0054] In this embodiment of the invention, the feature extraction process for engineering testing data specifically includes the following steps: Step 1: Classify and read the detection data Read the dam body engineering inspection records and categorize them into two types according to the source of the inspection: Periodic routine testing data: Test results conducted periodically in accordance with the operation and management system; Problem-triggered special inspection data: Special inspection results conducted after anomalies are discovered through monitoring data, inspections, or visual recognition.

[0055] Both types of detection data are uniformly entered into the feature extraction process.

[0056] Step 2: Standardization of testing indicators

[0057] For different types of testing items (such as concrete strength, dam density, seepage state, structural crack width, etc.), their test values ​​are extracted and compared with the corresponding engineering design standards or allowable values ​​to construct standardized testing deviation indicators. ; in, For the first Item test measured value, For corresponding design reference values ​​or specification limits, Standardized deviation index; Step 3: Determine the level of anomaly detection The test results are classified into different levels based on the deviation index: ; in, For detection level, This is the allowable deviation threshold for the project.

[0058] Step 4: Detect the health score mapping

[0059] Map the detection level to the detection health score: ; Step 5: Detection Feature Aggregation When multiple detection indicators exist, a weighted average method is used to obtain the comprehensive health score. ; in, ; For the first Weight of each detection indicator.

[0060] This score is used as a detection feature in the calculation of the comprehensive health index.

[0061] Based on the environmental condition scoring function, text semantic scoring function, visual image scoring function, and detection data scoring function, an auxiliary health index is calculated using a weighted summation model.

[0062] Specifically, the calculation of the multimodal assisted health index involves constructing quantitative scoring functions for each assisted modality and calculating the assisted health index. Environmental condition scoring function : ; In the formula, This is the normal water level. The design flood level or warning water level.

[0063] Text semantic scoring function : ; In the formula, This is a set of high-risk semantic keywords (such as "abnormal" and "dangerous"). This is a set of semantic keywords for medium-risk situations (such as "attention" and "change").

[0064] Visual image scoring function : ; In the formula, This represents the overall risk probability value identified visually. This represents the risk sensitivity coefficient.

[0065] Detection data scoring function For the test data, the principle of "emergency priority, periodic backup" is adopted to determine the feature values, and the scoring rules are as follows: .

[0066] Multimodal Assisted Health Index Fusion: Calculating Assisted Health Indexes using a weighted summation model : ; In the formula, , , , These are the normalized weights for environment, text, vision, and detection factors, respectively, and satisfy the following conditions: When a data item is missing, the weights are automatically reassigned.

[0067] Specifically, the comprehensive health index calculation incorporates structural dominant factors. By integrating the structural health index and the auxiliary health index, the comprehensive health index for the nth period is calculated. : ; In the formula, This is to ensure that physical structure indicators play a dominant role in safety assessment.

[0068] Specifically, the safety status of dams is classified and determined based on a comprehensive health index, including: establishing a comprehensive health index. Mapping relationship with security level : ; In the formula, Thresholds are set for security levels, and .

[0069] Specifically, this also includes: Reverse attribution diagnosis: When the dam's safety status is determined to be level two or below, a reverse attribution mechanism is activated to traverse and backtrack the factors that led to the overall health index. The decreasing controlling factor generates a structured diagnostic report containing disease type, location, and specific numerical values.

[0070] This invention constructs a multi-factor dataset matrix and a normalized relational function to uniformly model features such as structural deformation, environmental hydrology, periodic monitoring, visual appearance, and textual semantics. In particular, for small reservoirs or key local areas lacking automated monitoring instruments, it can utilize non-contact methods such as visual imaging to achieve state perception and quantitative analysis, thereby filling the blind spots of traditional single monitoring methods and improving the accuracy, sensitivity, and robustness of safety diagnosis for small reservoir dams.

[0071] In one specific embodiment of the present invention, a reservoir dam safety classification diagnosis and early warning system includes: The data acquisition module is used to collect multi-source heterogeneous data from the reservoir dam. The dataset construction module is used to preprocess the multi-source heterogeneous data and perform spatiotemporal rasterization alignment on the preprocessed data to construct a multi-factor feature dataset. The structural health index calculation module is used to construct a structural health mapping function based on engineering safety thresholds for structural deformation features in a multi-factor feature dataset, and obtain the structural health index. The auxiliary health index calculation module is used to construct quantitative scoring functions for each auxiliary modality data and calculate the auxiliary health index. The comprehensive health index calculation module is used to incorporate the structural dominant factor and fuse the structural health index with the auxiliary health index to obtain the comprehensive health index. The result judgment module is used to classify and determine the safety status of the dam based on the comprehensive health index, and output the corresponding diagnosis and early warning results.

[0072] In one specific embodiment of the present invention, the monitoring object is the "east auxiliary dam" of a small reservoir, where GNSS surface displacement gauges, reservoir water level gauges, and intelligent monitoring cameras are deployed. Figure 1 As shown, this embodiment of the invention achieves safety classification diagnosis and early warning of small reservoir dams based on spatiotemporal fusion of multimodal data through the following steps: S1: To address the issue of cluttered multi-source data, this embodiment of the invention first establishes a full-dimensional sensing network. Monitoring data at time t is read via an API interface to construct a raw dataset containing automated sensor values, appearance images, and manual inspection text. For the automated sensor data, a time-series dataset is constructed. ;in, For GNSS vertical displacement, This refers to the reservoir water level. To obtain accurate physical quantities, the system sets physical limit thresholds. Scan the original sequence, and once detected This means that the data is identified as electrical drift noise and removed. Then, a linear interpolation algorithm is used to repair the missing data, generating a corrected, high-confidence dataset. Meanwhile, for unstructured data, a deep learning semantic segmentation model was used to extract the percentage of floating debris obstructing the spillway from the monitoring footage. Visual features were analyzed, and semantic parsing of manual inspection records was performed using a dedicated sentiment lexicon for water conservancy projects (containing high-risk terms such as "piping" and "leakage") to extract textual risk features. Ultimately, this forms a standardized multidimensional perception data space.

[0073] S2: To address the challenge of misaligned sampling frequencies for data such as water level (daily measurement), displacement (weekly measurement), and inspection (monthly measurement), this embodiment of the invention establishes a standard time window W (in this embodiment, a monthly window is used). The spatiotemporal raster matrix is ​​indexed by . For example, when processing window data from February 2022, the system first employs an extreme value aggregation strategy for high-frequency continuous data, starting from . Extract the maximum absolute value of displacement within the set for that month. and highest water level This serves as a representative state value for that time period. For low-frequency or discrete data, the system employs an event-driven update and long-term persistence strategy: within time tolerance... Search for the most recent inspection and visual features within a 360-day range; for extremely low-frequency engineering inspection data. By default, the conclusions from the previous period are used. However, if sudden emergency-specific testing data is entered into this window, the new data will immediately overwrite the old data. Through this dynamic alignment mechanism, the spatiotemporal alignment feature vector of the nth time window is constructed. This achieves logical closure of multi-source heterogeneous data within the same spatiotemporal dimension.

[0074] S3: This step is based on the three-level safety threshold sequence determined in the "Technical Specification for Safety Monitoring of Earth and Rockfill Dams". (In this embodiment, the measurements are 1.0mm, 3.0mm, and 5.0mm respectively), a piecewise mapping function is established. The system uses the feature values ​​extracted in step S2. Mapped to a normalized structural health index The measurements obtained in this embodiment For example, because it falls in The system automatically calculates the "focus" range. This index directly reflects the deformation health of the dam's physical structure and serves as the foundation for subsequent integrated calculations.

[0075] S4: To comprehensively consider environmental and apparent risks, this embodiment of the invention constructs quantitative scoring functions for each auxiliary modality. The specific calculation process is as follows: First, compare with the current water level. With the design normal water level Because 45.5m < 48.0m, the environmental condition score is... Recorded as 1.0; secondly, the existence of the spillway was detected based on visual features. Floating debris obstruction, combined with risk sensitivity coefficient Computer vision scoring Secondly, since no risk keywords were found during this period's text inspection and the concrete strength test met the standards, both the text and the test scores were 1.0. Finally, the system utilizes a preset weight vector. The auxiliary health index is calculated by weighting and summing the above scores. This process transforms qualitative appearances into calculable quantitative indicators.

[0076] S5: This embodiment of the invention introduces a structural dominant factor. (Set to 0.6) To ensure the dominant position of physical structural safety, the structural health index will be... With auxiliary health index By integrating the data, a comprehensive health index can be calculated. Based on the preset security level mapping relationship ( ; The system determines that the dam is currently in a "Level II (Attention)" state.

[0077] Specifically, when in this state, the system automatically initiates a reverse attribution diagnosis mechanism: traversing the constituent parts. The system accurately identifies the lowest-scoring "structural displacement factor (0.7)" and the second-lowest "visual floating object factor (0.95)" as the primary causes of the problem and generates a structured diagnostic report. Furthermore, if a severe rainstorm occurs in the next period, causing serious leakage to be discovered during inspections (a sharp drop in text scores), the system will trigger the event-driven update mechanism described in S2, resulting in... The drop below the red alert line directly triggered a Level IV anomaly warning, achieving a closed-loop process from data perception to cause analysis and dynamic warning.

[0078] In one specific embodiment of the present invention, the safety monitoring of the east auxiliary dam of a small reservoir during the flood season is taken as the object. The monitoring period is 30 consecutive days, and the time window ΔT is 1 day. Two GNSS surface displacement monitoring points and one reservoir water level gauge are set up in the dam section, and fixed video monitoring equipment is deployed in the dam surface and spillway area. At the same time, the method of the embodiment of the present invention is verified by combining manual inspection records and engineering test data.

[0079] (1) Multi-source data acquisition and analysis

[0080] During the monitoring period, the following multi-source heterogeneous data were acquired: Structural monitoring data: GNSS surface displacement gauges collect radial displacement data of the dam crest daily to obtain the maximum displacement sequence Δu(t); Environmental operating data: The reservoir water level gauge obtains the daily maximum water level H(t); Manual inspection text data: The inspection frequency is twice a week, and the text content includes descriptions such as "no obvious abnormalities on the dam surface" and "local seepage traces"; Visual imagery data: Video surveillance and drone footage are used to identify cracks on the dam surface, signs of seepage, and the accumulation of floating debris in the spillway; Engineering inspection data: On the 18th day, a special inspection was triggered due to the detection of water seepage risk by video, and the results of seepage flow and local density were obtained.

[0081] After the above data has been cleaned, outlier removed, and standardized, it will proceed to the next step of the alignment process.

[0082] (2) Spatiotemporal rasterization alignment and feature construction

[0083] Using the East Auxiliary Dam as a single monitoring unit and a one-day time window, multi-source data were spatiotemporally aligned. For GNSS displacement and water level data, an in-window extreme value aggregation strategy is adopted to extract the most unfavorable features; A mechanism of "long-term retention + event-triggered update" is adopted for inspection texts and detection data; For visual image data, extract the crack identification probability value Pv(t).

[0084] Construct the multi-factor feature vector for day n: ; in, The structural displacement extreme value characteristics within the nth time window; For environmental water level characteristics, To inspect the semantic features of the text. For visual image risk characteristics, To detect the feature values ​​of the data.

[0085] (3) Calculation of structural health index (corresponding to step S4)

[0086] The structural displacement safety threshold is set as follows: Level 1 threshold: Δu1 = 5mm; Secondary threshold: Δu2 = 10 mm; Level 3 threshold: Δu3 = 15mm; When the maximum displacement Δu on the dam crest was monitored on the 21st day 21 When the thickness is 9.2 mm, the following is calculated using the structural health mapping function: ; This indicates that the structural state has entered the "attention-alert" transition zone.

[0087] (4) Calculation of multimodal health index

[0088] Environmental condition rating: Water level is close to the warning level, rating value 0.75; Text semantic scoring: Keywords such as "water seepage" and "needs attention" appeared during the inspection, with a score of 0.65; Visual imaging assessment: The combined risk probability of cracks and water seepage is 0.58; Test data score: Specialized testing showed that the seepage flow was too high, with a score of 0.60.

[0089] After weighted fusion, the following is obtained: .

[0090] (5) Comprehensive health index and safety level determination

[0091] Taking the dominant structural factor α = 0.6, calculate the comprehensive health index: ; Based on the safety level mapping rules, the safety level of the East Auxiliary Dam was determined to be Level II (Alert) on that day, and the system automatically generated an early warning message.

[0092] (6) Reverse attribution diagnostic results

[0093] The system performs reverse analysis on the decline in the overall health index and determines the main causes in the following order: Structural displacement anomaly (with the largest weight contribution); Visual and detection indicators related to water seepage; High water level environment conditions.

[0094] The final output is a structured diagnostic report that includes the anomaly type, location, trigger time, and recommended handling measures, enabling interpretable early warning.

[0095] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0096] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for safety classification diagnosis and early warning of reservoir dams, characterized in that, include: Collect multi-source heterogeneous data on reservoir dams; The multi-source heterogeneous data is preprocessed, and the preprocessed data is then spatiotemporally rasterized and aligned to construct a multi-factor feature dataset. For structural deformation features in a multi-factor feature dataset, a structural health mapping function based on engineering safety thresholds is constructed to obtain a structural health index. Construct quantitative scoring functions for each auxiliary modality data and calculate the auxiliary health index; By introducing a structural dominant factor and fusing the structural health index and auxiliary health index, a comprehensive health index is obtained. The dam's safety status is classified and determined based on the comprehensive health index, and corresponding diagnostic and early warning results are output.

2. The method for safety classification diagnosis and early warning of reservoir dams according to claim 1, characterized in that, The preprocessed data undergoes spatiotemporal rasterization and alignment, including: Using dam monitoring units as spatial indexes and standard time windows as time indexes, a spatiotemporal raster matrix is ​​established, and the preprocessed data is aligned. Within the same time window, an extreme value aggregation strategy is used to extract time-period characteristics of high-frequency or continuously collected data. A long-term retention and event-driven update strategy is adopted to process low-frequency or event-triggered data; The aligned multimodal data are concatenated into vectors within the same spatiotemporal raster to construct the multi-factor feature vector for the nth time window. .

3. The method for safety classification diagnosis and early warning of reservoir dams according to claim 2, characterized in that, The long-term retention and event-driven update strategy includes: using the previous conclusions when there is no new data; when new data is entered, mapping it to the current window using the time tolerance nearest neighbor matching method; if the time difference exceeds the preset tolerance, introducing a confidence decay factor or marking it as missing.

4. The method for safety classification diagnosis and early warning of reservoir dams according to claim 1, characterized in that, The structural health mapping function will incorporate structural displacement characteristics. Mapped to a normalized structural health index The expression is as follows: ； In the formula, , , These are preset safety thresholds for structural displacement at levels one, two, and three; This represents the corresponding health score coefficient.

5. The method for safety classification diagnosis and early warning of reservoir dams according to claim 1, characterized in that, Construct quantitative scoring functions for each auxiliary modality data, and calculate the auxiliary health index, including: Environmental condition scoring function, text semantic scoring function, visual image scoring function, and detection data scoring function are constructed respectively. Based on the environmental condition scoring function, text semantic scoring function, visual image scoring function, and detection data scoring function, an auxiliary health index is calculated using a weighted summation model.

6. The method for safety classification diagnosis and early warning of reservoir dams according to claim 1, characterized in that, By introducing a structural dominant factor and fusing structural health index with auxiliary health index, a comprehensive health index is obtained, including: ； in, Indicates the dominant structural factor. Indicates an auxiliary health index, This indicates the structural health index.

7. The method for safety classification diagnosis and early warning of reservoir dams according to claim 1, characterized in that, The dam's safety status is classified and determined based on a comprehensive health index, including: establishing a comprehensive health index. Mapping relationship with security level : ； In the formula, Thresholds are set for security levels, and .

8. The method for safety classification diagnosis and early warning of reservoir dams according to claim 1, characterized in that, Also includes: When the dam's safety status is determined to be level two or below, a reverse attribution mechanism is activated to iterate and trace back the factors that led to the comprehensive health index. The decreasing controlling factor generates a structured diagnostic report containing disease type, location, and specific numerical values.

9. A reservoir dam safety classification diagnosis and early warning system, employing the reservoir dam safety classification diagnosis and early warning method according to any one of claims 1-8, characterized in that, include: The data acquisition module is used to collect multi-source heterogeneous data from the reservoir dam. The dataset construction module is used to preprocess the multi-source heterogeneous data and perform spatiotemporal rasterization alignment on the preprocessed data to construct a multi-factor feature dataset. The structural health index calculation module is used to construct a structural health mapping function based on engineering safety thresholds for structural deformation features in a multi-factor feature dataset, and obtain the structural health index. The auxiliary health index calculation module is used to construct quantitative scoring functions for each auxiliary modality data and calculate the auxiliary health index. The comprehensive health index calculation module is used to incorporate the structural dominant factor and fuse the structural health index with the auxiliary health index to obtain the comprehensive health index. The result judgment module is used to classify and determine the safety status of the dam based on the comprehensive health index, and output the corresponding diagnosis and early warning results.

Images